A Study On Application Of Support Vector Machine In GPC With Real Test Analysis

This paper mainly involves the applications of Support Vector Machine (SVM) in Generalized Predictive Control (GPC), and focuses on some of the application problems with real-test analysis. The basic idea of SVM is to map linear inseparable input data into a high dimensional and linear separable feature space via a nonlinear mapping technique (kernel dot product), and classification or regression is done in the feature space. With good ability in generalization and nonlinear function estimation, SVM is being widely used in control theory areas such as system identification. GPC as one part of modern control theories has good control performance, excellent disturbance regression ability and strong robustness. So it has been widely used in industrial process control areas. The paper combines the SVM system identification method with GPC based on predictive model, which aims to exert the virtues of both SVM and GPC. In detail, this paper mainly involves:1. Build the testing platform for advanced process control algorithm based on Matlab7.0 and KINGVIEW. This platform is used for real-test analysis of advanced process control methods in this paper.2. The CARIMA model which the basic GPC is based on is identified by SVM modeling with linear kernel function for linear system and weakly nonlinear system. Then the GPC method based on CARIMA model is implemented to control the linear or weakly nonlinear system. Both simulation and real test show the effectiveness of the method. As to strongly nonlinear system, its model is identified by SVM with polynomial kernel function. Secondly, the model is linearized at each sample time, which makes the strongly nonlinear system into a time-varying linear system with CARIMA expression. Then the basic GPC method is introduced to control the strongly nonlinear system. To overcome the instability caused by model mismatchwhen linearizing the system, β -type incremental GPC is introduced to improve thesystem's stability.3. To restrain the instability of the basic GPC caused by model mismatch, this paper presents an error prediction model based on SVM for model correction according to past error data. The error prediction model is used for error compensation. With both simulation and real test in liquid level control, we show the effectiveness of theimproved GPC method.With the study of this paper, we not only broaden the application scope of the basicGPC method, but also improve the performance of GPC method to some extent.