Studies On The Algorithms And Applications Of SVM And PSO Based Nonlinear Model Predictive Control Systems

The purpose of this study is to develop a novel nonlinear model predictive control system with computational efficiency for real-time optimization and good robustness when the model parameters change. In addition to providing the real-time optimization framework, the study also compares the optimization and closed loop control performance between linear and nonlinear MPC.Since the superior performances of SVM in generalization and the capability of dealing with high dimensional systems over neural networks, the SVM modeling is selected in designing NMPC.The particle swarm optimization (PSO) algorithms are based on artificial life and evolutionary computation. The PSO is applied in on-line nonlinear optimization, since its merits in computation speed, required memory and tuning parameters. On other hand, the good search capability and fast convergence make PSO applicable to real optimization control. Finally, the ability of dealing with constraints and good robustness are another factors to select PSO in this study.The proposed SVM-PSO nonlinear model predictive control has also been verified through two benchmark plants. The simulation results show the feasibility and applicability of the proposed NMPC in advanced process control.