The method of adaptive inverse control is researched deeply in the thesis. Based on the analysis of respective merits and defects of evolutionary programming and simplex method, a hybrid algorithm of evolutionary programming and simplex method-EPSM algorithm is proposed in this thesis. The convergence of the hybrid algorithm is proved by mathematical method. Because the approach includes both the stochastic searching and determinate searching, to a certain extent, it can enhance the optimal ability of EP, so that it is a potential optimization. The traditional methods have some limits in selecting parameters of networks when is used to train RBF neural networks. Moreover, EPSM algorithm has few parameters and is applied easily. This approach avoids the shortcomings of other methods. EPSM and EP are used to train RBF neural networks. The validity and practicability of EPSM algorithm is verified by simulation. The inverse system method needs the precise models of the plant, however, it is difficult for these precise models to be described by accurate mathematical ones. The inverse model of a plant is built by RBF neural network, which is trained by EPSM algorithm. The mult-variable linear and nonlinear couping system is decouped .by the RBFNN inverse controller. The direct inverse controller combined with PID controller is implemented to adaptive inverse control of system. Finally we compare the result of simulation with that of conventional PID in linear system and with that of direct inverse control in nonlinear system.The RBFNN inverse controller is combined with traditional PID controller in this thesis, and this method of adaptive inverse control provides a new approach for nonlinear system control. |