RBF Neural Network based Model Predictive Control Algorithm and its Application to a CSTR Process

Abstract

This paper proposes a model predictive control (MPC) algorithm based on radius basis function (RBF) neural network model, and applies the algorithm to a nonlinear CSTR process. Firstly, the first principle model of CSTR is established based on mass and energy conservations. Then, a RBF-NARX model is trained and validated. Two nonlinear MPC algorithms based on RBF neural network model are derived. One is RBF-MPC based on nonlinear model and nonlinear optimization. The other is based on linearized model at current sampling time, which derives a linear model with time-varying parameters, and the optimization used in MPC can be transformed to a Quadratic Programming, where the global optimality and the online calculating time can be guaranteed. The MPC algorithms are verified by Matlab simulations.