A Model-Predictive Control Method Based on Hybrid Neural Networks for Nonlinear Systems Described by Hammerstein Model

向微,陈宗海,盛捷

2008-01-01

Abstract:A model-predictive control strategy based on hybrid neural networks for the multi-input and multi-output Hammerstein model is presented, which uses a linear optimal component and radial basis function neural networks in series. By this method, the nonlinear block is not limited to a polynomial equation, thus the requirement of real roots of the polynomial equation in traditional control designs is removed with no deterioration of the optimal performance. On the other hand, the hybrid neural networks are more efficient than the traditional neural networks which require complicated net structures in approximating dynamic mappings and show poor real-time performance.