Application And Research Of Supervisory Predictive Control Algorithm

Predictive control as a powerful tool in processing modern industrial process control has attracted wide attention. And Supervisory Predictive Control (SPC) algorithm is a typical predictive control algorithm, because it has good control performance and economic performance, based on previous research, for a variety of applications in industrial process, a certain degree of improvement and research is done.The main points and innovation of this thesis can be summarized as follows:(1) For nonlinear systems using multiple models, based on the special condition, the local predictive controller is designed for each linear model independently, and then weighted linear output according to the modified weighted method. The selection method of weighted value is improved, the supervision predictive control algorithm is firstly induced in object function, not only contains the error indicator, but also contains economic indicators, compared with the traditional design method, it can prevent the appearance of error, and at the same time control amount of fluctuation can be suppressed, thereby the predictive output overshoot is prevented. And the simulation results show that the feasibility and effectiveness of control algorithm, which provides a strong theoretical basis for industrial process.(2) For a class of nonlinear time-varying systems described by T-S model, using multi-step linear prediction of fuzzy control strategy, linear-ed into a linear state-space model at each sampling point, and then uses supervisory predictive control algorithm to simulate and verify the feasibility of the algorithm. In the process of optimizing set-point, combined with the Genetic Algorithm(GA). Through single-step fuzzy and multi-step fuzzy SPC and traditional linear and nonlinear simulations to verify multi-step fuzzy SPC based on GA has better control performance.(3) To establish system model by using Least Squares Support Vector Machine (LS-SVM) regression thought. LS-SVM based on Cauchy distribution weighted is proposed to establish the system model, and then uses the model identified in SPC algorithm, Simulation results show that control method based on LS-SVM based on Cauchy distribution weighted has good control performance compared with the control method based on support vector machine and the traditional weighted LS-SVM.(4) LS-SVM regression model established as the model of SPC algorithm, and in the process of optimizing set-point the improved Particle Swarm Optimization is used to calculate the set-point value dynamically. And to simulate through practical examples.(5) For the overshoot problems of predictive output, based on the existing literates, stepped SPC algorithm is improved, and single-step predictive output of stepped SPC is proposed. By introducing single-step predictive output difference item so as to achieve a certain purpose, and combined with examples to simulate.