Adaptive Model Predictive Control Using Diagonal Recurrent Neural Network

Abstract

A neural network-based model predictive control scheme is proposed for nonlinear systems. In this scheme an adaptive diagonal recurrent neural network (DRNN) is used for modeling of nonlinear processes. A recursive estimation algorithm using the extended Kalman filter (EKF) is proposed to calculate Jacobian matrix in the model adaptation so that the algorithm is simple and converges fast. Particle swarm optimization (PSO) is adopted to obtain optimal future control inputs over a prediction horizon, which overcomes effectively the shortcoming of descent-based nonlinear programming method on the initial condition sensitivity. A case study of biochemical fermentation process shows that the performance of the proposed control scheme is better than that of PI controller.