Title of Invention	PLANT CONTROL APPARATUS
Abstract	A plant control system includes: a numerical calculation execution part which calculates the operation characteristic of the plant; a model for simulating the plant control characteristic according to information on the numerical calculation result; a learning part which learns the plant operation method by using the model; a learning information database which stores learning information data on the learning part; a pattern generation part which generates pattern data expressing a state input based on the learning information data in the learning part with a smaller input number than the model input dimension; a pattern database which stores the pattern data generated in the pattern generation part; and a learning result determination part which selects a learning result having a preferable control effect from the learning result obtained by using a plurality of patterns.

Title of Invention

PLANT CONTROL APPARATUS

Abstract

A plant control system includes: a numerical calculation execution part which calculates the operation characteristic of the plant; a model for simulating the plant control characteristic according to information on the numerical calculation result; a learning part which learns the plant operation method by using the model; a learning information database which stores learning information data on the learning part; a pattern generation part which generates pattern data expressing a state input based on the learning information data in the learning part with a smaller input number than the model input dimension; a pattern database which stores the pattern data generated in the pattern generation part; and a learning result determination part which selects a learning result having a preferable control effect from the learning result obtained by using a plurality of patterns.

Full Text	BACKGROUND OF THE INVENTION The present invention relates to a plant control system and in particular or system, to a thermal power plant. The plant control system processes a measurement signal of an operation state amount obtained from a plant to be controlled, calculates an operation signal to be supplied to the plant to be controlled, and transmits it as a control instruction to the control system. The plant control system uses an algorithm for calculating the operation signal so that the measurement signal of the plant operation state amount satisfies its target value. It is possible to use the PI (proportion/integration) control algorithm for controlling the plant. In the PI control, the deviation of the plant measurement signal from its target value is multiplied by the proportion gain. The obtained value is added by the value obtained by time- integrating the deviation so as to obtain the operation signal to be supplied to the object to be controlled. On the other hand, in order to automatically correct the control algorithm in accordance with a change of the plant operation state and the environment, the plant may be controlled by using the adaptive control and the learning algorithm. JP-A-2000-35956 discloses a control apparatus using a reinforced learning method using a model as a method for obtaining the operation signal of the control apparatus for controlling the plant by using the learning algorithm. In the method using the technique of the reinforced learning method, the control apparatus includes a model for predicting the characteristic of the object to be controlled and a learning part for learning the method for generating such a model input that a model output can achieve the target value. By inputting the model input learned by the learning part to the model, it is possible to obtain the effect that the model output approaches the target value. The learning type adaptive control corrects the model by using the measurement signal which has measured the plant operation state and performs re- learning by using the corrected model so as to perform online correction/modification of the control algorithm for accurately controlling the plant. Accordingly, it is necessary to correct the model within a cycle (control cycle) during which the operation signal outputted from the control apparatus is modified for the plant and use the corrected model for the re-learning so as to complete the learning and correct/modify the control algorithm. The control cycle can be considered as a time from the plant operation completion to the moment when the plant enters a static state, which is normally several minutes or several tens of minutes. For example, when controlling a complicated plant such as a thermal power plant, the model input dimension is several tens or several hundreds and the number of combinations of the learning model inputs (search space) increases, which in turn requires a long learning time. As a result, it becomes difficult to online-correct/modify the control algorithm required for accurately controlling the plant. Consequently, in order to complete the learning within the control cycle, it is necessary to correct the model in accordance with increase of the model input dimension and increase the learning speed of re-learning by using the corrected model. JP-A-7-160661 discloses a technique for increasing the learning speed in a neural network learning by classifying teacher data into a plurality of patterns according to the combinations of the plant measurement information and extracting a learning pattern according to the control result so as to learn only the teacher data on the extracted pattern. SUMMARY OF THE INVENTION By using the technique of the reinforced learning method using the model disclosed in JP-A-2000- 35956, it is possible to automatically learn the method for generating an operation signal which can achieve the control target. However, when a model is corrected according to the measurement signal which has measured the plant operation state and the corrected model is used for re- learning so as to correct the control algorithm, the model input dimension increases in a complicated plant, which makes it difficult to perform the learning within the control cycle for controlling the plant. Moreover, when the technique disclosed in JP- A-7-160661 is used, it is possible to divide the teacher data by patterning so as to reduce the search space and increase the learning speed. That is, even if the model input dimension increases and the search space (number of teacher data) increases, it is possible to perform learning within the control cycle by executing the appropriate patterning. However, when the teacher data to be classified is biased, the number of teacher data may be too small depending on the learning pattern and it may not be possible to acquire a desired learning result. Moreover, since the type of patterns to be classified is decided according to the knowledge of a designer of the control apparatus, there is a problem that the pattern generation requires a large load on the operator. It is therefore an object of the present invention to provide a plant control system which corrects a model in accordance with a measurement signal which has measured a plant operation state and uses the corrected model for re-learning, which enables execution of the high-speed learning for correcting the plant control algorithm and accurately controlling the plant. A plant control system for controlling a plant including an operation signal generation part which calculates an control instruction operation signal for a plant to be controlled by using a measurement signal which has measured an operation state of the plant, apparatus comprising: an operation signal database which stores the measurement signal of the measured plant; an operation signal database which stores an operation signal for the plant; a numerical calculation execution part which calculates the operation characteristic of the plant; a numerical calculation database which stores the numerical calculation result from the numerical calculation execution part; a simulation model which simulates control characteristic of a plant when the operation signal is supplied to the plant according to information on the numerical calculation result from the numerical calculation database; a learning part which learns the plant operation method by using the model; a learning information database which stores learning information data on the learning part; a control logic database which contains information used for the operation signal outputted from the operation signal generation part; a pattern generation part which generates pattern data expressing a state input based on the learning information data in the learning part with a smaller input number than the model input dimension; a pattern database which stores the pattern data generated in the pattern generation part; and a learning result determination part which selects a learning result having a preferable control effect from the learning result obtained by using a plurality of patterns; wherein the operation signal is calculated from the operation signal generation part according to the learning result selected by the learning result determination part. According to the present invention, it is possible to provide a plant control system which corrects a model according to a measurement signal which has measured the plant operation state and performs re-learning at a high speed by using the corrected model, thereby correcting the plant control algorithm so as to accurately control the plant. Other objects, features and advantages of the invention will become apparent from the following description of the embodiments of the invention taken in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS Fig. 1 is a control block diagram showing an entire configuration of a plant control system according to an embodiment of the present invention. Fig. 2 is control block diagram indicating a learning part of a plant control system according to the embodiment of the present invention shown in Fig. 1. Fig. 3 explains the patterning operation in the plant control system according to the embodiment of the present invention shown in Fig. 1. Fig. 4 is a flowchart showing the operation of the plant control system according to the embodiment of the present invention shown in Fig. 1. Fig. 5 is a flowchart showing the pre- learning operation in the flowchart showing the operation of the plant control system according to the embodiment of the present invention shown in Fig. 4. Fig. 6 is a flowchart showing the pattern generation operation in the flowchart showing the operation of the plant control system according to the embodiment of the present invention shown in Fig. 4. Fig. 7 explains an example of the information stored in a learning information database in the plant control system according to the embodiment of the present invention shown in Fig. 1. Fig. 8 explains an example of the information stored in a pattern database in the plant control system according to the embodiment of the present invention shown in Fig. 1. Fig. 9 is a flowchart for detailed explanation of the pattern search algorithm as a part of the pattern generation operation in the flowchart showing the operation of the plant control system according to the embodiment of the present invention shown in Fig. 4. Fig. 10 explains the pattern search algorithm for calculating a patterning error in the flowchart showing the operation of the plant control system according to the embodiment of the present invention shown in Fig. 9. Fig. 11 is a flowchart showing the re- learning operation in the flowchart showing the operation of the plant control system according to the embodiment of the present invention shown in Fig. 4. Fig. 12 is a flowchart showing a detailed explanation of a part of the re-learning operation in the flowchart showing the operation of the plant control system according to the embodiment of the present invention shown in Fig. 4. Fig. 13 shows an example of an initial screen displayed on a display mounted on the plant control system according to the embodiment of the present invention shown in Fig. 1. Fig. 14 shows an example of a screen displayed on the display when a numerical calculation is executed in the plant control system according to the embodiment of the present invention shown in Fig. 1. Fig. 15 shows an example of a screen displayed on the display when a database is referenced in the plant control system according to the embodiment of the present invention shown in Fig. 1. Fig. 16 shows an example of a screen displayed on the display when patterning and learning are executed in the plant control system according to the embodiment of the present invention shown in Fig. 1. Fig. 17 shows an example of a screen displayed on the display when an operation is executed in the plant control system according to the embodiment of the present invention shown in Fig. 1. Fig. 18 shows an example of a screen displayed on the display when the pre-learning and the pattern generation are executed in the plant control system according to the embodiment of the present invention shown in Fig. 1. Fig. 19 shows an example of a screen displayed on the display when re-learning is executed in the plant control system according to the embodiment of the present invention shown in Fig. 1. Fig. 20 shows an example of a screen displayed on the display when pattern generation is executed in the plant control system according to the embodiment of the present invention shown in Fig. 1. Fig. 21 shows a configuration outline of a thermal power plant including boiler to be controlled by employing the plant control system according to the embodiment of the present invention shown in Fig. 1. Fig. 22 is an enlarged view of an air heater portion in the boiler of the thermal power plant to be controlled shown in Fig. 21. Fig. 23 is a flowchart showing a part of the re-learning process in the control apparatus of the thermal power plant having the boiler according to the embodiment of the present invention shown in Fig. 21. Fig. 24 is an example of a screen shown on the display when an operation is performed for the control apparatus of the thermal power plant having the boiler according to the embodiment of the present invention shown in Fig. 1. Figs. 25A-25C explain examples of patterned grouping in the control apparatus of the thermal power plant having the boiler according to the embodiment of the present invention shown in Fig. 1. DETAILED DESCRIPTION OF THE EMBODIMENTS Description will now be directed to a plant control system according to an embodiment of the present invention with reference to the attached drawings. [Embodiment 1] Fig. 1 is a block diagram showing the entire configuration of the plant control system according to an embodiment of the present invention. In Fig. 1, the plant of the present embodiment is a plant 100 constituting a thermal power plant including a boiler using coal as fuel. The thermal power plant 100 including the boiler is configured so as to be controlled by a control apparatus 200. The control apparatus 200 for controlling the plant 100 as a control object includes, as calculation apparatuses, a numerical calculation part 220, a model 230, a learning part 260, an operation signal generation part 280, a learning result judging part 300, and a pattern generation part 400. The control apparatus 200 includes as databases, a measurement signal database 210, a numerical calculation database 240, an operation signal database 250, a learning information database, a control logic database 290, and a pattern database 500. Furthermore, the control apparatus 200 includes, as an external interface, an external input interface 201 and an external output interface 202. The control apparatus 200 acquires the measurement signal 1 which has measured various state amounts of a thermal power plant from the thermal power plant 100 via the external input interface 291. Moreover, the control apparatus 200 sends an operation signal 15 for controlling, for example, a burner of the boiler and an air flow rate of an air port via the external output interface 202 to the plant 100 as the thermal power plant of the control object. In the control apparatus 200, the measurement signal 1 which has measured various state amounts of the plant 100 is sent to the external input interface 201 and stored as the measurement signal 2 in the measurement signal database 210 as a database of the control apparatus 200. An operation signal 14 generated by an operation signal generation part 280 as a calculation apparatus provided in the control apparatus 200 is sent to the external output interface 202 and stored in the operation signal database 250 as a database provided in the control apparatus 200. The operation signal generation part 280 uses control logic data 13 stored in the control logic database 290 and learning data 12 outputted from the learning result determination part 300 as a calculation apparatus provided in the control apparatus 200 so as to generate the operation signal 14 functioning as the operation signal 15 for the plant 100, so that the measurement signal 1 of the plant 100 achieves an operation target value and outputs the signal to the external interface 202. The control logic database 290 contains a control circuit and a control parameter used for calculating the control logic data 13 so as to output the control logic data. The known PI control can be used in the control circuit for calculating the control logic data 13. The learning data stored in the learning information database 270 as a database provided in the control apparatus 200 is generated by a learning part 260 provided in the control apparatus 200. The learning part 260 is connected to the model 230 as a calculation apparatus provided in the control apparatus 200. The model 230 has a function for simulating the control characteristic of the plant 100 as the thermal power plant. That is, the model 230 performs a calculation for simulation of the following: the operation signal 15 serving as a control instruction is supplied to the plant 100 so as to obtain the measurement signal 1 as the control result. For this simulation calculation, a model input 7 for operating the model 230 is received from the learning part 260 and the model 230 performs simulation calculation of the characteristic change by control of the plan so as to obtain a model output 8 as the result of the simulation calculation. Here, the model output 8 is a predicted value of the measurement signal 1 of the plant 100. The model 230 is configured according to the numerical calculation result 6 stored in the numerical calculation database 240 as a database provided in the control apparatus 2 00. The numerical calculation part 220 as a calculation apparatus provided in the control apparatus 200 analyzes the operation characteristic of the plant 100 by using a physical model simulating the plant 100. The calculation result 4 obtained by analyzing the characteristic of the plant 100 by using the physical model in the numerical calculation part 220 is stored in the numerical calculation database 240. The model 230 calculates a model output 8 corresponding to the model input 7 by using information stored in the numerical calculation database 240 and the measurement signal database 210 and employing the statistical method such as the neural network. The model 230 extracts the numerical calculation result 6 required for calculating the model output 8 corresponding to the model input 7 from the numerical calculation database 240 by using the model information data 5 and interpolates the result. Moreover, the model 230 can be corrected by using the measurement signal 3 stored in the measurement database 210 so that the control characteristics of the model 230 and the plant 100 coincide with each other when the characteristic of the physical model of the numerical calculation part 220 is different from the characteristic of the plant 100. The learning part 260 learns a method for generating the model input 7 so that the model output 8 which is simulation-calculated by the model 230 achieves the model output target value set in advance by an operation staff. Here, the current model input is inputted to the learning part 2 60 so as to output its model input change width. Here, the input to the learning part 2 60 will be called a state input and the output will be called an operation amount. The constriction condition used for learning and the learning information data 9 containing the model output target value are stored in the learning information database 270. The learning part 260 directly learns the state input before start of the plant operation and, after the start of the plant operation, performs learning by patterning the state input by using the pattern data stored in a pattern database 500 which will be detailed later. Moreover, the learning data 10 obtained as a result of the learning in the learning part 260 is stored in the learning information database 270 provided in the control apparatus 200. The learning data 10 includes the state input before the model input modification, the operation amount at the state input, and information on the model output change width obtained as a result of the operation. It should be noted that the functions of the learning part 260 will be detailed later. The pattern generation part 400 as a calculation apparatus provided in the control apparatus 200 generates pattern data 17 by using the learning data 16 learned before the start of the plant operation in the learning part 260 and stores the pattern data 17 in the pattern database 500. After operation of the plant 100 is started, optimal pattern data 18 similar to the measurement signal 3 is selected from the pattern data stored in the pattern database 500 and the learning part 260 patterns the state input contained in the measurement signal 3. The learning data 11 obtained by the learning in the learning part 260 and stored in the learning information database 270 is inputted to a learning result determination part 300 as a calculation apparatus provided in the control apparatus 200. When a plurality of pattern data 18 are used for the learning, the learning result determination part 300 selects the learning data which makes the control effect to be maximum from the respective pattern learning data 11. When only one pattern is used, the learning data on the pattern is selected. The learning data 12 selected by the learning result determination part 300 is inputted the operation signal generation part 280. The operation staff of the plant 100 can access information stored in various databases provided in the control apparatus 200 by using an external input device 600 formed by a keyboard 601 and a mouse 602, a data processing apparatus 610 having a data send/receive part 612 which can send and receive data to/from the control apparatus 200, and a display 620. Moreover, the operation staff of the plant 100 can use the aforementioned apparatuses to input setting parameters used in the numerical calculation part 200, the learning part 260, and the pattern generation part 400 which are provided in the control apparatus 200. The data processing apparatus 610 is formed by an external input interface 611, a data send/receive part 612, and an external output interface 613. The input signal 61 of the data processing apparatus generated in the external input apparatus 600 is inputted to the data processing apparatus 610 via the external input interface 611. The data send/receive part 612 of the data processing apparatus 610 acquires the I/O data information 60 provided in the control apparatus 200 according to the information in the data processing apparatus input signal 62. Moreover, according to the information in the data processing apparatus input signal 62, the data send/receive part 612 outputs the I/O data information 60 containing the parameter setting value used in the numerical calculation part 220, the learning part 260, and the pattern generation part 400 provided in the control apparatus 200. The data send/receive part 610 transmits data processing apparatus output signal 63 obtained by processing the I/O data information 60, to the external output interface 613. A data processing apparatus output signal 64 is sent from the external output interface 613 to the display 620 and displayed. Accordingly, information stored in the respective databases provided in the control apparatus 200 can be displayed on the display 620. It should be noted that the plant control system 200 according to the embodiment of the present invention includes the measurement signal database 210, the numerical calculation database 240, the operation signal database 250, the learning information database 270, the control logic database 290, and the pattern database 500 which are arranged so as to constitute the calculation apparatus of the control apparatus 200. However, it is also possible to arrange all or some of them outside the control apparatus 200. Moreover, in the aforementioned example, the numerical calculation part 220 is built in the control apparatus 200. However, it is also possible to arrange it outside the control apparatus 200. For example, it is also possible to arrange the numerical calculation part 220 and the numerical calculation database 240 outside the control apparatus 200 and the numerical calculation result 6 is sent to the control apparatus 2 00 via the Internet. Next, the learning part 260 provided in the control apparatus 200 will be detailed with reference to Fig. 2.. As shown in Fig. 2, the learning part 2 60 includes a model input generation part 2 61, a pre- learning part 262, a pattern selection part 263, a pattern transforming part 264, and a re-learning part 2 65. The model input generation part 2 61 constituting the learning part 260 applies an operation amount 19 to the current model input condition of the model 230 so as to generate the model input 7 and the state input 20 after the operation, by the model input generation part. Moreover, the state input 20 is extracted by using the model input generation part 2 61 from the measurement signal 3 of the plant 100 stored in the measurement signal database 210. The pre-learning part 262 constituting the learning part 260 uses the learning information data 9 stored in the learning information database 270 before start of the operation of the plant 100, so as to learn the method for generating the model input 7 so that the model output 8 simulated/calculated by the model 230 can achieve the model output target value. Here, the state input 20 is used as it is. The model input 7 is generated by applying the operation amount 19 learned by the pre-learning part 2 62, to the current model input condition, inputs it to the model 230, and inputs it as the state input 20 after the operation to the pre-learning part 262. The learning data 10 obtained as a result of the learning in the pre-learning part 262 is stored in the learning information database 270. According to the learning information 16 stored in the learning information database 270, the pattern generation part 400 generates pattern data and stores the pattern data 17 generated by the pattern generation part 400, in the pattern database 500. The pattern selection part 263 constituting the learning part 260 selects the pattern data 23 from the pattern data 18 stored in the pattern database 500 according to the state input 20 and supplies it to the pattern transforming part 264. The pattern transforming part 264 constituting the learning part 250 patterns the pattern data 23 selected by the pattern selection part 263 and reduces the input dimension. Moreover, the pattern transforming part 264 executes inverse operation on the operation amount 21 learned by the patterning so as to increase the input dimension. Referring to Fig. 3, explanation will be given on the patterning. As shown in Fig. 3, the pattern transforming part 264 processes the pattern data as follows: the input information value to be used in the patterning is set to 1 and the input information value not to be used is set to 0. Accordingly, when patterning the state input 20, only the input having the pattern data information value 1 for the state input 20 is selected. Moreover, in the inverse operation, linear interpolation is performed between the inputs used in the patterning for the operation amount 21 patterned after the learning so as to obtain an operation amount of other inputs. In Fig. 2, the re-learning part 265 constituting the learning part 2 60 learns the generation method of the model input 7 after the operation of the plant is started, by using the learning information data 9 stored in the learning information database 270, so that the model output 8 to be simulated/calculated by the model 230 corrected by the measurement signal 3 can achieve the model output target value. Here, the pattern transforming part 264 patterns the state input 20. The patterned state input 22 is supplied to the re-learning part 265 corresponding to the pattern. As a result of learning in the re-learning part, a patterned operation amount 21 is outputted from the re-learning part 265. The patterned operation amount is transformed again into the operation amount 19 by the pattern transforming part 2 64 and inputted to the model input generation part 2 61 so as to generate the model input 7 and the state input 20 after the operation. With the configuration thus far explained, the learning part 260 realizes the pre-learning for the pattern generation before starting the operation of the plant and the re-learning by the patterning of the state input after the operation start of the plant 100. Fig. 4 is a flowchart showing the operation of the control apparatus 200 for controlling a plant as the embodiment of the present invention explained with reference to Fig. 1 to Fig. 3. In the flowchart of Fig. 4, a combination of steps 1000, 1010, 1020, 1030, 1040, and 1050 is executed. Hereinafter, explanation will be given on each of the steps. After start of the operation of the control apparatus 200 for controlling the plant 100, firstly, in step 1000, numerical calculation is executed by the numerical calculation part 220 of the control apparatus 200 and the numerical calculation data 4 is sent to the numerical calculation database 240 so as to be stored in it. Next, in step 1010, the learning part 260 and the model 230 of the control apparatus 200 are used to execute pre-learning of the plant operation method and transmit the learning data 10 to the learning information database 270 so as to be stored in it. Next, in step 1020, the pattern generation part 400 of the control apparatus generates pattern data 17 by using the learning data 16 obtained by pre- learning stored in the learning information database 270 and transmits the pattern data 17 to the pattern database 500 so as to be stored in it. The aforementioned operation is executed before starting the operation of the plant 100. After the operation of the plant 100 is started, step 1030 inputs the measurement signal 1 of the plant 100 to the control apparatus 200 by using the external input interface 201 and transmits it to the measurement signal database 210 so as to be stored in it. Step 1040 patterns the state input extracted by the acquired measurement signal 3 by using the pattern database stored in the pattern database 500 and executes re-learning in the learning part 2 60. After the learning in the learning part 260, by using the learning result determination part 300 of the control apparatus 200, appropriate learning data 12 is selected from the learning data 11 obtained by the used pattern is selected and outputted from the learning result determination part 300 to the operation signal generation part 280. In the subsequent step 1050, the operation signal generation part 280 generates an operation signal 14 by using the learning data 12 and the control logic data 13 and outputs it as the operation signal 15 from the operation signal generation part 280 via the external input interface 202 to the plant 100. The aforementioned steps 1030 to 1050 are repeatedly executed each time a measurement signal is inputted, thereby controlling the plant 100. Next, the operations of the steps 1010, 1020, and 1040 in Fig. 4 will be detailed with reference to a flowchart. Fig. 5 is a flowchart showing the operation of the pre-learning of step 1010. As shown in Fig. 5, the flowchart of the pre- learning operation is executed by a combination of steps 2000, 2010, 2020, 2030, 2040, 2050, 2060, 2070, and 2080. Hereinafter, a detailed explanation will be given on each of the steps. Step 2000 initializes (sets I=1) the number of initialization times I which indicates the number of repetitions of the steps 2010 to 2060. Next, step 2010 sets an initial value of the state input. An arbitrary model input value can be selected as the initial value of the state input. Step 2020 initializes (sets J = 1) the number of operation times J indicating the number of operation times J indicating the number of repetitions of steps 2030 to 2050. Next, step 2030 obtains an operation amount of the model input by using the pre-learning part 2 62 of the learning part 260 constituting the control apparatus 2 00. Step 2040 adds the operation amount to the state input by using the model input generation part of the learning part 260 to operate the model 230 and update the state input. Step 2050 learns the operation method of the model 230 by using the neural network and various learning algorithms such as the reinforced learning according to the model output obtained as a result of the operation of the model 2230. Step 2060 is a branching section where the number of operation times J is compared to a predetermined threshold value. If J is smaller than the threshold value, J is incremented by 1 and control is returned to step 2030. If J is greater than the threshold value, control is passed to step 2070 which is a branching section. Step 2070 compares the number of initialization times I to the predetermined threshold value. If I is smaller than the threshold value, I is incremented by 1 and control is returned to step 2010. If I is greater than the threshold value, control is passed to step 2080. Step 2080 transmits the learning result obtained by the learning part 2 60 to the learning information database 270 so as to be stored in it. Then, control is passed to step to terminate the pre- learning operation. By the aforementioned operation, the pre- learning starts learning by using an arbitrary model input as the initial value of the state input in step 2010. Accordingly, it is possible to acquire an operation method to reach the model output target value from an arbitrary model input condition. Fig. 6 is a flowchart showing the pattern generation of step 1020. As shown in Fig. 6, the flowchart of the pattern generation operation executes a combination of steps 4000, 4010, 4020, and 4030. Hereinafter, each of the steps will be detailed with reference to Fig. 7 and Fig. 8, Step 4000 initializes (sets I=1) the data reference number I as the value indicating the number of repetitions of the steps 4010 to 4020. Next, step 4010 references the learning data obtained by pre-learning stored in the learning information database 270 according to I so as to generate a pattern. It should be noted that a detailed explanation will be given later on the pattern generation algorithm. Fig. 7 shows the learning data stored in the learning information database 270 of the control apparatus 2 00. As shown in Fig. 7, the learning information database 270 contains the state input value, the operation amount value in its state input, and the model output amount obtained as the result of the operation. A pattern is generated according to the operation amount in the learning data. It should be noted that in Fig. 7, S_0001 is a number assigned for distinguishing the state input. Step 4020 in Fig. 6 transmits the generated pattern and the state input value of the referenced learning data to the pattern database 500 so as to be stored in it. Fig. 8 shows the learning data stored in the pattern database 500 of the control apparatus 200. As shown in Fig. 8, the pattern database 500 contains the pattern data and the state input value referenced upon generation of the pattern. When storing the generated pattern, if the generated pattern coincides with the pattern data already stored, it is stored as the state input corresponding to the pattern data. Accordingly, there is a case that one pattern data is correlated to a plurality of state inputs. It should be noted that in Fig. 8, P 0001 is a number assigned to distinguish pattern data. Step 4030 in Fig. 6 is a branching section where the data reference number I is compared to the predetermined threshold value. I f J is smaller than the threshold value, J is incremented by 1 and control is returned to step 4010. On the contrary, if J is equal to or greater than the threshold value, control is passed to the step to terminate the operation of the pattern generation. Next, a detailed explanation will be given on the pattern generation algorithm of step 4010 with reference to the flowchart of Fig. 9 and Fig. 10. As shown in Fig. 10, the pattern generation algorithm starts search from the pattern input dimension 1 in order to suppress the input dimension to be patterned to a minimum value and repeats the search to generate a pattern while increasing the input dimension until the end condition is satisfied. Moreover, in order to effectively perform the pattern search, the evolution search algorithm is used for the search. That is, the pattern database expression is made to be a gene (solution candidate) and the generated solution candidates are subjected to genetic operations such as mixing and mutation so as to search for an optimal pattern. As shown in Fig. 9, the flowchart of step 4010 is formed by a combination of steps 4011, 4012, 4013, 4014, 4015, 4016, 4017, 4018, and 4019. Hereinafter, each of the steps will be detailed. Step 4011 initializes J (J = 1) as a value indicating the input dimension used for patterning. Step 4012 initializes the number of generations K (set K =1) as a value indicating the number of repetitions of steps 4010 to 4015. Next, step 4013 generates L solution candidates under the constriction condition that J is the input dimension which makes the pattern data information value. Step 4014 calculates a patterning error for each of the solution candidates. As shown in Fig. 10, the patterning error is calculated as an operation amount obtained by linearly interpolating the operation amount of the learning data to be referenced and the input used in patterning. Step 4015 performs genetic operations such as crossover and mutation on the solution candidates under the constriction condition according to the calculated patterning error so as to generate a new solution candidate. Step 4016 is a branching section . If the number of generations K is equal to or smaller than a predetermined threshold value, K is incremented by 1 and control is returned to step 4010. Otherwise, control is passed to step 4017. Step 4017 selects a solution candidate which makes the patterning error to be minimum for the solution candidate group which has passed through the predetermined number of generations and control is passed to step 4018. Step 4018 is a branching section. If the patterning error of the solution candidate selected in step 4017 is equal to or greater than a predetermined threshold value, or the input dimension J is equal to or smaller than a predetermined threshold value, J is incremented by 1 and control is returned to step 4012. Otherwise, control is passed to step 4019. Step 4019 uses the selected solution candidate as a generation pattern and transmits it to the pattern database 500 so as to be stored in it. Control is passed to the step to terminate the operation of the pattern generation algorithm. With the aforementioned operation, the generated pattern data is stored in the pattern database 500. The operation staff of the plant 100 uses the delta processing apparatus 610 to display these information on the display 620. Thus, the operation staff can know what kind of state input can be currently used for patterning. Moreover,, the operation staff can input threshold values J and K as the setting parameters via the display 620. Fig. 11 is a flowchart showing the re- learning operation of step 1040. As shown in Fig. 11, the flowchart of the re- learning operation is formed by a combination of steps 2100, 2110, 2120, 2130, 2140, 2150, 2160, 2170, and 2180. Hereinafter, each of the steps will be detailed. Step 2100 extracts the state input to be inputted to the re-learning part 2 65 from the measurement signal by using the model input generation part 261 of the learning part 260 constituting the control apparatus 260. Next, step 2110 uses the pattern selection part 263 of the learning part 260 to calculate the minimum state input error between the extracted state input and the pattern data stored in the pattern database 500 of the control apparatus 200. As shown in Fig. 8, the pattern database 500 contains a plurality of state input values for each of patterns and the minimum state input error is a minimum difference between these values and the current state input value. The minimum state input error is the similarity between the current state input and the state input experienced in the pre-learning. Accordingly, it is possible to select a pattern corresponding to the state input having a high similarity with the current state input value according to the minimum state input error. As a result, by using the selected pattern, it is possible to learn the operation amount which can obtain the control effect equivalent to the pre- learning result for the current state input. Step 2120 is a branching section to decide whether any of the pattern minimum state input errors calculated in step 2110 is equal to or smaller than the predetermined threshold value. If yes, control is passed to step 2130. Otherwise, control is passed to step 2140. Step 2130 selects all the patterns having the minimum state input error equal to or smaller than the threshold value and then control is passed to step 2150. Step 2140 selects only the pattern having the smallest minimum state input error and control is passed to step 2150. Next, step 2150 initializes (set 1=1) the I as the value indicating the number of repetitions of the steps 2160. Step 2160 learns the model operation method of the model 230 by using the re-learning part 265 of the learning part 260 corresponding to the selected pattern. Step 2170 is a branching section. If I is equal to or smaller than the selected number of patterns, I is incremented by 1 and control is returned to step 2160 and learning is performed by using the re- learning part 2 65 corresponding to the another selected pattern. Step 2180 decides the appropriate operation method among the re-learning results obtained by the selected pattern by using the learning result determination part 300 of the control apparatus 200 and outputs it to the operation signal generation part 280 of the control apparatus 200. Then, control is passed to the step to terminate the operation of re-learning. With the aforementioned operation, re- learning can learn the operation method equivalent to the pre-learning result in a short time by using the pattern selected according to the pre-learning result for the current state input. Moreover, it is possible to input the threshold value of the minimum state input error as the set parameter via the display 620. Next, a detailed explanation will be given on the re-learning algorithm of step 2160 with reference to the flowchart of Fig. 12. As shown in Fig. 12, the flowchart of the operation of the re-learning algorithm is executed by a combination of steps 2200, 2210, 2220, 2230, 2240, 2250, 2260, 2270, 2280, 2290, and 2300. Hereinafter, each of the steps will be detailed. Step 2200 initializes (set J = 1) the number of initializations J as a value indicating the number of repetitions of steps 2210 to 2280. Next, step 2210 initializes the state input to an input value to be extracted from the measurement signal 3. Step 2220 initializes (K = 1) the number of repetitions K as a value indicating the number of repetitions of steps 2230 to 2270. Next, step 2230 obtains the operation amount of the patterned model input by using re-learning part 265 of the learning part 260 corresponding to the pattern to be used. Step 2240 linearly interpolates the operation amount of the patterned model input by using the pattern transforming part 264 of the learning part 260 so as to obtain the operation amount. Step 2250 adds the operation amount to the state input by using the model input generation part 261 of the learning part 260 so as to operate the model and update the state input. Step 2260 patterns the state input according to the pattern data to be used, by using the pattern transforming part 264 of the learning part 260. Step 2270 learns the model operation method by using the neural network and various learning algorithms such as reinforced learning according to the model output obtained as a result of the model operation. Step 2280 is a branching section where the number of operations K is compared to a predetermined threshold value. If K is smaller than the threshold value, K is incremented by 1 and control is returned to step 2230. Otherwise, control is passed to step 2290. Step 2290 compares the number of initializations J to a predetermined threshold value. If J is smaller than the threshold value, J is incremented by 1 and control is returned to step 2210. Otherwise,, control is passed to step 2300. Step 2300 transmits the learned result to the learning information database 270 of the control apparatus 200 so as to be stored in it and control is passed to the step to terminate the operation of re- learning. With the aforementioned operation, the re- learning uses the patterned state input to learn the operation method when the current state input is used as the initial value. This completes the detailed explanation of the operations of the steps 1010, 1020, and 1040 shown in Fig. 4. Next, explanation will be given on the screen displayed on the display 620 for displaying the data processing apparatus output signal 64 outputted from the data processing apparatus 610 constituting the plant control system 200 according to the embodiment of the present invention. Fig. 13 to Fig. 20 show examples of the screen displayed on the display 620 attached to the plant control system 200 shown in Fig. 1. Fig. 13 shows an example of an initial screen. By operating a mouse 602 to superimpose the cursor on a button and clicking the mouse 602 while the screen of Fig. 13 is displayed on the display 620, it is possible to select (press) the button. When the button 701 of "numerical calculation", the button 702 of "database reference", the button 703 of "patterning/learning", and the button 704 of "operation execution" are selected, the screens of Fig. 14, Fig. 15, Fig. 16, and Fig. 17 are displayed on the display 620, respectively. Moreover, when the button 705 of "end" is selected, the initial screen is terminated. Fig. 14 shows the case when the button 711 of "calculation condition setting" is selected so as to input/set various calculation conditions required for executing the calculation by the numerical calculation execution part 220 of the control apparatus 200. Moreover, when the button 712 of "numerical calculation execution" is selected, it is possible to start calculation by the numerical calculation execution part 220. Moreover, when the button 713 "return" is selected, control is passed to Fig. 13. On the screen of the display 620 shown in Fig. 15, it is possible to select a database containing information to be displayed. By selecting the button 721 of "measurement signal database", the button 722 of "operation signal database", the button 723 of "numerical calculation database", the button 724 of "learning information database", the button 725 of "control logic database", and the button 726 of "pattern database", it is possible to access the measurement signal database 210, the operation signal database 250, the numerical calculation database 240, the learning database 270, the control logic database 290, and the pattern database 500, respectively. It is possible to display the information in the respective databases on the display 62 0 and add new information to the information in the database and to modify or delete information in the database. Moreover, by selecting the button 727 of "return", the screen is returned to Fig. 13. On the screen of Fig. 16, when the button 731 of "pre-learning" is selected, the pre-learning screen shown in Fig. 18 appears. Moreover, when the button 732 of "re- learning" is selected on the screen of Fig. 16, the re- learning screen shown in Fig. 19 appears on the display 620. When the button 733 of "return" is selected, the screen of Fig. 13 appears. The screen of Fig. 17 shows the model input 741 before operation, the model output 742, the model input 743 after the operation, model output 744, and the learned operation amount 745 as a guidance display, so that the plant operation staff can select whether to execute an operation on the guidance display. When an operation is to be executed, the button 746 of "yes" is selected. Otherwise, the button 747 of "no" is selected. On the display screen shown in Fig. 1, the operation staff can confirm the relationship between the pattern and the operation signal before operating the plant, which increases the reliability of the plant operation.. That is, the plant control method shown in Fig. 17 can be detailed as follows. In the plant control method of the control apparatus for controlling the plant by using a model simulating the plant, as shown in Fig. 8, the control apparatus has a pattern database containing the state input pattern which patterns a plurality of operation signals to be inputted to the model. Moreover, as shown in Fig. 7, the control apparatus includes a learning information database containing a plurality of operation signals to be inputted to the model and the output from the model. As shown as the model input 743 and the model output 744 after the operation, the plurality of operation signals stored in the learning information database and the output from the model are outputted to the display. As shown as the learned operation amount 745, the operation signal to the plant according to the plurality of operation signals inputted to the model and the state input pattern stored in the pattern database are superimposed and outputted to the display. As shown as the button 74 6 to be selected when an operation is to be executed, the plant is controlled by the operation signal to the plant according to permission of operation on the plant. In Fig. 18, when the button 751 of "learning start" is selected, the pre-learning part 262 of the learning part 260 and the model 230 of the control apparatus 200 are operated according to the flowchart shown in Fig. 5 so as to execute the pre-learning of the operation method. Moreover, when the button 742 of "pattern generation" is selected the pattern generation screen shown in Fig. 20 appears on the display 620. When the button 753 of "return" is selected, control is returned to Fig. 16. In Fig. 19, it is possible to input the set value of the minimum state input error threshold value to be used in the flowchart of Fig. 11, into the data input column 761. After the set value is inputted in the data input column 7 61, if the button 7 62 "execution" is selected, the learning part 260 of the control apparatus 200 and the model 230 are operated according to the flowchart shown in Fig. 11 so as to execute re- learning. Moreover, if the button 763 of "return" is selected, control is returned to Fig. 16. In Fig. 20, it is possible to input a set value of the patterning error threshold value used in the flowchart shown in Fig. 9 into the data input column 771, and the set value of the pattern input dimension threshold value into the data input column 772. If the button 773 of "execution" is selected after the setting values are inputted into the data input columns 771 and 772, the pattern generation part 400 of the control apparatus 200 is operated according to the flowchart shown in Fig. 6 so as to execute pattern generation. Moreover, if the button 774 of "return" is selected, control is returned to Fig. 18. This completes the explanation on the screens displayed on the display 620. As has been described above with reference to the flowcharts in Fig. 11 and Fig. 12, in the embodiment of the present invention, the state input inputted to the learning part of the plant control system is patterned with a smaller number of inputs than the model input dimension so as to reduce the model search space, thereby enabling a high-speed learning. Moreover, a desired learning result can always be obtained by generating a pattern by using a pre-learning result without reducing the input dimension in advance before start of the plant operation as has been detailed with reference to the flowchart in Fig. 5, and comparing the similarity of the current state input to the state input contained in the pre-learning result after start of the plant operation so that a pattern generated from the state input having a higher similarity is selected for learning the operation method as has been detailed with reference to the flowchart of Fig. 11. Moreover, as has been detailed with reference to the flowchart of Fig. 9, it is possible to automatically generate a pattern input dimension and pattern information by an evolutionary search algorithm., which enables pattern generation not requiring any manual operation while preventing an erroneous patterning. Furthermore, the plant operation staff can visually check the state of the patterning and the effect of the patterned learning because the apparatus has the function to reference the information stored in the respective databases on the display and the function to input setting parameters to be used in the learning part and the pattern generation part via the display as has been detailed with reference to Fig. 13 to Fig. 20. Next, explanation will be given on a case that the plant control system 200 as the embodiment of the present invention shown in Fig. 1 to Fig. 20 is applied to a thermal power plant 100 having a boiler. It should be noted that the plant control system 200 may be used for control of other plants than the thermal power plant. Fig. 21 shows an outline of the thermal power plant having a boiler. Firstly, explanation will be given on the mechanism of the thermal power plant having the boiler. In the thermal power plant shown in Fig. 21, coal as fuel is crushed into fine coal powder by a mill 110 and introduced via a burner 102 into a boiler 101 together v/ith a primary air for feed of the coal powder and a secondary air for combustion adjustment. The coal as the fuel burns in a furnace of the boiler 101. The coal as the fuel and the primary air are introduced via a pipe 134 while the secondary air is introduced via a pipe 141 into the burner 102. Moreover, the after-air for two-stage combustion is introduced into the boiler 101 via an after-air port 103 installed in the boiler 101. The after-air is introduced into the after-air port 103 from a pipe 142. The high-temperature combustion gas generated by combustion of the fuel coal inside the furnace of the boiler flows downward along a route shown by arrows in the furnace of the boiler 101 and passes through a heat exchanger 106. After the heat exchange, the combustion exhaust gas is discharged from the boiler 101 and flows downward toward an air heater 104 installed outside the boiler 101. The combustion exhaust gas which has passed through the air heater 104 comes into an exhaust gas treating device (not depicted). After harmful substances are removed from the combustion exhaust gas, the exhaust gas is discharged into the atmosphere through a chimney. Water circulating in the boiler 101 is introduced from a condenser (not depicted) installed in a turbine 108 into the boiler 101 via a water feed pump 105. The water is heated in the heat exchanger 106 by the combustion gas flowing downward in the furnace of the boiler 101 and becomes vapor of a high temperature. It should be noted that in this embodiment, only one heat exchanger is depicted. However, it is also possible to arrange a plurality of heat exchangers. The vapor of a high temperature and a high pressure generated in the heat exchanger 106 is introduced via a turbine governor valve 107 to the vapor turbine 108 so that energy of the vapor drives the vapor turbine 108 to rotate a generator 109 coupled to the vapor turbine 108 and generate electricity. Next, explanation will be given on routes of the primary air and the secondary air introduced into the furnace of the boiler 101 from the burner 102 installed in the furnace of the boiler 101 and the route of the after-air introduced into the furnace of the boiler 101 from the after-air port 103 installed in the furnace of the boiler 101. The primary air is introduced from a fan 120 into a pipe 130 which is branched to a pipe 132 which passes through the air heater 104 and a pipe 131 which bypasses the pipe 132 and the air heater 104. The primary air which has flown downward through the pipes 132 and 131 is again mixed in a pipe 133 and introduced into the mill 110. The air which passes through the air heater 104 is heated by the combustion exhaust gas discharged from the furnace of the boiler 101. The coal (fine coal powder) generated in the mill 110 by using the primary air is fed through the pipe 133 to the burner 102. The secondary air and the after-air are introduced from a fan 121 into a pipe 140 and flows downward in the pipe 140 which passes through the air heater 104, so as to be heated. After the heating, the pipe 140 is branched into a pipe 141 for the secondary air and a pipe 142 for the after-air, which are respectively introduced to the burner 102 and the after-air port 103 installed in the furnace of the boiler 101. The control apparatus 200 for the thermal power plant 100 having the boiler according to the present embodiment has the function to adjust the air amount to be introduced from the burner 102 into the boiler 101 and the air amount to be introduced from the after-air port 103 into the boiler 101 so as to reduce the NOx and CO concentrations in the exhaust gas from the boiler. The thermal power plant 100 includes various measurement devices for detecting the operation state of the thermal power plant. A measurement signal obtained from the measurement devices is sent as a measurement signal 1 to the control apparatus 200. The measurement devices for detecting the operation state of the thermal power plant 100 may be, for example, a flowmeter 150, a thermometer 151, a pressure indicator 152, a generated electricity measuring device 153, and a concentration measuring device 154 for measuring O2 concentration and/or CO concentration as shown in Fig. 2. The flowmeter 150 measures the flow rate of water supplied from a water feed pump 105 to the boiler 101. Moreover, the thermometer 151 and the pressure indicator 152 respectively measure the temperature and the pressure of the vapor generated by the heat exchange with the combustion gas flowing downward in the boiler 101 in the heat exchanger 106 in the boiler 101 and supplied to the vapor turbine 108. The generated electricity measuring device 153 measures the electricity generated by a generator 109 rotated by the vapor turbine 108 which is driven by the vapor generated by the heat exchanger 106. Moreover, information on the concentrations of the components (such as CO and NOx) contained in the combustion gas flowing downward in the boiler 101 is measured by the concentration measuring device 154 arranged at the exit of the boiler at the downstream side of the boiler 101 so as to measure the O2 concentration and/or the CO concentration. It should be noted that a plenty of other measuring devices than those shown in Fig. 2 are arranged on the thermal power plant 100 but they are not depicted. Fig. 22 is a partial enlarged diagram showing the air heater 104 arranged at the downstream side of the boiler 101 constituting the thermal power plant 100 and the pipes arranged in the air heater 104. As shown in Fig. 22, air dampers 162, 163, 161, 160 are respectively arranged in a pipe 131 which bypasses the secondary air pipe 132, the after-air pipe 142, the pipe 132 arranged inside the air heater 104, and the air heater 104 which are branched at the downstream side of the pipe 140 arranged inside the air heater 104. By operating the air dampers 160 to 163, it is possible to change the area where the air passes inside the pipes 131, 132, 141, 142 so as to separately adjust the air flow rate in the pipes 132, 132, 141, and 142. By using the operation signal 15 outputted to the thermal power plant 100 by the control apparatus 200 which controls the thermal power plant 100, it is possible to operate the water feed pump 105, the mill 110, the air dampers 160, 161, 162, 163, and other devices. It should be noted that in the control apparatus 200 for the thermal power plant 100 according to the embodiment, devices for adjusting the state amount of the thermal power plant such as the water feed pump 105, the mill 110, the air dampers 160, 161, 162, 163 are called operation ends and instruction signals required for operating them are called operation signals. Fig. 23 is a flowchart showing the process of the thermal power plant control method in the control apparatus 200 of the thermal power plant 100 having the boiler in the embodiment of the present invention shown in Fig. 21. In Fig. 23, the control apparatus 200 for the thermal plant 100 having the boiler in the embodiment executes control of the thermal power plant 100 by combining the steps 2181 and 2182. Step 2181 references the learning data obtained by each of the patterns selected by the pattern selection part 263 constituting the learning part 260 of the control apparatus 200 which controls the thermal power plant 100, compares the total of improvement values of the CO and NOx as the model output amount after the operation for current state input, and selects the maximum total of the improvement values. Moreover, when the number of patterns selected by the pattern selection part 263 is 1, control is passed to step 2182 without performing the aforementioned process. Step 2182 transmits the learning data 12 of the pattern selected by the pattern selection part 263 to the operation signal generation part 280 and passes control to step to terminate the operation of the learning result determination part 300 constituting the control apparatus 200. With the aforementioned operation, the control apparatus 200 uses the learning data to obtain the best improvement values of CO and COx as the result of re-learning so as to generate an operation signal. Fig. 24 shows an example of the screen displayed on the display 620 in the control apparatus 2 00 of the thermal power plant 100 having the boiler according to the embodiment of the present invention. Fig. 24 shows an example of screen which shows the air amount of the burner 102 of the boiler 101 used as the model input 7 inputted to the model 230 constituting the control apparatus 200, the air amount of the air port 103, guidance display of the concentrations of CO, NOx used in the model output 8 by simulation-calculating the thermal power plant to be controlled by the model 230 before and after the operation, and the operation amounts of the burner 102, and the air port 103 learned by the learning part 265 by the patterning of the pattern transforming part 264. The model input 7 displays distribution of the air amount of the burner 102/air port 103 arranged before and after a tank of the boiler 101 while correlating them with the boiler structure diagram. It should be noted that in Fig. 25, the burner 102 and the air port 103 are arranged before and after the tank of the boiler 101 respectively as one stage each having five air entrances. However, the numbers can be modified to an arbitrary value. The operation staff of the plant can check the control effect obtained by the patterned operation amount and the learned operation method to decide presence/absence of the operation while watching the screen shown in Fig. 24. Moreover, as shown in Figs. 25A to 25C, patterning in the control apparatus 200 for the thermal power plant 100 having the boiler according to the embodiment of the present invention may be realized, for example, as (1) grouping of the operation ends for each of the burner 102 and the air port 103 of the boiler 101 (Fig. 25A) , (2) grouping of the operation ends before/after the tank of the boiler 101 (Fig. 25B), and (3) grouping of all the operation ends of the boiler 101 (Fig. 25C). As is clear from the aforementioned, the operation staff of the thermal power plant 100 having the boiler can control the thermal power plant by selecting the appropriate patterning method by considering the type of the operation end, the number of operation ends, and the combustion characteristic of the boiler of the thermal power plant to be controlled. According to the embodiment of the present invention,, it is possible to provide a plant control system, which corrects the model according to the measurement signal which has measured the operation state of the plant, executes the high-speed learning so as to perform re-learning by using the corrected model and correct the plant control algorithm, thereby accurately control the plant. The present invention can be applied to a plant control system and in particular, to a control apparatus for a thermal power plant. It should be further understood by those skilled in the art that although the foregoing description has been made on embodiments of the invention, the invention is not limited thereto and various changes and modifications may be made without departing from the spirit of the invention and the scope of the appended claims. CLAIMS: 1. A plant control system for controlling a plant including an operation signal generation part which calculates an control instruction operation signal for a plant to be controlled by using a measurement signal which has measured an operation state of the plant, the apparatus comprising: an operation signal database which stores the measurement signal of the measured plant; an operation signal database which stores an operation signal for the plant; a numerical calculation execution part which calculates the operation characteristic of the plant; a numerical calculation database which stores the numerical calculation result from the numerical calculation execution part; a simulation model which simulates control characteristic of a plant when the operation signal is supplied to the plant according to information on the numerical calculation result from the numerical calculation database; a learning part which learns the plant operation method by using the model; a learning information database which stores learning information data on the learning part; a control logic database which contains information used for the operation signal outputted from the operation signal generation part; a pattern generation part which generates pattern data expressing a state input based on the learning information data in the learning part with a smaller input number than the model input dimension; a pattern database which stores the pattern data generated in the pattern generation part; and a learning result determination part which selects a learning result having a preferable control effect from the learning result obtained by using a plurality of patterns; wherein the operation signal is calculated from the operation signal generation part according to the learning result selected by the learning result determination part. 2. The plant control system as claimed in claim 1, wherein the learning part includes: a model input generation part which generates an input to be applied to the model and extracts a model input from a plant measurement signal; a pre-learning part which uses the learning information data stored in the learning information database before start of the plant operation so as to learn the model input generation part so that the model output simulated/calculated by the model achieves its target value; a pattern database which contains a pattern generated by using the pattern generation part for the learning information data obtained by the pre-learning by the pre-learning part; a pattern selection part which selects a pattern similar to the measurement signal from the pattern data stored in the pattern database after start of the plant operation; a pattern transforming part which patterns the state input to be inputted to the learning part by using the selected pattern, linearly interpolates the inter-inputs for the learned and patterned operation amount; and a re-learning part which performs learning with a reduced input dimension of the model input by using the selected pattern so that the model output simulated/calculated in the model by using the learning information data stored in the learning information database after start of the plant operation achieves its target value. 3. The plant control system as claimed in claim 1, the pattern generation part constituting the control apparatus has: a function of performing repeated search while successively incrementing the input dimension by 1 starting from the input dimension 1 until the end condition is satisfied in order to generate pattern data for a model operation amount contained in the learning information data learned by using the pre- learning part before start of the plant operation; and a function of generating a plurality of pattern data candidates as the pattern data search part, performing a genetic operation such as mixing and mutation on them, and repeating the search so as to search for the pattern data. 4. The plant control system as claimed in claim 2, wherein the learning part constituting the control apparatus compares the current state input to the state input contained in the pre-learning result in the pattern selection part after the plant operation start and selects the pattern generated from the state input having the higher similarity so as to learn the operation method. 5. The plant control system as claimed in claim 1, wherein the control apparatus includes a data processing device and a display which displays information stored in a measurement signal database, an operation signal database, a control logic database, a learning information database, a numerical calculation database, and a pattern database. 6. The plant control system as claimed in claim 5, wherein setting parameters used in the learning part and the pattern generation part constituting the control apparatus are inputted via an input part connected to the data processing apparatus. 7. The plant control system as claimed in claim 1, wherein the plant to be controlled is a thermal power plant including a boiler configured as follows: operation ends to which the plant operation signal is outputted are a burner which generates a high- temperature gas by a combustion reaction using fossil fuel and air and an air port which supplies air to the combustion gas from the rear side of the flow of the combustion gas generated by mixing the fossil fuel and the air supplied from the burner; the measurement signal of the measured plant is a measurement signal of the concentration of carbon monoxide and the concentration of nitrogen oxide obtained by measuring the carbon monoxide concentration and nitrogen oxide concentration of the fuel gas burned in the boiler by using a measurement device; the learning result determination part constituting the control apparatus selects such learning information data by a selected pattern that the total of the improvement values indicating the differences between the concentrations of the carbon monoxide and the concentrations of the nitrogen oxide in the combustion gas before and after the operation is maximum so as to generate the operation signal. 8. The plant control system as claimed in claim 7, wherein the state input parameter to be patterned in the learning result judging part constituting the control apparatus is the air flow rate introduced into the boiler separately from the burner and the air port arranged in the boiler. 9. The plant control system as claimed in claim 7, wherein the learning result judging part constituting the control apparatus uses one of the following methods: grouping of state inputs to be patterned; grouping for the burner and the air port, respectively; grouping of the entire boiler. 10. A plant control method for a control apparatus which controls a plant by using a model simulating the plant, the method comprising steps of: providing a pattern base containing a state input pattern in which a plurality of operations signals to be inputted to the model by the control apparatus are patterned; providing a learning information database containing a plurality of operation signals to be inputted to the model and an output from the model; outputting the plurality of operation signals stored in the learning information database and the output from the model to a display; outputting to the display, an operation signal to the plant based on the plurality of operation signals inputted to the model and the state input pattern stored in the pattern data base while superimposing them on each other; and controlling the plant by the operation signal to the plant according to permission of the operation execution on the plant. A plant control system includes: a numerical calculation execution part which calculates the operation characteristic of the plant; a model for simulating the plant control characteristic according to information on the numerical calculation result; a learning part which learns the plant operation method by using the model; a learning information database which stores learning information data on the learning part; a pattern generation part which generates pattern data expressing a state input based on the learning information data in the learning part with a smaller input number than the model input dimension; a pattern database which stores the pattern data generated in the pattern generation part; and a learning result determination part which selects a learning result having a preferable control effect from the learning result obtained by using a plurality of patterns.

Full Text

BACKGROUND OF THE INVENTION
The present invention relates to a plant
control system and in particular or system, to a
thermal power plant.
The plant control system processes a
measurement signal of an operation state amount
obtained from a plant to be controlled, calculates an
operation signal to be supplied to the plant to be
controlled, and transmits it as a control instruction
to the control system.
The plant control system uses an algorithm
for calculating the operation signal so that the
measurement signal of the plant operation state amount
satisfies its target value.
It is possible to use the PI
(proportion/integration) control algorithm for
controlling the plant. In the PI control, the
deviation of the plant measurement signal from its
target value is multiplied by the proportion gain. The
obtained value is added by the value obtained by time-
integrating the deviation so as to obtain the operation
signal to be supplied to the object to be controlled.
On the other hand, in order to automatically
correct the control algorithm in accordance with a
change of the plant operation state and the

environment, the plant may be controlled by using the
adaptive control and the learning algorithm.
JP-A-2000-35956 discloses a control apparatus
using a reinforced learning method using a model as a
method for obtaining the operation signal of the
control apparatus for controlling the plant by using
the learning algorithm.
In the method using the technique of the
reinforced learning method, the control apparatus
includes a model for predicting the characteristic of
the object to be controlled and a learning part for
learning the method for generating such a model input
that a model output can achieve the target value.
By inputting the model input learned by the
learning part to the model, it is possible to obtain
the effect that the model output approaches the target
value.
The learning type adaptive control corrects
the model by using the measurement signal which has
measured the plant operation state and performs re-
learning by using the corrected model so as to perform
online correction/modification of the control algorithm
for accurately controlling the plant.
Accordingly, it is necessary to correct the
model within a cycle (control cycle) during which the
operation signal outputted from the control apparatus
is modified for the plant and use the corrected model
for the re-learning so as to complete the learning and

correct/modify the control algorithm.
The control cycle can be considered as a time
from the plant operation completion to the moment when
the plant enters a static state, which is normally
several minutes or several tens of minutes.
For example, when controlling a complicated
plant such as a thermal power plant, the model input
dimension is several tens or several hundreds and the
number of combinations of the learning model inputs
(search space) increases, which in turn requires a long
learning time. As a result, it becomes difficult to
online-correct/modify the control algorithm required
for accurately controlling the plant.
Consequently, in order to complete the
learning within the control cycle, it is necessary to
correct the model in accordance with increase of the
model input dimension and increase the learning speed
of re-learning by using the corrected model.
JP-A-7-160661 discloses a technique for
increasing the learning speed in a neural network
learning by classifying teacher data into a plurality
of patterns according to the combinations of the plant
measurement information and extracting a learning
pattern according to the control result so as to learn
only the teacher data on the extracted pattern.
SUMMARY OF THE INVENTION
By using the technique of the reinforced

learning method using the model disclosed in JP-A-2000-
35956, it is possible to automatically learn the method
for generating an operation signal which can achieve
the control target.
However, when a model is corrected according
to the measurement signal which has measured the plant
operation state and the corrected model is used for re-
learning so as to correct the control algorithm, the
model input dimension increases in a complicated plant,
which makes it difficult to perform the learning within
the control cycle for controlling the plant.
Moreover, when the technique disclosed in JP-
A-7-160661 is used, it is possible to divide the
teacher data by patterning so as to reduce the search
space and increase the learning speed. That is, even
if the model input dimension increases and the search
space (number of teacher data) increases, it is
possible to perform learning within the control cycle
by executing the appropriate patterning.
However, when the teacher data to be
classified is biased, the number of teacher data may be
too small depending on the learning pattern and it may
not be possible to acquire a desired learning result.
Moreover, since the type of patterns to be
classified is decided according to the knowledge of a
designer of the control apparatus, there is a problem
that the pattern generation requires a large load on
the operator.

It is therefore an object of the present
invention to provide a plant control system which
corrects a model in accordance with a measurement
signal which has measured a plant operation state and
uses the corrected model for re-learning, which enables
execution of the high-speed learning for correcting the
plant control algorithm and accurately controlling the
plant.
A plant control system for controlling a
plant including an operation signal generation part
which calculates an control instruction operation
signal for a plant to be controlled by using a
measurement signal which has measured an operation
state of the plant, apparatus comprising: an operation
signal database which stores the measurement signal of
the measured plant; an operation signal database which
stores an operation signal for the plant; a numerical
calculation execution part which calculates the
operation characteristic of the plant; a numerical
calculation database which stores the numerical
calculation result from the numerical calculation
execution part; a simulation model which simulates
control characteristic of a plant when the operation
signal is supplied to the plant according to
information on the numerical calculation result from
the numerical calculation database; a learning part
which learns the plant operation method by using the
model; a learning information database which stores

learning information data on the learning part; a
control logic database which contains information used
for the operation signal outputted from the operation
signal generation part; a pattern generation part which
generates pattern data expressing a state input based
on the learning information data in the learning part
with a smaller input number than the model input
dimension; a pattern database which stores the pattern
data generated in the pattern generation part; and a
learning result determination part which selects a
learning result having a preferable control effect from
the learning result obtained by using a plurality of
patterns; wherein the operation signal is calculated
from the operation signal generation part according to
the learning result selected by the learning result
determination part.
According to the present invention, it is
possible to provide a plant control system which
corrects a model according to a measurement signal
which has measured the plant operation state and
performs re-learning at a high speed by using the
corrected model, thereby correcting the plant control
algorithm so as to accurately control the plant.
Other objects, features and advantages of the
invention will become apparent from the following
description of the embodiments of the invention taken
in conjunction with the accompanying drawings.

BRIEF DESCRIPTION OF DRAWINGS
Fig. 1 is a control block diagram showing an
entire configuration of a plant control system
according to an embodiment of the present invention.
Fig. 2 is control block diagram indicating a
learning part of a plant control system according to
the embodiment of the present invention shown in Fig.
1.
Fig. 3 explains the patterning operation in
the plant control system according to the embodiment of
the present invention shown in Fig. 1.
Fig. 4 is a flowchart showing the operation
of the plant control system according to the embodiment
of the present invention shown in Fig. 1.
Fig. 5 is a flowchart showing the pre-
learning operation in the flowchart showing the
operation of the plant control system according to the
embodiment of the present invention shown in Fig. 4.
Fig. 6 is a flowchart showing the pattern
generation operation in the flowchart showing the
operation of the plant control system according to the
embodiment of the present invention shown in Fig. 4.
Fig. 7 explains an example of the information
stored in a learning information database in the plant
control system according to the embodiment of the
present invention shown in Fig. 1.
Fig. 8 explains an example of the information
stored in a pattern database in the plant control

system according to the embodiment of the present
invention shown in Fig. 1.
Fig. 9 is a flowchart for detailed
explanation of the pattern search algorithm as a part
of the pattern generation operation in the flowchart
showing the operation of the plant control system
according to the embodiment of the present invention
shown in Fig. 4.
Fig. 10 explains the pattern search algorithm
for calculating a patterning error in the flowchart
showing the operation of the plant control system
according to the embodiment of the present invention
shown in Fig. 9.
Fig. 11 is a flowchart showing the re-
learning operation in the flowchart showing the
operation of the plant control system according to the
embodiment of the present invention shown in Fig. 4.
Fig. 12 is a flowchart showing a detailed
explanation of a part of the re-learning operation in
the flowchart showing the operation of the plant
control system according to the embodiment of the
present invention shown in Fig. 4.
Fig. 13 shows an example of an initial screen
displayed on a display mounted on the plant control
system according to the embodiment of the present
invention shown in Fig. 1.
Fig. 14 shows an example of a screen
displayed on the display when a numerical calculation

is executed in the plant control system according to
the embodiment of the present invention shown in Fig.
1.
Fig. 15 shows an example of a screen
displayed on the display when a database is referenced
in the plant control system according to the embodiment
of the present invention shown in Fig. 1.
Fig. 16 shows an example of a screen
displayed on the display when patterning and learning
are executed in the plant control system according to
the embodiment of the present invention shown in Fig.
1.
Fig. 17 shows an example of a screen
displayed on the display when an operation is executed
in the plant control system according to the embodiment
of the present invention shown in Fig. 1.
Fig. 18 shows an example of a screen
displayed on the display when the pre-learning and the
pattern generation are executed in the plant control
system according to the embodiment of the present
invention shown in Fig. 1.
Fig. 19 shows an example of a screen
displayed on the display when re-learning is executed
in the plant control system according to the embodiment
of the present invention shown in Fig. 1.
Fig. 20 shows an example of a screen
displayed on the display when pattern generation is
executed in the plant control system according to the

embodiment of the present invention shown in Fig. 1.
Fig. 21 shows a configuration outline of a
thermal power plant including boiler to be controlled
by employing the plant control system according to the
embodiment of the present invention shown in Fig. 1.
Fig. 22 is an enlarged view of an air heater
portion in the boiler of the thermal power plant to be
controlled shown in Fig. 21.
Fig. 23 is a flowchart showing a part of the
re-learning process in the control apparatus of the
thermal power plant having the boiler according to the
embodiment of the present invention shown in Fig. 21.
Fig. 24 is an example of a screen shown on
the display when an operation is performed for the
control apparatus of the thermal power plant having the
boiler according to the embodiment of the present
invention shown in Fig. 1.
Figs. 25A-25C explain examples of patterned
grouping in the control apparatus of the thermal power
plant having the boiler according to the embodiment of
the present invention shown in Fig. 1.
DETAILED DESCRIPTION OF THE EMBODIMENTS
Description will now be directed to a plant
control system according to an embodiment of the
present invention with reference to the attached
drawings.

[Embodiment 1]
Fig. 1 is a block diagram showing the entire
configuration of the plant control system according to
an embodiment of the present invention.
In Fig. 1, the plant of the present
embodiment is a plant 100 constituting a thermal power
plant including a boiler using coal as fuel. The
thermal power plant 100 including the boiler is
configured so as to be controlled by a control
apparatus 200.
The control apparatus 200 for controlling the
plant 100 as a control object includes, as calculation
apparatuses, a numerical calculation part 220, a model
230, a learning part 260, an operation signal
generation part 280, a learning result judging part
300, and a pattern generation part 400.
The control apparatus 200 includes as
databases, a measurement signal database 210, a
numerical calculation database 240, an operation signal
database 250, a learning information database, a
control logic database 290, and a pattern database 500.
Furthermore, the control apparatus 200
includes, as an external interface, an external input
interface 201 and an external output interface 202.
The control apparatus 200 acquires the
measurement signal 1 which has measured various state
amounts of a thermal power plant from the thermal power
plant 100 via the external input interface 291.

Moreover, the control apparatus 200 sends an
operation signal 15 for controlling, for example, a
burner of the boiler and an air flow rate of an air
port via the external output interface 202 to the plant
100 as the thermal power plant of the control object.
In the control apparatus 200, the measurement
signal 1 which has measured various state amounts of
the plant 100 is sent to the external input interface
201 and stored as the measurement signal 2 in the
measurement signal database 210 as a database of the
control apparatus 200.
An operation signal 14 generated by an
operation signal generation part 280 as a calculation
apparatus provided in the control apparatus 200 is sent
to the external output interface 202 and stored in the
operation signal database 250 as a database provided in
the control apparatus 200.
The operation signal generation part 280 uses
control logic data 13 stored in the control logic
database 290 and learning data 12 outputted from the
learning result determination part 300 as a calculation
apparatus provided in the control apparatus 200 so as
to generate the operation signal 14 functioning as the
operation signal 15 for the plant 100, so that the
measurement signal 1 of the plant 100 achieves an
operation target value and outputs the signal to the
external interface 202.
The control logic database 290 contains a

control circuit and a control parameter used for
calculating the control logic data 13 so as to output
the control logic data.
The known PI control can be used in the
control circuit for calculating the control logic data
13.
The learning data stored in the learning
information database 270 as a database provided in the
control apparatus 200 is generated by a learning part
260 provided in the control apparatus 200.
The learning part 260 is connected to the
model 230 as a calculation apparatus provided in the
control apparatus 200.
The model 230 has a function for simulating
the control characteristic of the plant 100 as the
thermal power plant. That is, the model 230 performs a
calculation for simulation of the following: the
operation signal 15 serving as a control instruction is
supplied to the plant 100 so as to obtain the
measurement signal 1 as the control result.
For this simulation calculation, a model
input 7 for operating the model 230 is received from
the learning part 260 and the model 230 performs
simulation calculation of the characteristic change by
control of the plan so as to obtain a model output 8 as
the result of the simulation calculation.
Here, the model output 8 is a predicted value
of the measurement signal 1 of the plant 100.

The model 230 is configured according to the
numerical calculation result 6 stored in the numerical
calculation database 240 as a database provided in the
control apparatus 2 00.
The numerical calculation part 220 as a
calculation apparatus provided in the control apparatus
200 analyzes the operation characteristic of the plant
100 by using a physical model simulating the plant 100.
The calculation result 4 obtained by
analyzing the characteristic of the plant 100 by using
the physical model in the numerical calculation part
220 is stored in the numerical calculation database
240.
The model 230 calculates a model output 8
corresponding to the model input 7 by using information
stored in the numerical calculation database 240 and
the measurement signal database 210 and employing the
statistical method such as the neural network.
The model 230 extracts the numerical
calculation result 6 required for calculating the model
output 8 corresponding to the model input 7 from the
numerical calculation database 240 by using the model
information data 5 and interpolates the result.
Moreover, the model 230 can be corrected by
using the measurement signal 3 stored in the
measurement database 210 so that the control
characteristics of the model 230 and the plant 100
coincide with each other when the characteristic of the

physical model of the numerical calculation part 220 is
different from the characteristic of the plant 100.
The learning part 260 learns a method for
generating the model input 7 so that the model output 8
which is simulation-calculated by the model 230
achieves the model output target value set in advance
by an operation staff.
Here, the current model input is inputted to
the learning part 2 60 so as to output its model input
change width. Here, the input to the learning part 2 60
will be called a state input and the output will be
called an operation amount.
The constriction condition used for learning
and the learning information data 9 containing the
model output target value are stored in the learning
information database 270.
The learning part 260 directly learns the
state input before start of the plant operation and,
after the start of the plant operation, performs
learning by patterning the state input by using the
pattern data stored in a pattern database 500 which
will be detailed later.
Moreover, the learning data 10 obtained as a
result of the learning in the learning part 260 is
stored in the learning information database 270
provided in the control apparatus 200.
The learning data 10 includes the state input
before the model input modification, the operation

amount at the state input, and information on the model
output change width obtained as a result of the
operation. It should be noted that the functions of
the learning part 260 will be detailed later.
The pattern generation part 400 as a
calculation apparatus provided in the control apparatus
200 generates pattern data 17 by using the learning
data 16 learned before the start of the plant operation
in the learning part 260 and stores the pattern data 17
in the pattern database 500.
After operation of the plant 100 is started,
optimal pattern data 18 similar to the measurement
signal 3 is selected from the pattern data stored in
the pattern database 500 and the learning part 260
patterns the state input contained in the measurement
signal 3.
The learning data 11 obtained by the learning
in the learning part 260 and stored in the learning
information database 270 is inputted to a learning
result determination part 300 as a calculation
apparatus provided in the control apparatus 200.
When a plurality of pattern data 18 are used
for the learning, the learning result determination
part 300 selects the learning data which makes the
control effect to be maximum from the respective
pattern learning data 11. When only one pattern is
used, the learning data on the pattern is selected.
The learning data 12 selected by the learning

result determination part 300 is inputted the operation
signal generation part 280.
The operation staff of the plant 100 can
access information stored in various databases provided
in the control apparatus 200 by using an external input
device 600 formed by a keyboard 601 and a mouse 602, a
data processing apparatus 610 having a data
send/receive part 612 which can send and receive data
to/from the control apparatus 200, and a display 620.
Moreover, the operation staff of the plant
100 can use the aforementioned apparatuses to input
setting parameters used in the numerical calculation
part 200, the learning part 260, and the pattern
generation part 400 which are provided in the control
apparatus 200.
The data processing apparatus 610 is formed
by an external input interface 611, a data send/receive
part 612, and an external output interface 613.
The input signal 61 of the data processing
apparatus generated in the external input apparatus 600
is inputted to the data processing apparatus 610 via
the external input interface 611. The data
send/receive part 612 of the data processing apparatus
610 acquires the I/O data information 60 provided in
the control apparatus 200 according to the information
in the data processing apparatus input signal 62.
Moreover, according to the information in the
data processing apparatus input signal 62, the data

send/receive part 612 outputs the I/O data information
60 containing the parameter setting value used in the
numerical calculation part 220, the learning part 260,
and the pattern generation part 400 provided in the
control apparatus 200.
The data send/receive part 610 transmits data
processing apparatus output signal 63 obtained by
processing the I/O data information 60, to the external
output interface 613. A data processing apparatus
output signal 64 is sent from the external output
interface 613 to the display 620 and displayed.
Accordingly, information stored in the respective
databases provided in the control apparatus 200 can be
displayed on the display 620.
It should be noted that the plant control
system 200 according to the embodiment of the present
invention includes the measurement signal database 210,
the numerical calculation database 240, the operation
signal database 250, the learning information database
270, the control logic database 290, and the pattern
database 500 which are arranged so as to constitute the
calculation apparatus of the control apparatus 200.
However, it is also possible to arrange all or some of
them outside the control apparatus 200.
Moreover, in the aforementioned example, the
numerical calculation part 220 is built in the control
apparatus 200. However, it is also possible to arrange
it outside the control apparatus 200.

For example, it is also possible to arrange
the numerical calculation part 220 and the numerical
calculation database 240 outside the control apparatus
200 and the numerical calculation result 6 is sent to
the control apparatus 2 00 via the Internet.
Next, the learning part 260 provided in the
control apparatus 200 will be detailed with reference
to Fig. 2..
As shown in Fig. 2, the learning part 2 60
includes a model input generation part 2 61, a pre-
learning part 262, a pattern selection part 263, a
pattern transforming part 264, and a re-learning part
2 65.
The model input generation part 2 61
constituting the learning part 260 applies an operation
amount 19 to the current model input condition of the
model 230 so as to generate the model input 7 and the
state input 20 after the operation, by the model input
generation part.
Moreover, the state input 20 is extracted by
using the model input generation part 2 61 from the
measurement signal 3 of the plant 100 stored in the
measurement signal database 210.
The pre-learning part 262 constituting the
learning part 260 uses the learning information data 9
stored in the learning information database 270 before
start of the operation of the plant 100, so as to learn
the method for generating the model input 7 so that the

model output 8 simulated/calculated by the model 230
can achieve the model output target value.
Here, the state input 20 is used as it is.
The model input 7 is generated by applying the
operation amount 19 learned by the pre-learning part
2 62, to the current model input condition, inputs it to
the model 230, and inputs it as the state input 20
after the operation to the pre-learning part 262.
The learning data 10 obtained as a result of
the learning in the pre-learning part 262 is stored in
the learning information database 270.
According to the learning information 16
stored in the learning information database 270, the
pattern generation part 400 generates pattern data and
stores the pattern data 17 generated by the pattern
generation part 400, in the pattern database 500.
The pattern selection part 263 constituting
the learning part 260 selects the pattern data 23 from
the pattern data 18 stored in the pattern database 500
according to the state input 20 and supplies it to the
pattern transforming part 264.
The pattern transforming part 264
constituting the learning part 250 patterns the pattern
data 23 selected by the pattern selection part 263 and
reduces the input dimension.
Moreover, the pattern transforming part 264
executes inverse operation on the operation amount 21
learned by the patterning so as to increase the input

dimension.
Referring to Fig. 3, explanation will be
given on the patterning. As shown in Fig. 3, the
pattern transforming part 264 processes the pattern
data as follows: the input information value to be used
in the patterning is set to 1 and the input information
value not to be used is set to 0.
Accordingly, when patterning the state input
20, only the input having the pattern data information
value 1 for the state input 20 is selected.
Moreover, in the inverse operation, linear
interpolation is performed between the inputs used in
the patterning for the operation amount 21 patterned
after the learning so as to obtain an operation amount
of other inputs.
In Fig. 2, the re-learning part 265
constituting the learning part 2 60 learns the
generation method of the model input 7 after the
operation of the plant is started,
by using the learning information data 9 stored in the
learning information database 270, so that the model
output 8 to be simulated/calculated by the model 230
corrected by the measurement signal 3 can achieve the
model output target value.
Here, the pattern transforming part 264
patterns the state input 20. The patterned state input
22 is supplied to the re-learning part 265
corresponding to the pattern. As a result of learning

in the re-learning part, a patterned operation amount
21 is outputted from the re-learning part 265.
The patterned operation amount is transformed
again into the operation amount 19 by the pattern
transforming part 2 64 and inputted to the model input
generation part 2 61 so as to generate the model input 7
and the state input 20 after the operation.
With the configuration thus far explained,
the learning part 260 realizes the pre-learning for the
pattern generation before starting the operation of the
plant and the re-learning by the patterning of the
state input after the operation start of the plant 100.
Fig. 4 is a flowchart showing the operation
of the control apparatus 200 for controlling a plant as
the embodiment of the present invention explained with
reference to Fig. 1 to Fig. 3.
In the flowchart of Fig. 4, a combination of
steps 1000, 1010, 1020, 1030, 1040, and 1050 is
executed. Hereinafter, explanation will be given on
each of the steps.
After start of the operation of the control
apparatus 200 for controlling the plant 100, firstly,
in step 1000, numerical calculation is executed by the
numerical calculation part 220 of the control apparatus
200 and the numerical calculation data 4 is sent to the
numerical calculation database 240 so as to be stored
in it.
Next, in step 1010, the learning part 260 and

the model 230 of the control apparatus 200 are used to
execute pre-learning of the plant operation method and
transmit the learning data 10 to the learning
information database 270 so as to be stored in it.
Next, in step 1020, the pattern generation
part 400 of the control apparatus generates pattern
data 17 by using the learning data 16 obtained by pre-
learning stored in the learning information database
270 and transmits the pattern data 17 to the pattern
database 500 so as to be stored in it. The
aforementioned operation is executed before starting
the operation of the plant 100.
After the operation of the plant 100 is
started, step 1030 inputs the measurement signal 1 of
the plant 100 to the control apparatus 200 by using the
external input interface 201 and transmits it to the
measurement signal database 210 so as to be stored in
it.
Step 1040 patterns the state input extracted
by the acquired measurement signal 3 by using the
pattern database stored in the pattern database 500 and
executes re-learning in the learning part 2 60.
After the learning in the learning part 260,
by using the learning result determination part 300 of
the control apparatus 200, appropriate learning data 12
is selected from the learning data 11 obtained by the
used pattern is selected and outputted from the
learning result determination part 300 to the operation

signal generation part 280.
In the subsequent step 1050, the operation
signal generation part 280 generates an operation
signal 14 by using the learning data 12 and the control
logic data 13 and outputs it as the operation signal 15
from the operation signal generation part 280 via the
external input interface 202 to the plant 100.
The aforementioned steps 1030 to 1050 are
repeatedly executed each time a measurement signal is
inputted, thereby controlling the plant 100.
Next, the operations of the steps 1010, 1020,
and 1040 in Fig. 4 will be detailed with reference to a
flowchart.
Fig. 5 is a flowchart showing the operation
of the pre-learning of step 1010.
As shown in Fig. 5, the flowchart of the pre-
learning operation is executed by a combination of
steps 2000, 2010, 2020, 2030, 2040, 2050, 2060, 2070,
and 2080.
Hereinafter, a detailed explanation will be
given on each of the steps.
Step 2000 initializes (sets I=1) the number
of initialization times I which indicates the number of
repetitions of the steps 2010 to 2060.
Next, step 2010 sets an initial value of the
state input. An arbitrary model input value can be
selected as the initial value of the state input.
Step 2020 initializes (sets J = 1) the number

of operation times J indicating the number of operation
times J indicating the number of repetitions of steps
2030 to 2050.
Next, step 2030 obtains an operation amount
of the model input by using the pre-learning part 2 62
of the learning part 260 constituting the control
apparatus 2 00.
Step 2040 adds the operation amount to the
state input by using the model input generation part of
the learning part 260 to operate the model 230 and
update the state input.
Step 2050 learns the operation method of the
model 230 by using the neural network and various
learning algorithms such as the reinforced learning
according to the model output obtained as a result of
the operation of the model 2230.
Step 2060 is a branching section where the
number of operation times J is compared to a
predetermined threshold value. If J is smaller than
the threshold value, J is incremented by 1 and control
is returned to step 2030. If J is greater than the
threshold value, control is passed to step 2070 which
is a branching section.
Step 2070 compares the number of
initialization times I to the predetermined threshold
value. If I is smaller than the threshold value, I is
incremented by 1 and control is returned to step 2010.
If I is greater than the threshold value, control is

passed to step 2080.
Step 2080 transmits the learning result
obtained by the learning part 2 60 to the learning
information database 270 so as to be stored in it.
Then, control is passed to step to terminate the pre-
learning operation.
By the aforementioned operation, the pre-
learning starts learning by using an arbitrary model
input as the initial value of the state input in step
2010. Accordingly, it is possible to acquire an
operation method to reach the model output target value
from an arbitrary model input condition.
Fig. 6 is a flowchart showing the pattern
generation of step 1020.
As shown in Fig. 6, the flowchart of the
pattern generation operation executes a combination of
steps 4000, 4010, 4020, and 4030.
Hereinafter, each of the steps will be
detailed with reference to Fig. 7 and Fig. 8,
Step 4000 initializes (sets I=1) the data
reference number I as the value indicating the number
of repetitions of the steps 4010 to 4020.
Next, step 4010 references the learning data
obtained by pre-learning stored in the learning
information database 270 according to I so as to
generate a pattern. It should be noted that a detailed
explanation will be given later on the pattern
generation algorithm.

Fig. 7 shows the learning data stored in the
learning information database 270 of the control
apparatus 2 00.
As shown in Fig. 7, the learning information
database 270 contains the state input value, the
operation amount value in its state input, and the
model output amount obtained as the result of the
operation.
A pattern is generated according to the
operation amount in the learning data. It should be
noted that in Fig. 7, S_0001 is a number assigned for
distinguishing the state input.
Step 4020 in Fig. 6 transmits the generated
pattern and the state input value of the referenced
learning data to the pattern database 500 so as to be
stored in it.
Fig. 8 shows the learning data stored in the
pattern database 500 of the control apparatus 200.
As shown in Fig. 8, the pattern database 500
contains the pattern data and the state input value
referenced upon generation of the pattern.
When storing the generated pattern, if the
generated pattern coincides with the pattern data
already stored, it is stored as the state input
corresponding to the pattern data.
Accordingly, there is a case that one pattern
data is correlated to a plurality of state inputs. It
should be noted that in Fig. 8, P 0001 is a number

assigned to distinguish pattern data.
Step 4030 in Fig. 6 is a branching section
where the data reference number I is compared to the
predetermined threshold value. I f J is smaller than
the threshold value, J is incremented by 1 and control
is returned to step 4010. On the contrary, if J is
equal to or greater than the threshold value, control
is passed to the step to terminate the operation of the
pattern generation.
Next, a detailed explanation will be given on
the pattern generation algorithm of step 4010 with
reference to the flowchart of Fig. 9 and Fig. 10.
As shown in Fig. 10, the pattern generation
algorithm starts search from the pattern input
dimension 1 in order to suppress the input dimension to
be patterned to a minimum value and repeats the search
to generate a pattern while increasing the input
dimension until the end condition is satisfied.
Moreover, in order to effectively perform the
pattern search, the evolution search algorithm is used
for the search. That is, the pattern database
expression is made to be a gene (solution candidate)
and the generated solution candidates are subjected to
genetic operations such as mixing and mutation so as to
search for an optimal pattern.
As shown in Fig. 9, the flowchart of step
4010 is formed by a combination of steps 4011, 4012,
4013, 4014, 4015, 4016, 4017, 4018, and 4019.

Hereinafter, each of the steps will be
detailed.
Step 4011 initializes J (J = 1) as a value
indicating the input dimension used for patterning.
Step 4012 initializes the number of
generations K (set K =1) as a value indicating the
number of repetitions of steps 4010 to 4015.
Next, step 4013 generates L solution
candidates under the constriction condition that J is
the input dimension which makes the pattern data
information value.
Step 4014 calculates a patterning error for
each of the solution candidates.
As shown in Fig. 10, the patterning error is
calculated as an operation amount obtained by linearly
interpolating the operation amount of the learning data
to be referenced and the input used in patterning.
Step 4015 performs genetic operations such as
crossover and mutation on the solution candidates under
the constriction condition according to the calculated
patterning error so as to generate a new solution
candidate.
Step 4016 is a branching section . If the
number of generations K is equal to or smaller than a
predetermined threshold value, K is incremented by 1
and control is returned to step 4010. Otherwise,
control is passed to step 4017.
Step 4017 selects a solution candidate which

makes the patterning error to be minimum for the
solution candidate group which has passed through the
predetermined number of generations and control is
passed to step 4018.
Step 4018 is a branching section. If the
patterning error of the solution candidate selected in
step 4017 is equal to or greater than a predetermined
threshold value, or the input dimension J is equal to
or smaller than a predetermined threshold value, J is
incremented by 1 and control is returned to step 4012.
Otherwise, control is passed to step 4019.
Step 4019 uses the selected solution
candidate as a generation pattern and transmits it to
the pattern database 500 so as to be stored in it.
Control is passed to the step to terminate the
operation of the pattern generation algorithm.
With the aforementioned operation, the
generated pattern data is stored in the pattern
database 500.
The operation staff of the plant 100 uses the
delta processing apparatus 610 to display these
information on the display 620.
Thus, the operation staff can know what kind
of state input can be currently used for patterning.
Moreover,, the operation staff can input threshold
values J and K as the setting parameters via the
display 620.
Fig. 11 is a flowchart showing the re-

learning operation of step 1040.
As shown in Fig. 11, the flowchart of the re-
learning operation is formed by a combination of steps
2100, 2110, 2120, 2130, 2140, 2150, 2160, 2170, and
2180.
Hereinafter, each of the steps will be
detailed.
Step 2100 extracts the state input to be
inputted to the re-learning part 2 65 from the
measurement signal by using the model input generation
part 261 of the learning part 260 constituting the
control apparatus 260.
Next, step 2110 uses the pattern selection
part 263 of the learning part 260 to calculate the
minimum state input error between the extracted state
input and the pattern data stored in the pattern
database 500 of the control apparatus 200.
As shown in Fig. 8, the pattern database 500
contains a plurality of state input values for each of
patterns and the minimum state input error is a minimum
difference between these values and the current state
input value.
The minimum state input error is the
similarity between the current state input and the
state input experienced in the pre-learning.
Accordingly, it is possible to select a pattern
corresponding to the state input having a high
similarity with the current state input value according

to the minimum state input error.
As a result, by using the selected pattern,
it is possible to learn the operation amount which can
obtain the control effect equivalent to the pre-
learning result for the current state input.
Step 2120 is a branching section to decide
whether any of the pattern minimum state input errors
calculated in step 2110 is equal to or smaller than the
predetermined threshold value. If yes, control is
passed to step 2130. Otherwise, control is passed to
step 2140.
Step 2130 selects all the patterns having the
minimum state input error equal to or smaller than the
threshold value and then control is passed to step
2150.
Step 2140 selects only the pattern having the
smallest minimum state input error and control is
passed to step 2150.
Next, step 2150 initializes (set 1=1) the I
as the value indicating the number of repetitions of
the steps 2160.
Step 2160 learns the model operation method
of the model 230 by using the re-learning part 265 of
the learning part 260 corresponding to the selected
pattern.
Step 2170 is a branching section. If I is
equal to or smaller than the selected number of
patterns, I is incremented by 1 and control is returned

to step 2160 and learning is performed by using the re-
learning part 2 65 corresponding to the another selected
pattern.
Step 2180 decides the appropriate operation
method among the re-learning results obtained by the
selected pattern by using the learning result
determination part 300 of the control apparatus 200 and
outputs it to the operation signal generation part 280
of the control apparatus 200. Then, control is passed
to the step to terminate the operation of re-learning.
With the aforementioned operation, re-
learning can learn the operation method equivalent to
the pre-learning result in a short time by using the
pattern selected according to the pre-learning result
for the current state input.
Moreover, it is possible to input the
threshold value of the minimum state input error as the
set parameter via the display 620.
Next, a detailed explanation will be given on
the re-learning algorithm of step 2160 with reference
to the flowchart of Fig. 12.
As shown in Fig. 12, the flowchart of the
operation of the re-learning algorithm is executed by a
combination of steps 2200, 2210, 2220, 2230, 2240,
2250, 2260, 2270, 2280, 2290, and 2300.
Hereinafter, each of the steps will be
detailed.
Step 2200 initializes (set J = 1) the number

of initializations J as a value indicating the number
of repetitions of steps 2210 to 2280.
Next, step 2210 initializes the state input
to an input value to be extracted from the measurement
signal 3.
Step 2220 initializes (K = 1) the number of
repetitions K as a value indicating the number of
repetitions of steps 2230 to 2270.
Next, step 2230 obtains the operation amount
of the patterned model input by using re-learning part
265 of the learning part 260 corresponding to the
pattern to be used.
Step 2240 linearly interpolates the operation
amount of the patterned model input by using the
pattern transforming part 264 of the learning part 260
so as to obtain the operation amount.
Step 2250 adds the operation amount to the
state input by using the model input generation part
261 of the learning part 260 so as to operate the model
and update the state input.
Step 2260 patterns the state input according
to the pattern data to be used, by using the pattern
transforming part 264 of the learning part 260.
Step 2270 learns the model operation method
by using the neural network and various learning
algorithms such as reinforced learning according to the
model output obtained as a result of the model
operation.

Step 2280 is a branching section where the
number of operations K is compared to a predetermined
threshold value. If K is smaller than the threshold
value, K is incremented by 1 and control is returned to
step 2230. Otherwise, control is passed to step 2290.
Step 2290 compares the number of
initializations J to a predetermined threshold value.
If J is smaller than the threshold value, J is
incremented by 1 and control is returned to step 2210.
Otherwise,, control is passed to step 2300.
Step 2300 transmits the learned result to the
learning information database 270 of the control
apparatus 200 so as to be stored in it and control is
passed to the step to terminate the operation of re-
learning.
With the aforementioned operation, the re-
learning uses the patterned state input to learn the
operation method when the current state input is used
as the initial value.
This completes the detailed explanation of
the operations of the steps 1010, 1020, and 1040 shown
in Fig. 4.
Next, explanation will be given on the screen
displayed on the display 620 for displaying the data
processing apparatus output signal 64 outputted from
the data processing apparatus 610 constituting the
plant control system 200 according to the embodiment of
the present invention.

Fig. 13 to Fig. 20 show examples of the
screen displayed on the display 620 attached to the
plant control system 200 shown in Fig. 1.
Fig. 13 shows an example of an initial
screen. By operating a mouse 602 to superimpose the
cursor on a button and clicking the mouse 602 while the
screen of Fig. 13 is displayed on the display 620, it
is possible to select (press) the button.
When the button 701 of "numerical
calculation", the button 702 of "database reference",
the button 703 of "patterning/learning", and the button
704 of "operation execution" are selected, the screens
of Fig. 14, Fig. 15, Fig. 16, and Fig. 17 are displayed
on the display 620, respectively. Moreover, when the
button 705 of "end" is selected, the initial screen is
terminated.
Fig. 14 shows the case when the button 711 of
"calculation condition setting" is selected so as to
input/set various calculation conditions required for
executing the calculation by the numerical calculation
execution part 220 of the control apparatus 200.
Moreover, when the button 712 of "numerical
calculation execution" is selected, it is possible to
start calculation by the numerical calculation
execution part 220. Moreover, when the button 713
"return" is selected, control is passed to Fig. 13.
On the screen of the display 620 shown in
Fig. 15, it is possible to select a database containing

information to be displayed.
By selecting the button 721 of "measurement
signal database", the button 722 of "operation signal
database", the button 723 of "numerical calculation
database", the button 724 of "learning information
database", the button 725 of "control logic database",
and the button 726 of "pattern database", it is
possible to access the measurement signal database 210,
the operation signal database 250, the numerical
calculation database 240, the learning database 270,
the control logic database 290, and the pattern
database 500, respectively.
It is possible to display the information in
the respective databases on the display 62 0 and add new
information to the information in the database and to
modify or delete information in the database.
Moreover, by selecting the button 727 of "return", the
screen is returned to Fig. 13.
On the screen of Fig. 16, when the button 731
of "pre-learning" is selected, the pre-learning screen
shown in Fig. 18 appears.
Moreover, when the button 732 of "re-
learning" is selected on the screen of Fig. 16, the re-
learning screen shown in Fig. 19 appears on the display
620. When the button 733 of "return" is selected, the
screen of Fig. 13 appears.
The screen of Fig. 17 shows the model input
741 before operation, the model output 742, the model

input 743 after the operation, model output 744, and
the learned operation amount 745 as a guidance display,
so that the plant operation staff can select whether to
execute an operation on the guidance display.
When an operation is to be executed, the
button 746 of "yes" is selected. Otherwise, the button
747 of "no" is selected.
On the display screen shown in Fig. 1, the
operation staff can confirm the relationship between
the pattern and the operation signal before operating
the plant, which increases the reliability of the plant
operation..
That is, the plant control method shown in
Fig. 17 can be detailed as follows. In the plant
control method of the control apparatus for controlling
the plant by using a model simulating the plant, as
shown in Fig. 8, the control apparatus has a pattern
database containing the state input pattern which
patterns a plurality of operation signals to be
inputted to the model. Moreover, as shown in Fig. 7,
the control apparatus includes a learning information
database containing a plurality of operation signals to
be inputted to the model and the output from the model.
As shown as the model input 743 and the model output
744 after the operation, the plurality of operation
signals stored in the learning information database and
the output from the model are outputted to the display.
As shown as the learned operation amount 745, the

operation signal to the plant according to the
plurality of operation signals inputted to the model
and the state input pattern stored in the pattern
database are superimposed and outputted to the display.
As shown as the button 74 6 to be selected when an
operation is to be executed, the plant is controlled by
the operation signal to the plant according to
permission of operation on the plant.
In Fig. 18, when the button 751 of "learning
start" is selected, the pre-learning part 262 of the
learning part 260 and the model 230 of the control
apparatus 200 are operated according to the flowchart
shown in Fig. 5 so as to execute the pre-learning of
the operation method.
Moreover, when the button 742 of "pattern
generation" is selected the pattern generation screen
shown in Fig. 20 appears on the display 620.
When the button 753 of "return" is selected,
control is returned to Fig. 16.
In Fig. 19, it is possible to input the set
value of the minimum state input error threshold value
to be used in the flowchart of Fig. 11, into the data
input column 761.
After the set value is inputted in the data
input column 7 61, if the button 7 62 "execution" is
selected, the learning part 260 of the control
apparatus 200 and the model 230 are operated according
to the flowchart shown in Fig. 11 so as to execute re-

learning.
Moreover, if the button 763 of "return" is
selected, control is returned to Fig. 16.
In Fig. 20, it is possible to input a set
value of the patterning error threshold value used in
the flowchart shown in Fig. 9 into the data input
column 771, and the set value of the pattern input
dimension threshold value into the data input column
772.
If the button 773 of "execution" is selected
after the setting values are inputted into the data
input columns 771 and 772, the pattern generation part
400 of the control apparatus 200 is operated according
to the flowchart shown in Fig. 6 so as to execute
pattern generation.
Moreover, if the button 774 of "return" is
selected, control is returned to Fig. 18.
This completes the explanation on the screens
displayed on the display 620.
As has been described above with reference to
the flowcharts in Fig. 11 and Fig. 12, in the
embodiment of the present invention, the state input
inputted to the learning part of the plant control
system is patterned with a smaller number of inputs
than the model input dimension so as to reduce the
model search space, thereby enabling a high-speed
learning.
Moreover, a desired learning result can

always be obtained by generating a pattern by using a
pre-learning result without reducing the input
dimension in advance before start of the plant
operation as has been detailed with reference to the
flowchart in Fig. 5, and comparing the similarity of
the current state input to the state input contained in
the pre-learning result after start of the plant
operation so that a pattern generated from the state
input having a higher similarity is selected for
learning the operation method as has been detailed with
reference to the flowchart of Fig. 11.
Moreover, as has been detailed with reference
to the flowchart of Fig. 9, it is possible to
automatically generate a pattern input dimension and
pattern information by an evolutionary search
algorithm., which enables pattern generation not
requiring any manual operation while preventing an
erroneous patterning.
Furthermore, the plant operation staff can
visually check the state of the patterning and the
effect of the patterned learning because the apparatus
has the function to reference the information stored in
the respective databases on the display and the
function to input setting parameters to be used in the
learning part and the pattern generation part via the
display as has been detailed with reference to Fig. 13
to Fig. 20.
Next, explanation will be given on a case

that the plant control system 200 as the embodiment of
the present invention shown in Fig. 1 to Fig. 20 is
applied to a thermal power plant 100 having a boiler.
It should be noted that the plant control system 200
may be used for control of other plants than the
thermal power plant.
Fig. 21 shows an outline of the thermal power
plant having a boiler. Firstly, explanation will be
given on the mechanism of the thermal power plant
having the boiler.
In the thermal power plant shown in Fig. 21,
coal as fuel is crushed into fine coal powder by a mill
110 and introduced via a burner 102 into a boiler 101
together v/ith a primary air for feed of the coal powder
and a secondary air for combustion adjustment. The
coal as the fuel burns in a furnace of the boiler 101.
The coal as the fuel and the primary air are
introduced via a pipe 134 while the secondary air is
introduced via a pipe 141 into the burner 102.
Moreover, the after-air for two-stage
combustion is introduced into the boiler 101 via an
after-air port 103 installed in the boiler 101. The
after-air is introduced into the after-air port 103
from a pipe 142.
The high-temperature combustion gas generated
by combustion of the fuel coal inside the furnace of
the boiler flows downward along a route shown by arrows
in the furnace of the boiler 101 and passes through a

heat exchanger 106. After the heat exchange, the
combustion exhaust gas is discharged from the boiler
101 and flows downward toward an air heater 104
installed outside the boiler 101.
The combustion exhaust gas which has passed
through the air heater 104 comes into an exhaust gas
treating device (not depicted). After harmful
substances are removed from the combustion exhaust gas,
the exhaust gas is discharged into the atmosphere
through a chimney.
Water circulating in the boiler 101 is
introduced from a condenser (not depicted) installed in
a turbine 108 into the boiler 101 via a water feed pump
105. The water is heated in the heat exchanger 106 by
the combustion gas flowing downward in the furnace of
the boiler 101 and becomes vapor of a high temperature.
It should be noted that in this embodiment,
only one heat exchanger is depicted. However, it is
also possible to arrange a plurality of heat
exchangers.
The vapor of a high temperature and a high
pressure generated in the heat exchanger 106 is
introduced via a turbine governor valve 107 to the
vapor turbine 108 so that energy of the vapor drives
the vapor turbine 108 to rotate a generator 109 coupled
to the vapor turbine 108 and generate electricity.
Next, explanation will be given on routes of
the primary air and the secondary air introduced into

the furnace of the boiler 101 from the burner 102
installed in the furnace of the boiler 101 and the
route of the after-air introduced into the furnace of
the boiler 101 from the after-air port 103 installed in
the furnace of the boiler 101.
The primary air is introduced from a fan 120
into a pipe 130 which is branched to a pipe 132 which
passes through the air heater 104 and a pipe 131 which
bypasses the pipe 132 and the air heater 104. The
primary air which has flown downward through the pipes
132 and 131 is again mixed in a pipe 133 and introduced
into the mill 110.
The air which passes through the air heater
104 is heated by the combustion exhaust gas discharged
from the furnace of the boiler 101.
The coal (fine coal powder) generated in the
mill 110 by using the primary air is fed through the
pipe 133 to the burner 102.
The secondary air and the after-air are
introduced from a fan 121 into a pipe 140 and flows
downward in the pipe 140 which passes through the air
heater 104, so as to be heated. After the heating, the
pipe 140 is branched into a pipe 141 for the secondary
air and a pipe 142 for the after-air, which are
respectively introduced to the burner 102 and the
after-air port 103 installed in the furnace of the
boiler 101.
The control apparatus 200 for the thermal

power plant 100 having the boiler according to the
present embodiment has the function to adjust the air
amount to be introduced from the burner 102 into the
boiler 101 and the air amount to be introduced from the
after-air port 103 into the boiler 101 so as to reduce
the NOx and CO concentrations in the exhaust gas from
the boiler.
The thermal power plant 100 includes various
measurement devices for detecting the operation state
of the thermal power plant. A measurement signal
obtained from the measurement devices is sent as a
measurement signal 1 to the control apparatus 200.
The measurement devices for detecting the
operation state of the thermal power plant 100 may be,
for example, a flowmeter 150, a thermometer 151, a
pressure indicator 152, a generated electricity
measuring device 153, and a concentration measuring
device 154 for measuring O2 concentration and/or CO
concentration as shown in Fig. 2.
The flowmeter 150 measures the flow rate of
water supplied from a water feed pump 105 to the boiler
101. Moreover, the thermometer 151 and the pressure
indicator 152 respectively measure the temperature and
the pressure of the vapor generated by the heat
exchange with the combustion gas flowing downward in
the boiler 101 in the heat exchanger 106 in the boiler
101 and supplied to the vapor turbine 108.
The generated electricity measuring device

153 measures the electricity generated by a generator
109 rotated by the vapor turbine 108 which is driven by
the vapor generated by the heat exchanger 106.
Moreover, information on the concentrations
of the components (such as CO and NOx) contained in the
combustion gas flowing downward in the boiler 101 is
measured by the concentration measuring device 154
arranged at the exit of the boiler at the downstream
side of the boiler 101 so as to measure the O2
concentration and/or the CO concentration.
It should be noted that a plenty of other
measuring devices than those shown in Fig. 2 are
arranged on the thermal power plant 100 but they are
not depicted.
Fig. 22 is a partial enlarged diagram showing
the air heater 104 arranged at the downstream side of
the boiler 101 constituting the thermal power plant 100
and the pipes arranged in the air heater 104.
As shown in Fig. 22, air dampers 162, 163,
161, 160 are respectively arranged in a pipe 131 which
bypasses the secondary air pipe 132, the after-air pipe
142, the pipe 132 arranged inside the air heater 104,
and the air heater 104 which are branched at the
downstream side of the pipe 140 arranged inside the air
heater 104.
By operating the air dampers 160 to 163, it
is possible to change the area where the air passes
inside the pipes 131, 132, 141, 142 so as to separately

adjust the air flow rate in the pipes 132, 132, 141,
and 142.
By using the operation signal 15 outputted to
the thermal power plant 100 by the control apparatus
200 which controls the thermal power plant 100, it is
possible to operate the water feed pump 105, the mill
110, the air dampers 160, 161, 162, 163, and other
devices.
It should be noted that in the control
apparatus 200 for the thermal power plant 100 according
to the embodiment, devices for adjusting the state
amount of the thermal power plant such as the water
feed pump 105, the mill 110, the air dampers 160, 161,
162, 163 are called operation ends and instruction
signals required for operating them are called
operation signals.
Fig. 23 is a flowchart showing the process of
the thermal power plant control method in the control
apparatus 200 of the thermal power plant 100 having the
boiler in the embodiment of the present invention shown
in Fig. 21.
In Fig. 23, the control apparatus 200 for the
thermal plant 100 having the boiler in the embodiment
executes control of the thermal power plant 100 by
combining the steps 2181 and 2182.
Step 2181 references the learning data
obtained by each of the patterns selected by the
pattern selection part 263 constituting the learning

part 260 of the control apparatus 200 which controls
the thermal power plant 100, compares the total of
improvement values of the CO and NOx as the model
output amount after the operation for current state
input, and selects the maximum total of the improvement
values.
Moreover, when the number of patterns
selected by the pattern selection part 263 is 1,
control is passed to step 2182 without performing the
aforementioned process.
Step 2182 transmits the learning data 12 of
the pattern selected by the pattern selection part 263
to the operation signal generation part 280 and passes
control to step to terminate the operation of the
learning result determination part 300 constituting the
control apparatus 200.
With the aforementioned operation, the
control apparatus 200 uses the learning data to obtain
the best improvement values of CO and COx as the result
of re-learning so as to generate an operation signal.
Fig. 24 shows an example of the screen
displayed on the display 620 in the control apparatus
2 00 of the thermal power plant 100 having the boiler
according to the embodiment of the present invention.
Fig. 24 shows an example of screen which
shows the air amount of the burner 102 of the boiler
101 used as the model input 7 inputted to the model 230
constituting the control apparatus 200, the air amount

of the air port 103, guidance display of the
concentrations of CO, NOx used in the model output 8 by
simulation-calculating the thermal power plant to be
controlled by the model 230 before and after the
operation, and the operation amounts of the burner 102,
and the air port 103 learned by the learning part 265
by the patterning of the pattern transforming part 264.
The model input 7 displays distribution of the air
amount of the burner 102/air port 103 arranged before
and after a tank of the boiler 101 while correlating
them with the boiler structure diagram.
It should be noted that in Fig. 25, the
burner 102 and the air port 103 are arranged before and
after the tank of the boiler 101 respectively as one
stage each having five air entrances. However, the
numbers can be modified to an arbitrary value.
The operation staff of the plant can check
the control effect obtained by the patterned operation
amount and the learned operation method to decide
presence/absence of the operation while watching the
screen shown in Fig. 24.
Moreover, as shown in Figs. 25A to 25C,
patterning in the control apparatus 200 for the thermal
power plant 100 having the boiler according to the
embodiment of the present invention may be realized,
for example, as (1) grouping of the operation ends for
each of the burner 102 and the air port 103 of the
boiler 101 (Fig. 25A) , (2) grouping of the operation

ends before/after the tank of the boiler 101 (Fig.
25B), and (3) grouping of all the operation ends of the
boiler 101 (Fig. 25C).
As is clear from the aforementioned, the
operation staff of the thermal power plant 100 having
the boiler can control the thermal power plant by
selecting the appropriate patterning method by
considering the type of the operation end, the number
of operation ends, and the combustion characteristic of
the boiler of the thermal power plant to be controlled.
According to the embodiment of the present
invention,, it is possible to provide a plant control
system, which corrects the model according to the
measurement signal which has measured the operation
state of the plant, executes the high-speed learning so
as to perform re-learning by using the corrected model
and correct the plant control algorithm, thereby
accurately control the plant.
The present invention can be applied to a
plant control system and in particular, to a control
apparatus for a thermal power plant.
It should be further understood by those
skilled in the art that although the foregoing
description has been made on embodiments of the
invention, the invention is not limited thereto and
various changes and modifications may be made without
departing from the spirit of the invention and the
scope of the appended claims.

CLAIMS:
1. A plant control system for controlling a
plant including an operation signal generation part
which calculates an control instruction operation
signal for a plant to be controlled by using a
measurement signal which has measured an operation
state of the plant, the apparatus comprising:
an operation signal database which stores the
measurement signal of the measured plant;
an operation signal database which stores an
operation signal for the plant;
a numerical calculation execution part which
calculates the operation characteristic of the plant;
a numerical calculation database which stores
the numerical calculation result from the numerical
calculation execution part;
a simulation model which simulates control
characteristic of a plant when the operation signal is
supplied to the plant according to information on the
numerical calculation result from the numerical
calculation database;
a learning part which learns the plant
operation method by using the model;
a learning information database which stores
learning information data on the learning part;
a control logic database which contains
information used for the operation signal outputted
from the operation signal generation part;

a pattern generation part which generates
pattern data expressing a state input based on the
learning information data in the learning part with a
smaller input number than the model input dimension;
a pattern database which stores the pattern
data generated in the pattern generation part; and
a learning result determination part which
selects a learning result having a preferable control
effect from the learning result obtained by using a
plurality of patterns;
wherein the operation signal is calculated
from the operation signal generation part according to
the learning result selected by the learning result
determination part.
2. The plant control system as claimed in claim
1, wherein the learning part includes:
a model input generation part which generates
an input to be applied to the model and extracts a
model input from a plant measurement signal;
a pre-learning part which uses the learning
information data stored in the learning information
database before start of the plant operation so as to
learn the model input generation part so that the model
output simulated/calculated by the model achieves its
target value;
a pattern database which contains a pattern
generated by using the pattern generation part for the
learning information data obtained by the pre-learning

by the pre-learning part;
a pattern selection part which selects a
pattern similar to the measurement signal from the
pattern data stored in the pattern database after start
of the plant operation;
a pattern transforming part which patterns
the state input to be inputted to the learning part by
using the selected pattern, linearly interpolates the
inter-inputs for the learned and patterned operation
amount; and
a re-learning part which performs learning
with a reduced input dimension of the model input by
using the selected pattern so that the model output
simulated/calculated in the model by using the learning
information data stored in the learning information
database after start of the plant operation achieves
its target value.
3. The plant control system as claimed in claim
1, the pattern generation part constituting the control
apparatus has:
a function of performing repeated search
while successively incrementing the input dimension by
1 starting from the input dimension 1 until the end
condition is satisfied in order to generate pattern
data for a model operation amount contained in the
learning information data learned by using the pre-
learning part before start of the plant operation; and
a function of generating a plurality of

pattern data candidates as the pattern data search
part, performing a genetic operation such as mixing and
mutation on them, and repeating the search so as to
search for the pattern data.
4. The plant control system as claimed in claim
2, wherein the learning part constituting the control
apparatus compares the current state input to the state
input contained in the pre-learning result in the
pattern selection part after the plant operation start
and selects the pattern generated from the state input
having the higher similarity so as to learn the
operation method.
5. The plant control system as claimed in claim
1, wherein the control apparatus includes a data
processing device and a display which displays
information stored in a measurement signal database, an
operation signal database, a control logic database, a
learning information database, a numerical calculation
database, and a pattern database.
6. The plant control system as claimed in claim
5, wherein setting parameters used in the learning part
and the pattern generation part constituting the
control apparatus are inputted via an input part
connected to the data processing apparatus.
7. The plant control system as claimed in claim
1, wherein the plant to be controlled is a thermal
power plant including a boiler configured as follows:
operation ends to which the plant operation signal is

outputted are a burner which generates a high-
temperature gas by a combustion reaction using fossil
fuel and air and an air port which supplies air to the
combustion gas from the rear side of the flow of the
combustion gas generated by mixing the fossil fuel and
the air supplied from the burner; the measurement
signal of the measured plant is a measurement signal of
the concentration of carbon monoxide and the
concentration of nitrogen oxide obtained by measuring
the carbon monoxide concentration and nitrogen oxide
concentration of the fuel gas burned in the boiler by
using a measurement device; the learning result
determination part constituting the control apparatus
selects such learning information data by a selected
pattern that the total of the improvement values
indicating the differences between the concentrations
of the carbon monoxide and the concentrations of the
nitrogen oxide in the combustion gas before and after
the operation is maximum so as to generate the
operation signal.
8. The plant control system as claimed in claim
7, wherein the state input parameter to be patterned in
the learning result judging part constituting the
control apparatus is the air flow rate introduced into
the boiler separately from the burner and the air port
arranged in the boiler.
9. The plant control system as claimed in claim
7, wherein the learning result judging part

constituting the control apparatus uses one of the
following methods: grouping of state inputs to be
patterned; grouping for the burner and the air port,
respectively; grouping of the entire boiler.
10. A plant control method for a control
apparatus which controls a plant by using a model
simulating the plant, the method comprising steps of:
providing a pattern base containing a state
input pattern in which a plurality of operations
signals to be inputted to the model by the control
apparatus are patterned;
providing a learning information database
containing a plurality of operation signals to be
inputted to the model and an output from the model;
outputting the plurality of operation signals
stored in the learning information database and the
output from the model to a display;
outputting to the display, an operation
signal to the plant based on the plurality of operation
signals inputted to the model and the state input
pattern stored in the pattern data base while
superimposing them on each other; and
controlling the plant by the operation signal
to the plant according to permission of the operation
execution on the plant.

A plant control system includes: a numerical
calculation execution part which calculates the
operation characteristic of the plant; a model for
simulating the plant control characteristic according
to information on the numerical calculation result; a
learning part which learns the plant operation method
by using the model; a learning information database
which stores learning information data on the learning
part; a pattern generation part which generates pattern
data expressing a state input based on the learning
information data in the learning part with a smaller
input number than the model input dimension; a pattern
database which stores the pattern data generated in the
pattern generation part; and a learning result
determination part which selects a learning result
having a preferable control effect from the learning
result obtained by using a plurality of patterns.

Documents:

http://ipindiaonline.gov.in/patentsearch/GrantedSearch/viewdoc.aspx?id=c8U+l489nhI6f/ApbNuyIw==&loc=wDBSZCsAt7zoiVrqcFJsRw==

« Previous Patent

Next Patent »

Patent Number

272430

Indian Patent Application Number

1197/KOL/2008

PG Journal Number

14/2016

Publication Date

01-Apr-2016

Grant Date

31-Mar-2016

Date of Filing

11-Jul-2008

Name of Patentee

HITACHI, LTD.

Applicant Address

6-6, MARUNOUCHI 1-CHOME, CHIYODA-KU, TOKYO

Inventors:

#	Inventor's Name	Inventor's Address
1	EGUCHI TORU	C/O HITACHI, LTD., INTELLECTUAL PROPERTY GROUP, 12TH FLOOR, MARUNOUCHI CENTER BUILDING, 6-1, MARUNOUCHI 1-CHOME, CHIYODA-KU, TOKYO 100-8220
2	YAMADA AKIHIRO	C/O HITACHI, LTD., INTELLECTUAL PROPERTY GROUP, 12TH FLOOR, MARUNOUCHI CENTER BUILDING, 6-1, MARUNOUCHI 1-CHOME, CHIYODA-KU, TOKYO 100-8220
3	FUKAI MASAYUKI	C/O HITACHI, LTD., INTELLECTUAL PROPERTY GROUP, 12TH FLOOR, MARUNOUCHI CENTER BUILDING, 6-1, MARUNOUCHI 1-CHOME, CHIYODA-KU, TOKYO 100-8220
4	SHIMIZU SATORU	C/O HITACHI, LTD., INTELLECTUAL PROPERTY GROUP, 12TH FLOOR, MARUNOUCHI CENTER BUILDING, 6-1, MARUNOUCHI 1-CHOME, CHIYODA-KU, TOKYO 100-8220
5	SEKIAI TAKAAKI	C/O HITACHI, LTD., INTELLECTUAL PROPERTY GROUP, 12TH FLOOR, MARUNOUCHI CENTER BUILDING, 6-1, MARUNOUCHI 1-CHOME, CHIYODA-KU, TOKYO 100-8220

PCT International Classification Number

G05B13/02; F23N5/18; G05B19/42

PCT International Application Number

N/A

PCT International Filing date

PCT Conventions:

#	PCT Application Number	Date of Convention	Priority Country
1			NA