Title of Invention

METHOD AND DEVICE FOR PRECALCULATING INITIALLY UNKNOWN PARAMETERS OF AN INDUSTRIAL PROCESS

Abstract

This invention relates to a method for pre-calculating initially unknown parameters of an industrial process, in particular a plant in the basic materials industry, having parameters that vary, in particular abruptly, the initially unknown parameters to be predicted being determined by means of a process modal as a function of initially known parameters of the process, and the process model having at least a global process model, which constitutes a time-averaged image of the process, and at least one specialized process model, which constitutes an image of the process for a specific operating state or working point that is determined by means of the initially known parameters.

Full Text

-1A- Description The invention relates to a method and a device for precalculating initially unknown parameters of an industrial process, in particular a plant in the basic materials industry. In the closed-loop or open-loop control of industrial processes, in particular in plants in the basic materials industry, such as steel works, for example, it is often necessary to ascertain specific process parameters or states in the manner of a forecast, since they are not available at the point in time at which they are needed for the closed-loop or open-loop control. It is known to identify process parameters with the aid of a model. Here, input variables or the input variables relevant for the process parameters to be identified are fed to a process model, which is generally simplified. However, in the case of plants in the basic materials industry, this known method often leads to problems. It is a characteristic of plants in the basic materials industry, in particular of steel works, that errors in the identification or a lack of accuracy in the identification lead to high costs as a result of the production of scrap. This is promoted in particular by the fact that in plants in the basic materials industry, in particular in steel works, the operating states alter abruptly, so that during the time which the process model needs to adapt to the new input variables, the production of products of inadequate quality may occur. This problem relates in particular to rolling trains, in which the operating state changes abruptly as a result of rolling, for example, a new rolled strip that consists of a new material or which has a thickness that differs from the preceding strip. During rolling, in particular, important closed-loop control or open-loop control variables, such - 2 - as the rolling force, for example, are available too late as a measured variable for the corresponding open-loop or closed-loop control. It is the object of the invention to specify a method and a device which makes it possible to ascertain initially unknown parameters of an industrial process in the manner of a forecast and particularly precisely. In this case it is desirable if this ascertainment of the parameters is adaptable quickly to changing operating states or working points of the corresponding process. Here, parameters can be understood to be state variables or process parameters. According to the invention, the object is achieved by means of a method and a device for precalculating initially unknown parameters of an industrial process, in particular a plant in the basic materials industry, having parameters that vary, in particular abruptly, the initially unknown parameters to be predicted being determined by means of a process model as a function of initially known parameters of the process, and the process model having at least a global process model, which constitutes a time-averaged image of the process, and at least one specialized process model, which constitutes an image of the process for a specific operating state or working point that is determined by means of the initially known parameters. In this case, a model for predicting the initially unknown parameters for a specific point in time, that is to say either the global process model or a specialized process model, is advantageously selected as a function of the initially known parameters or of some of the initially known parameters. By means of this changing over in specific operating states to a model that is specialized for the corresponding operating state, a situation is reached in which the initially unknown parameters can be predicted in a particularly precise way. In addition, this procedure has been tried and tested in conjunction with online adaptation of the process model to the current process events. If a global process model, which repre- - 3 - sents the corresponding process in a time average, is used for the prediction of initially unknown process parameters for operating states that have not found entry into the modelling via the global process model, then under certain circumstances deficient prediction of the initially unknown parameters occurs for operating states of this type. If the corresponding global process model is adapted to these new operating states, then this model will possibly supply a deficient prediction of the initially unknown parameters for subsequent operating states. For a case of this type, a specialized process model is advantageously used for predicting the initially unknown parameters, the said process model being specially adapted for the corresponding operating state or working point. In a further advantageous refinement of the invention, specialized process models for specific operating states or working points are stored in the form of a library or database. In a further advantageous refinement of the invention there is a representative data set of parameters of the process, with which the global process model is adapted to the corresponding process at specific intervals. If, for example, the global process model is a neural network, then the latter is trained using the representative data set. Should operating states or working points which are not represented in this data set occur, then the appropriate parameters are appended to the data set. If the global process model is then adapted or trained using this new data set, then it also represents the new operating states or working points. In contrast to pure post-training using only new parameters, this training does not lead to a destruction of the knowledge, represented in the global process model, about operating states or working points that have already occurred. Further advantages and inventive details emerge from the following description of exemplary embodiments using the /drawing ami in conjunction with the subclaims. - 4 - In detail: FIG 1 shows the method according to the invention for predicting initially unknown parameters/ FIG 2 shows the interaction of prediction and adaptation. FIG 1 shows the method according to the invention for precalculating initially unknown parameters. In this case, initially known parameters are fed at least to a classifier 4 and, in particular optionally, to a global process model 5 and at least one specialized process model 6. The global process model 5 and the specialized process models 6 calculate the initially unknown parameters 3 as a function of initially known parameters 1 of the process. In this case, the prediction of the initially unknown parameters by the global process model 5 and the process models 6 that are specialized for specific operating states is carried out according to the operating state of the process. Whereas the specialized process models 6 are specialized for specific operating states of the industrial plant to be modelled, that is to say the process, the global process model 5 is a model that is representative of the average process. Different operating states may be, for example, different steel grades in a rolling mill. The selection of which of the models is used for the current point in time for predicting the initially unknown parameters 3 is made by the classifier 4, which selects the most suitable model. This is carried out, for example, by ascertaining a control variable 2 which indicates which model is the most suitable at the current point in time for identifying the process parameters 3. FIG 2 shows the interaction of prediction 15 and adaptation 16 in the method according to the invention. For a specific open-loop or closed-loop control application, parameters 12 which are initially unknown are necessary. However, these initially unknown parameters 12 depend on—i'nltially known parameters 10 about the process. -Stich initially known parameters 10 may be parameters in the strict sense, such as model parameters. - 5 - state variables, manipulated variables or presettings. On the basis of the initially known parameters 10, initially unknown parameters 14 are predicted, by means of a model 11, through the prediction 15. By means of these predicted, initially unknown parameters 14, the desired open-loop or closed-loop control task for the corresponding process is carried out. However, the initially unknown parameters for the previous point in time can be ascertained at a later point in time. As a function of the predicted, initially unknown parameters 14, and of the initially unknown parameters 13 measured at the previous point in time, and as a function of the initially known process parameters 10, the adaptation 16 ascertains a new model 11. In this case, a model may be a global process model or a specialized process model. For specialized process models, the adaptation is advantageously carried out immediately after the measured, initially unknown parameters 13 are available. 6. WE CLAIM: 1. Method for pre-calculating initially unkown parameters 2. Method for pre-calculating initially unknown parameters as claimed in claim 1 wherein a model for predicting the initially unknown parameters (3) for a specific point in time, that is to say either the global process model (3) or a specialized process model (6), is selected as a function of the initially known parameters (1) or of some of the initially known parameters (1)- 3. Method for pre—calculating initially unknown parameters as claimed in claim 1 or 2, wherein the global process model (5) or the specialized process model is an analytical model, a neural network or a hybrid model, that is to say a model having an analytical model and a neural network. 1 7. 4. Method for pre-calculating initially unknown parameters as claimed in claim 1, 2 or 3 wherein the specialized process model (6) is a variable. 3. Method for pre-calculating initially unknown parameters as claimed in claim 3, wherein the specialized process model (6) is A simple adaline, that is to say a neural network having one neuron. 6. Method for pre—calculating initially unknown parameters as claimed in claim 1,2,3,4 or 5 wherein the global process model (5) and the specialized process model (6) are adapted to the current process events, in particular by means of on—line training. 7. Method for pre-calculating initially unknown parameters as claimed in claim 6, wherein the adaptation (16) is performed as a function of the initially known parameters (10) and as a function of the initially unknown parameters (12). 8. Method for pre-calculating initially unknown parameters as claimed in claim 7, wherein the initially unknown parameters (17), which had been current at a specific applicable point in time in relation to the process sequence, are determined or measured at a later point in time, and in that the adaptation (16) is performed as a function of the parameters (13) at the 2 8. applicable point in time that were initially unknown and were determined at a later point in time, and of the parameters (14) that were initially unknown and were predicted far this applicable point in time. 9. Method for pre-calculating initially unknown parameters as claimed in one of claims 1 to 8, wherein the parameters predicted by means of the global process model or the specialized process model are fed to information processing based on neural networks, which improves the said parameters in the sense of fine adaptation, the information processing based on neural networks being adapted on-line to the current process events. 10. Method for pre-calculating initially unknown parameters as claimed in one of claims 1 to 9, wherein the parameters predicted by means of the global process model or the specialized process model are combined with a correction term, in particular additively or multiplicatively, the correction term being farmed by means of information processing based on neural networks, and the information processing based on neural networks being adapted on—1ine to the process. 9. 11- Device for carrying out the method as claimed in one of the preceding claims for pre-calculating initially unknown parameters of an industrical process, in particular a plant in the basic materials industry, having parameters that vary, in particular abruptly, the initially unknown parameters to be predicted being d«t«rmined by means of a process model as a function of initially known parameters of the process, and the process model having at least a global process model, which constitutes a time-averaged image of the process, and at least one specialized process model, which constitutes an image of the process for a specific operating state or working point that is determined by means of the initially known parameters. This invention relates to a method for pre-calculating initially unknown parameters of an industrial process, in particular a plant in the basic materials industry, having parameters that vary, in particular abruptly, the initially unknown parameters to be predicted being determined by means of a process modal as a function of initially known parameters of the process, and the process model having at least a global process model, which constitutes a time-averaged image of the process, and at least one specialized process model, which constitutes an image of the process for a specific operating state or working point that is determined by means of the initially known parameters.

Documents:

01780-cal-1997 abstract.pdf

01780-cal-1997 claims.pdf

01780-cal-1997 correspondence.pdf

01780-cal-1997 description(complete).pdf

01780-cal-1997 drawings.pdf

01780-cal-1997 form-1.pdf

01780-cal-1997 form-2.pdf

01780-cal-1997 form-3.pdf

01780-cal-1997 form-5.pdf

01780-cal-1997 gpa.pdf

01780-cal-1997 priority document.pdf

1780-CAL-1997-(04-10-2012)-FORM-27.pdf

1780-CAL-1997-CORRESPONDENCE 1.1.pdf

1780-CAL-1997-CORRESPONDENCE.pdf

1780-CAL-1997-FORM-27.pdf

1780-cal-1997-granted-abstract.pdf

1780-cal-1997-granted-claims.pdf

1780-cal-1997-granted-correspondence.pdf

1780-cal-1997-granted-description (complete).pdf

1780-cal-1997-granted-drawings.pdf

1780-cal-1997-granted-examination report.pdf

1780-cal-1997-granted-form 1.pdf

1780-cal-1997-granted-form 2.pdf

1780-cal-1997-granted-form 3.pdf

1780-cal-1997-granted-form 5.pdf

1780-cal-1997-granted-gpa.pdf

1780-cal-1997-granted-letter patent.pdf

1780-cal-1997-granted-priority document.pdf

1780-cal-1997-granted-reply to examination report.pdf

1780-cal-1997-granted-specification.pdf

1780-cal-1997-granted-translated copy of priority document.pdf

1780-CAL-1997-PA 1.1.pdf

1780-CAL-1997-PA.pdf