Nonlinear model predictive control based on multiple neural networks

Improved predictions can be obtained by using multiple neural networks instead of trying to find a single optimal network as usual. All the component networks of MNN model are selected using generalized information entropy, and the accuracy and reliability of overall model are significantly improved. Based on such an MNN model, a new nonlinear model predictive control algorithm is proposed. Simulation results of a pH CSTR demonstrates that the method is effective and practical.