A novel model predictive control method was proposed for a class of dynamic processes with modest nonlinearities in this paper. In this method, a diagonal recurrent neural network (DRNN) is used to compensate nonlinear modeling error that is caused because linear model is regarded as prediction model of nonlinear process. It is aimed at offsetting the effect of model mismatch on the control performance, strengthening the robustness of predictive control and the stability of control system. Under a certain assumption condition, linear model predictive control method is extended to nonlinear process, which doesn’t need solve nonlinear optimization problem. Consequently, the computational efforts are reduced drastically. The simulation example shows that the proposed method is an effective control strategy with excellent tracing characteristics and strong robustness.
Constrained Nonlinear Predictive Control Based on Input-Output Linearization Using a Neural Network
