Research On Diagnosis Method Of Predictive Control Performance Model Based On Data

Predictive control is a sort of advanced control strategy, therefore, the study on diagnosis technology of predictive control performance model has important theoretic and application value for maintaining and increasing predictive controller performance, enhancing the promotion and application of advanced control strategy.This paper firstly introduces predictive control, particle swarm optimization (PSO) and other relevant basic theories and methods. On this basis, mainly based on the two of the predictive controller performance diagnosis methods based on data puts forward a kind of predictive control performance diagnosis method based on the data and the simulation thought of identification model. Main contents are as follows:Based on performance assessment method of performance characteristics subspace’s predictive control model, the method introduces performance characteristics subspace to describe the characteristics of each performance type, establishes real time data and performance subspaces of various data and calculates the distance among them, using classification analysis and taking distance as measurement merit determines the type in accordance with diagnosis data and locates the reason that causes the performance reduction of predictive control model. In order to make up for the shortcoming of the above method, the paper comes up with a sort of performance assessment of performance characteristics subspace based on PSO. Adopting PSO when choosing performance characteristics subspace makes the selection of performance characteristics subspace more representative. Simulation study is used to test the effectiveness of the method.Based on the performance assessment method of PCA similar factor’s predictive control model, the method by introducing performance characteristics subspace to describe the characteristics of each performance type calculates real time data and PCA similar factor among performance subspace of various data, using classification analysis and taking PCA similar factor as measurement merit determines the type in accordance with diagnosis data and locates the reason that causes the performance reduction of predictive control model. The paper puts forward the performance assessment method that takes advantage of PSO to gain PCA similar factor parameter, and uses simulation results to test the effectiveness of the method. Based on principles, features and applicable range of the above two methods, the paper combines the two assessment methods by neural network classification and fully takes advantage of the indicator information of the two methods, raising a sort of mixed performance diagnosis method. And simulation study verifies the effectiveness of the method.