Optimal disturbance rejection control approach based on a compound neural network prediction method

For traditional MPC, the deviations between the predicted output and the real output of system may become large when meeting strong disturbances. Therefore, the MPC control law obtained by the objective function with a receding horizon optimization will not be optimal, which makes the dynamic performance of the closed-loop system become sluggish and the closed-loop system may even become unstable. The “feedforward compensation + feedback controller” compound method is only a simple superposition of two control effects, in which the feedforward compensation does not participate in the receding horizon optimization process, and how to coordinate the effects of the feedback and feedforward controllers is an important problem to address. In this paper, an optimal disturbance rejection control approach based on a compound neural network prediction method is proposed by considering the disturbances and model mismatches. The novelty of this paper lies in that the estimate of disturbances by a disturbance observer is introduced into the neural network predictive model, which makes the predictive model output more accurate in the presence of disturbances and model mismatches. Thus, the feedback control law obtained by the objective function with a receding horizon optimization is optimal with strong disturbance rejection ability. The proposed scheme is applied to control the temperature of a simplified jacketed stirred tank heater (JSTH). Simulation results demonstrate the effectiveness of the proposed control method.