The-line Model Of The Ls-svm-based Nonlinear Model Predictive Control

With the development of modern industry and progress of science and technology, the process industry is changed to more complex and with strong nonlinear characteristics so that linear model predictive control (LMPC) may not always obtain satisfactory control performance . Therefore, predictive control based nonlinear model (NMPC) has become an important research issue in the control engineering fields . Most nonlinear predictive control methods are based on off-line model, but due to the migration of system work domain, off-line data model can not accurately describe the system and the actual situation, which lead to the predictive control based on off-line model can not achieve the desired effect of the real-time control . To solve this problem, The problem for the online LS-SVM-based nonlinear model predictive control is studied in this thesis .Some problems of nonlinear predictive control are researched in this thesis based on the previous research work, and the main research works are as follows:(1) Aiming at the problem of LS-SVM lacking robustness, a weighted LS-SVM is proposed. The new method takes time and similar factors as a weighted factor, simulation results show that the robustness of the new method can be effectively improved .(2) A kind of NMPC algorithm based on online weighted LS-SVM is proposed to solve the defect of NMPC based on offline model . This method established online model of the system using the weighted LS-SVM, and then particle swarm optimization algorithm acted as rolling optimization strategy . Simulation results show that the effect of NMPC based on online model is better than that of the NMPC based on offline model.(3)Aiming at the defect of off-line clustering modeling, a modeling method based on weighted LS-SVM online clustering method is proposed. A kind of SVM based on weighted LS-SVM online clustering model of multiple model predictive function control (PFC) algorithm is proposed in order to solve the problem of single model of nonlinear predictive function control which is needed linearization in every sampling, and the predictive function control law of multiple input multiple output system is derived . The simulation results show that the control effect and the anti-interference ability of multiple model PFC based on model online clustering is better than single offline model PFC .