Optimal Control Based On Support Vector Machine

The batch reactor is a type of vessels widely used in the chemical industry. It has been one of the research hot issues in the control domain. The batch reactor has the characteristics of time-varying, nonlinear, disturbed by noise most of the time. The chemical process is often accompanied by the endothermic, exothermic phenomenon, and complex reaction mechanism.In the production, the main parameter affecting the chemical reaction is the temperature. Its control quality directly affects the quality of the products. This thesis uses the Generic Model Control (GMC) algorithm, which is essentially a kind of nonlinear optimal control methods based on the controlled object information. In the design, this thesis researches on the model of the reactor, and the control signal is generated from the model information and the error. With GMC algorithm, the reactor has the advantages of stability, anti-interference ability.The reactor model information in the GMC optimal control law is Qr, which is the heat released from the chemical reaction process. In the actual production, it can not measure accurately, but in theory, Qr could be calculated from the chemical formula according to the chemical meaning. In order to enhance the practicability of the GMC algorithm, this thesis uses the mixed kernel function SVM algorithm, inputting the related variables of Qr to predict it. In the training process, the parameters of mixed kernel function SVM will affect the accuracy of the model. This thesis adopts an improved inertia weight PSO algorithm to optimize these parameters of SVM model, and it not only enhances the global searching ability in the later period, but also avoiding the PSO algorithm into local extremum.In this thesis, MATLAB simulations of the GMC control and the classic PID control are carried out and compared. The results show that the optimal GMC based on mixed kernel function SVM control strategy has achieved a good performance.