Research Of Multi-objective Predictive Control For Model Mismatch

Predictive control is an advanced control technique based on model. Because of varying modeling method, strong robustness and effective control, predictive control has successfully applied in petroleum, chemical industry, metallurgy, mechanism and many other fields. As a control method based on controlled plant's feature information, high degree of accuracy of model is required in predictive control method. In the event that bias of prediction model is on the small side, predictive control method will obtain satisfactory control effect. But, in contrast, if prediction model is mismatch with controlled plant, control effect will influenced directly. Studying of predictive control method for model mismatch has important theoretical and realistic signification.This text carries on research on predictive control problem of model mismatch. Base on predictive control theory, neural network modeling and multi-objective optimization theory, a multi-objective predictive control method is proposed for model mismatch. The main points of research contents as follows:(1) The text introduces the related theories on predictive control and nonlinear predictive control. Then simulation research on predictive control method is carried on based on RBF neural network prediction model and particle swarm optimization algorithm. Simulation results showed that predictive control method has good control performance when the prediction model has high precision. But the predictive control method will not work in model mismatch. Then a new problem of predictive control method under model mismatch.(2) The text introduces multi-objective optimization and decision-making theory in detail and proposes an improved multi-objective particle swarm optimization algorithm (SMOPSO). The simulation result shows the effectiveness of the proposed method by simulation experiment of a couple of test function. Meanwhile, a preference multi-objective particle swarm optimization algorithm is proposed based on SMOPSO, weight coefficient probability density function and Pareto theory. The simulation result shows the effectiveness of the proposed algorithm by simulation experiment.(3) Based on above-mentioned theory, this text proposes a multi-objective predictive control algorithm which selecting controlled quantity form non-dominated set using model error probability distribution statistic information. The single objective predictive control method and multi-objective predictive control method proposed by this text are separately used to do simulation experiment for two test controlled plants. Simulation result shows that the method proposed by this text can improve control performance efficiently for model mismatch.