Design and application of predictive controller using neural networks

This paper presents a design methodology for predictive control of multivariable systems via neural network. The type of proposed controller has its efficiency to deal with the nonlinear multi-input multi-output dynamics. The neural-network-based predictive control law is developed based on a generalized predictive performance criterion. The stabilities of the neural network model and neural network predictive controller are shown via the Lyapunov stability theory. Simulation results reveal that the proposed control gives satisfactory tracking and disturbance rejection performance for two illustrative multivariable discrete-time systems with time-delay. Experimental results for the heating barrel of a plastic injection molding process are conducted which have shown usefulness of the proposed method under the conditions of set-points changes.