Disclosure of Invention
The invention provides an industrial process model parameter optimization method based on generalized predictive control, which is used for solving the technical problems of low prediction model precision and poor adaptability in the prior art.
The method comprises the following steps:
Basic information and targets of a control system are acquired, working condition situations are defined according to the basic information of the control system, a working condition change set is generated, and the working condition change set comprises a plurality of working condition change combinations;
Determining a prediction model to-be-replaced list corresponding to each working condition change combination based on a target of the control system, and analyzing model information of each prediction model in the prediction model to-be-replaced list;
when the working condition changes, model information corresponding to the working condition change combination is obtained, and a reference track is determined through the model information;
Based on basic information and target definition control quality index of the control system, controlling the control system in a buffering stage according to the reference track, and monitoring the control quality index in the buffering stage;
and optimizing the prediction model parameters of the control system by controlling the quality index.
In some embodiments of the present application, defining a working condition situation according to basic information of a control system, generating a working condition change set includes:
Acquiring historical data of a plurality of complete control processes of a control system, establishing a data time axis, and intercepting the data time axis between a starting node and an ending node of an original working condition to obtain a first data time axis;
The basic information of the control system comprises internal disturbance, external disturbance and performance parameters, and the overall disturbance quantity is determined according to the internal disturbance and the external disturbance;
determining a forward delay length and a backward delay length through the integral disturbance quantity, forward delaying an original working condition starting node of the first data time shaft through the forward delay length, and backward delaying an original working condition ending node of the first data time shaft through the backward delay length;
The extended first data time axis is recorded as a second data time axis, and various performance parameter change curves are established to obtain a performance curve;
screening steady-state parameters in the performance parameters, and determining that the original working condition is a steady-state working condition or an unsteady-state working condition according to characteristic values of the steady-state parameters;
and defining working condition situations based on the steady state condition and the performance curve of the original working condition, and combining each working condition situation according to a logic relation so as to obtain a plurality of working condition change combinations.
In some embodiments of the present application, defining the operating condition context based on the steady state condition and the performance curve of the original operating condition includes:
performing time scale alignment treatment on performance curves of different steady-state working conditions, comparing the same kind of performance curves of different steady-state working conditions, and determining the difference of the performance curves;
Taking the performance parameter with the difference of the performance curves exceeding the corresponding first threshold value as a steady-state performance parameter, integrating the steady-state performance parameter, and splitting the steady-state working condition to obtain a plurality of working condition situations;
Performing time scale alignment treatment on performance curves of different unsteady state working conditions, comparing the same kind of performance curves of different unsteady state working conditions, and determining the difference of the performance curves;
and taking the performance parameter with the difference of the performance curves exceeding the corresponding second threshold value as an unsteady performance parameter, integrating the unsteady performance parameter, and splitting the unsteady working condition to obtain a plurality of working condition situations.
In some embodiments of the present application, determining a list to be replaced of a prediction model corresponding to each working condition change combination based on a target of a control system includes:
the targets of the control system comprise stability, accuracy and rapidity, and the stability index target value, the accuracy index target value and the rapidity index target value are obtained by quantifying the stability index target value, the accuracy index target value and the rapidity index target value;
Analyzing the present value and the change trend of the stability index, the present value and the change trend of the accuracy index, and the present value and the change trend of the rapidness index of each working condition change combination;
determining the model ranking according to the difference between the present value of the stability index and the target value, the difference between the present value of the accuracy index and the target value and the difference between the present value of the rapidity index and the target value;
Based on the stability index present value and the change trend of the working condition change combination, the accuracy index present value and the change trend, the rapidness index present value and the matching degree of the prediction model in the change trend evaluation model library;
And sequencing the prediction models according to the matching degree, and adding the prediction models conforming to the model ranking into a list of the prediction models to be replaced.
In some embodiments of the present application, determining a reference trajectory from model information includes:
Determining the performance of each prediction model in the prediction model to-be-replaced list according to the model information;
and integrating a plurality of prediction models through the distribution weight of the performance determination prediction model to generate a reference track.
In some embodiments of the present application, defining a control quality index based on basic information and a target of a control system includes:
Determining a control quality index according to the performance parameter and a target of the control system; ; wherein P is a control quality index,/> Is the target value of stability index,/>Stability index real-time value,/>Is the target value of accuracy index,/>For the real-time value of accuracy index,/>Is a rapidity index target value,/>For the real-time value of the rapidity index, exp is an exponential function, n is the performance parameter category,/>For the weight corresponding to the ith performance parameter,/>For the i-th performance parameter, k is a preset constant,/> Is a control quality function with respect to the differences between the three target values and the real-time value.
In some embodiments of the present application, optimizing the predictive model parameters of the control system by controlling the quality index includes:
presetting a control quality index standard curve corresponding to a reference track;
Establishing a control quality index change curve through the control quality index, and comparing the control quality index standard curve with the control quality index change curve to obtain the similarity;
If the similarity exceeds the corresponding threshold, optimizing the prediction model parameters of the control system by analyzing the reference track;
Otherwise, optimizing the prediction model parameters of the control system according to the difference between the similarity and the corresponding threshold value and the basic information of the control system.
In some embodiments of the present application, optimizing prediction model parameters of a control system according to differences between similarities and corresponding thresholds and basic information of the control system includes:
The prediction model parameters comprise a prediction time domain, a control time domain and a softening coefficient;
The causal indexes of the prediction time domain, the control time domain and the softening coefficient are calculated respectively through basic information, and the causal indexes meeting the requirements are screened out;
Respectively constructing influence functions of a prediction time domain, a control time domain and a softening coefficient according to the respective corresponding causal indexes;
integrating influence functions of a prediction time domain, a control time domain and a softening coefficient to obtain a comprehensive influence function;
and adjusting the comprehensive influence function based on the difference between the similarity and the corresponding threshold value, thereby determining the set values corresponding to the prediction time domain, the control time domain and the softening coefficient, and optimizing the prediction model parameters of the control system.
By applying the technical scheme, basic information and targets of the control system are acquired, working condition situations are defined according to the basic information of the control system, a working condition change set is generated, and the working condition change set comprises a plurality of working condition change combinations; determining a prediction model to-be-replaced list corresponding to each working condition change combination based on a target of the control system, and analyzing model information of each prediction model in the prediction model to-be-replaced list; when the working condition changes, model information corresponding to the working condition change combination is obtained, and a reference track is determined through the model information; based on basic information and target definition control quality index of the control system, controlling the control system in a buffering stage according to the reference track, and monitoring the control quality index in the buffering stage; and optimizing the prediction model parameters of the control system by controlling the quality index. According to the application, the working condition situation is defined according to the basic information of the control system, the working condition change set is generated, all conditions of working condition change are defined, and the subsequent analysis of the working condition change is convenient. By defining and monitoring the control quality index, the prediction model parameters of the control system are optimized, and the accuracy and adaptability of the model are improved.
Detailed Description
The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
The embodiment of the application provides an industrial process model parameter optimization method based on generalized predictive control, which is shown in fig. 1 and comprises the following steps:
Step S101, basic information and targets of a control system are acquired, working condition situations are defined according to the basic information of the control system, a working condition change set is generated, and the working condition change set comprises a plurality of working condition change combinations.
In this embodiment, the basic information of the control system includes a system architecture, an input-output relationship, a constraint condition, an internal disturbance, an external disturbance, a performance parameter, and the like, and the objective of the control system is the final objective requirement of control. The working condition situation is defined through basic information, and is the original working condition situation which is more subdivided. The working condition change combination is a working condition situation change condition, such as a starting working condition, a single-parameter steady-state working condition, a multi-parameter steady-state working condition, wherein the single-parameter steady-state working condition is that only one or a plurality of main variables in the system are stable, and the multi-parameter steady-state working condition is that a plurality of variables are stable at the same time. The method comprises two or more working condition situation changes.
In some embodiments of the present application, defining a working condition situation according to basic information of a control system, generating a working condition change set includes:
Acquiring historical data of a plurality of complete control processes of a control system, establishing a data time axis, and intercepting the data time axis between a starting node and an ending node of an original working condition to obtain a first data time axis;
The basic information of the control system comprises internal disturbance, external disturbance and performance parameters, and the overall disturbance quantity is determined according to the internal disturbance and the external disturbance;
determining a forward delay length and a backward delay length through the integral disturbance quantity, forward delaying an original working condition starting node of the first data time shaft through the forward delay length, and backward delaying an original working condition ending node of the first data time shaft through the backward delay length;
The extended first data time axis is recorded as a second data time axis, and various performance parameter change curves are established to obtain a performance curve;
screening steady-state parameters in the performance parameters, and determining that the original working condition is a steady-state working condition or an unsteady-state working condition according to characteristic values of the steady-state parameters;
and defining working condition situations based on the steady state condition and the performance curve of the original working condition, and combining each working condition situation according to a logic relation so as to obtain a plurality of working condition change combinations.
In this embodiment, the length of the data time axis represents time, corresponding data exists on the axis, and the performance parameter curve is a curve with time on the abscissa and parameter size on the ordinate.
In this embodiment, the original working condition is the original working condition in the industrial control system. The integral disturbance quantity (weight product) is determined according to the internal disturbance and the external disturbance, the forward extension length and the backward extension length corresponding to different integral disturbance quantities are different, and the corresponding relation can be obtained according to historical experience or other mathematical relations. Therefore, the comprehensiveness of the data is ensured, and the data before and after the working condition occurs has great significance on definition.
In this embodiment, steady-state parameters in the performance parameters are screened, and the original working condition is determined to be a steady-state working condition or a non-steady-state working condition through the characteristic values of the steady-state parameters. Steady state parameters include steady state errors, controller parameters, system matrices, etc.
In some embodiments of the present application, defining the operating condition context based on the steady state condition and the performance curve of the original operating condition includes:
performing time scale alignment treatment on performance curves of different steady-state working conditions, comparing the same kind of performance curves of different steady-state working conditions, and determining the difference of the performance curves;
Taking the performance parameter with the difference of the performance curves exceeding the corresponding first threshold value as a steady-state performance parameter, integrating the steady-state performance parameter, and splitting the steady-state working condition to obtain a plurality of working condition situations;
Performing time scale alignment treatment on performance curves of different unsteady state working conditions, comparing the same kind of performance curves of different unsteady state working conditions, and determining the difference of the performance curves;
and taking the performance parameter with the difference of the performance curves exceeding the corresponding second threshold value as an unsteady performance parameter, integrating the unsteady performance parameter, and splitting the unsteady working condition to obtain a plurality of working condition situations.
In this embodiment, the performance curve comparison manner may include a least square method and polynomial fitting, and the least square method: the two curves were fitted using the least square method, and the goodness of fit was compared. Polynomial fitting: two curves were fitted using a polynomial and the differences of the fitted curves were compared. Other ways of comparing curve differences may be substituted as well.
In this embodiment, the first threshold and the second threshold may be set, for example, by calculating an average value and a standard deviation of the temperature under each steady-state working condition. Setting a first threshold: we decided to set the first threshold to ±2 times the standard deviation of the temperature average. Thus, we can exclude those performance parameters that are within the normal fluctuation range. And calculating the average value and standard deviation of the temperature under each unsteady state working condition. Setting a second threshold: since the temperature change under non-steady state conditions is large, we decided to set the second threshold to be + -3 standard deviations of the temperature average. In this way, more variations may be included. The average temperature under steady state conditions was assumed to be 25C with a standard deviation of 1 ℃. First threshold = 25 ± 2 x 1 =23-27 ℃. The average value of the temperature under the unsteady state working condition is assumed to be 30 ℃, and the standard deviation is assumed to be 3 ℃. Second threshold = 30 ± 3 =3 =21 °c-39 ℃.
In this embodiment, the type with the excessively large difference of the performance parameters is used as the splitting basis, when the unsteady state working condition is split or the steady state working condition is split, the steady state working condition is split into a plurality of working condition situations because of different fluctuation interference conditions, the unsteady state working condition is split into a plurality of working condition situations, and the accuracy of the working condition is ensured. For example, a steady state operating condition is split into 3, 4 operating condition scenarios, and an unsteady state operating condition is split into 9, 10 operating condition scenarios.
In this embodiment, steady-state performance parameters are integrated, and the steady-state conditions are split. And integrating the unsteady state performance parameters and splitting the unsteady state working condition. The comprehensive performance parameters are integrated into a comprehensive index, and the range of the comprehensive index is divided into a plurality of intervals, so that a plurality of working condition situations are obtained.
Step S102, determining a prediction model to-be-replaced list corresponding to each working condition change combination based on the target of the control system, and analyzing model information of each prediction model in the prediction model to-be-replaced list.
In this embodiment, it is necessary to determine in advance which of the adaptive prediction models correspond to the working condition changes, and consider the information of these prediction models, so as to help to optimize the model parameters.
In some embodiments of the present application, determining a list to be replaced of a prediction model corresponding to each working condition change combination based on a target of a control system includes:
the targets of the control system comprise stability, accuracy and rapidity, and the stability index target value, the accuracy index target value and the rapidity index target value are obtained by quantifying the stability index target value, the accuracy index target value and the rapidity index target value;
Analyzing the present value and the change trend of the stability index, the present value and the change trend of the accuracy index, and the present value and the change trend of the rapidness index of each working condition change combination;
determining the model ranking according to the difference between the present value of the stability index and the target value, the difference between the present value of the accuracy index and the target value and the difference between the present value of the rapidity index and the target value;
Based on the stability index present value and the change trend of the working condition change combination, the accuracy index present value and the change trend, the rapidness index present value and the matching degree of the prediction model in the change trend evaluation model library;
And sequencing the prediction models according to the matching degree, and adding the prediction models conforming to the model ranking into a list of the prediction models to be replaced.
In this embodiment, stability: stability refers to the ability of a system to maintain or recover to a steady state when subjected to external disturbances or parameter changes. Accuracy: accuracy refers to whether the output of the system can accurately track the input signal or reach a predetermined target value. Rapidity: described rapidly is the rate at which the system changes in response to an input, i.e., the time it takes for the system to transition from one state to another.
In this embodiment, the stability index target value, the accuracy index target value, and the rapidity index target value are set values to be realized. And determining the model ranking according to the difference between the present value of the stability index and the target value, the difference between the present value of the accuracy index and the target value and the difference between the present value of the rapidity index and the target value, wherein the three differences are in common correspondence with one model ranking.
And step S103, when the working condition changes, model information corresponding to the working condition change combination is obtained, and the reference track is determined through the model information.
In this embodiment, when the working condition changes, it is determined which working condition change combination belongs to, so as to obtain corresponding prediction model information, and a reference track is established, and the reference track mainly plays a role in buffering, so that subsequent model parameter adjustment is facilitated.
In some embodiments of the present application, determining a reference trajectory from model information includes:
Determining the performance of each prediction model in the prediction model to-be-replaced list according to the model information;
and integrating a plurality of prediction models through the distribution weight of the performance determination prediction model to generate a reference track.
In this embodiment, the system may be guided to transition to the control stage by setting the reference trajectory, so that subsequent optimization is facilitated. The trajectory may also include model parameters such as predicted time domain, control time domain, and softening coefficients.
Step S104, control quality indexes are defined based on basic information and targets of the control system, the control system is controlled in a buffering stage according to the reference track, and the control quality indexes in the buffering stage are monitored.
In this embodiment, the control quality index is used to describe the control effect, and is used as a basis for adjusting the model parameters.
In some embodiments of the present application, defining a control quality index based on basic information and a target of a control system includes:
Determining a control quality index according to the performance parameter and a target of the control system; ; wherein P is a control quality index,/> Is the target value of stability index,/>Stability index real-time value,/>Is the target value of accuracy index,/>For the real-time value of accuracy index,/>Is a rapidity index target value,/>For the real-time value of the rapidity index, exp is an exponential function, n is the performance parameter category,/>For the weight corresponding to the ith performance parameter,/>For the i-th performance parameter, k is a preset constant,/> Is a control quality function with respect to the differences between the three target values and the real-time value.
In the present embodiment of the present invention,Indicating the quality of control by modification of the performance parameter.
Step S105, optimizing the prediction model parameters of the control system by controlling the quality index.
In this embodiment, if the control quality index meets the expectation, the reference track is analyzed to adjust the prediction model parameters. And if the control quality index does not accord with the expectation, optimizing the prediction model parameters of the control system according to the difference between the similarity and the corresponding threshold value and the basic information of the control system.
In this embodiment, the corresponding threshold may be generally set according to the overshoot of the system, for example, if the control system requires no more than 5% of the overshoot, the corresponding threshold may be set to 5% of the overshoot.
In some embodiments of the present application, optimizing the predictive model parameters of the control system by controlling the quality index includes:
presetting a control quality index standard curve corresponding to a reference track;
Establishing a control quality index change curve through the control quality index, and comparing the control quality index standard curve with the control quality index change curve to obtain the similarity;
If the similarity exceeds the corresponding threshold, optimizing the prediction model parameters of the control system by analyzing the reference track;
Otherwise, optimizing the prediction model parameters of the control system according to the difference between the similarity and the corresponding threshold value and the basic information of the control system.
In some embodiments of the present application, optimizing prediction model parameters of a control system according to differences between similarities and corresponding thresholds and basic information of the control system includes:
The prediction model parameters comprise a prediction time domain, a control time domain and a softening coefficient;
The causal indexes of the prediction time domain, the control time domain and the softening coefficient are calculated respectively through basic information, and the causal indexes meeting the requirements are screened out;
Respectively constructing influence functions of a prediction time domain, a control time domain and a softening coefficient according to the respective corresponding causal indexes;
integrating influence functions of a prediction time domain, a control time domain and a softening coefficient to obtain a comprehensive influence function;
and adjusting the comprehensive influence function based on the difference between the similarity and the corresponding threshold value, thereby determining the set values corresponding to the prediction time domain, the control time domain and the softening coefficient, and optimizing the prediction model parameters of the control system.
In this embodiment, the prediction time domain, the control time domain, and the softening coefficient are predicted.
The prediction horizon determines how many steps in the future to consider in the control algorithm. When the predicted time domain is larger than the control time domain, the control quantity is kept unchanged, and the control quantity of the last step of the control time domain is used as a subsequent control input. The size of the predicted time domain directly affects the overshoot and the adjustment time of the system. A tradeoff typically needs to be made between overshoot and settling time, as a smaller system overshoot typically means a longer settling time.
The control time domain determines how many future control amounts are calculated at each sampling instant. The control amount in the control time domain is calculated based on a predictive model of GPC. If the prediction model is mismatched with the actual process, the longer the control time domain is, the larger the error accumulated by the model deviation is, and the calculated control quantity is inaccurate. Therefore, when the model is mismatched, the control horizon should be appropriately reduced to reduce the cumulative bias.
The softening factor is related to the tracking performance of the system. If the system is over-tuned, the tracking effect can be improved by adjusting the softening coefficient. The softening factor makes the set point changes input to the predictive model smoother than the step changes. Increasing the softening factor may make the system tracking smoother, while decreasing the softening factor may make the system tracking faster, but may increase the risk of concussion.
The above description refers to the context of the prediction time domain, the control time domain, and the softening coefficient, and the influence function is constructed by considering the above description.
In this embodiment, the causal indexes of the prediction time domain, the control time domain and the softening coefficient are calculated respectively through the basic information, and the causal indexes meeting the requirements are screened out. The basic information contains influencing factors influencing the three, and the accurate influencing function can be established by considering the contents. The specific implementation steps are as follows:
The causal index is a nonlinear interdependence index, and is a method based on state space reconstruction and neighbor distance, and is used for judging the direction and the size of the causal relation. For two independent systems or factors X and Y, the state space of the two systems is established according to the state space reconstruction theory.
For sample point X n,xrn,1,...,xrn,k in state space X. Representing k adjacent points of X n in a state space X, and calculating the Euclidean distance average value of X n and the k adjacent points; ; for a sample point Y n,ysn,1,...,ysn,k in the state space Y to represent k neighboring points of Y n in the state space Y, mapping the k neighboring points into the state space X, and calculating the euclidean distance average value of X n and k neighboring points X sn,1,...,xsn,k; /(I) ; To simplify the calculation, the average distance of x n from all N sample points can be used; /(I); The nonlinear interdependence index is a state space method, and the causal relationship between the systems judged according to the mapping relationship of the state space is defined as: ; is available according to definition,/> When/>Approaching 0, systems X and Y are independent of each other, when/>Significantly greater than 0, there is a causal relationship from system X to Y, the closer to 1 the stronger the causality.
It should be noted that, the other indexes capable of characterizing the causal relationship are all the same, and the application provides only one specific mode.
In this embodiment, the influence functions of the prediction time domain, the control time domain and the softening coefficient are respectively constructed according to the respective corresponding cause and effect indexes, and the form of the specific influence function is not limited here, as long as these satisfactory factors can be considered.
In this embodiment, the influence functions of the prediction time domain, the control time domain and the softening coefficient are integrated to obtain a comprehensive influence function. Integrating the three means that the three also affect each other, considering the effect of the three on each other. In practical applications, these three parameters are usually interdependent and need to be considered in combination to achieve optimal control. For example, increasing the prediction horizon may require a corresponding increase in the control horizon to ensure effective control over the entire prediction horizon. Meanwhile, the adjustment of the softening coefficient also needs to consider the selection of the prediction time domain and the control time domain so as to ensure the stability and the performance of the control system.
In this embodiment, the comprehensive influence function is adjusted based on the difference between the similarity and the corresponding threshold, and the product of the adjustment factor and the comprehensive influence function is corrected for different adjustment factors by different differences.
The plurality of correspondence relationships related to the present embodiment may be obtained from historical experience or may be obtained by solving from a mathematical model, and the specific mode is not limited.
By applying the technical scheme, basic information and targets of the control system are acquired, working condition situations are defined according to the basic information of the control system, a working condition change set is generated, and the working condition change set comprises a plurality of working condition change combinations; determining a prediction model to-be-replaced list corresponding to each working condition change combination based on a target of the control system, and analyzing model information of each prediction model in the prediction model to-be-replaced list; when the working condition changes, model information corresponding to the working condition change combination is obtained, and a reference track is determined through the model information; based on basic information and target definition control quality index of the control system, controlling the control system in a buffering stage according to the reference track, and monitoring the control quality index in the buffering stage; and optimizing the prediction model parameters of the control system by controlling the quality index. According to the application, the working condition situation is defined according to the basic information of the control system, the working condition change set is generated, all conditions of working condition change are defined, and the subsequent analysis of the working condition change is convenient. By defining and monitoring the control quality index, the prediction model parameters of the control system are optimized, and the accuracy and adaptability of the model are improved.
From the above description of the embodiments, it will be clear to those skilled in the art that the present invention may be implemented in hardware, or may be implemented by means of software plus necessary general hardware platforms. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (may be a CD-ROM, a U-disk, a mobile hard disk, etc.), and includes several instructions for causing a computer device (may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective implementation scenario of the present invention.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present application, and are not limiting; although the application has been described in detail with reference to the foregoing embodiments, it will be appreciated by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not drive the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present application.