Control Strategies Research For Nonlinear System Inverse Models Of RBF Neural Network

Non-linearity is a kind of general phenomena in the physical systems. And nonlinear control plays an important role in the control science. Because of the creation of the theory of inverse system, nonlinear control has improved greatly, especially in the recent years.It becomes the advancing science on which the international control regions emphasis. But the method of inverse system requires the nonlinear part analytical, which restricts its wide applications. Fortunately, neural networks approximate nonlinear mapping easily. This dissertation focuses on the incorporation of inverse system,RBF networks and nonlinear control in order to apply to engineering better. Great research work on algorithm and control strategy has been done by the author. The main contributions of the dissertation are stated as following:(1) The dynamic nearest neighbors clustering algorithm of obtaining the RBF neural network data centers of the hidden layers is raised in this article based on the problem that it is difficult to determine the centers of the Radial Basis Function when we study the RBF network training algorithm. The"crude regulation"and"fine regulation"methods which regulated the cubman radius automatically are introduced in to guarantee the rationality of clubman in order to make the cluster centers optimal and simple the network structure.Thus we can obtain a dynamic self-adaptive RBFN which can adjust parameters and structure adaptively in the same time.(2) By analyzing profoundly the inverse system method and adopting the ANNαth-order inverse system method which has widespread sense, a direct controller based on the RBF networks is presented by the author and derives strict proof of the existence of the controller and reversibility of the SISO system and the MIMO system. A controller can be designed further for the pseudo linear system constructed by the inverse model and the plant. The model of controller and the plant are in series, which forms a dynamic pseudo linear system. An inverse control strategy based on RBFN with feedforward was presented in this paper. The dynamic pseudo linear plant can be controlled by PD controller along with feed-forward control method. The feedforward's contribution is to compensate the effects that the disturbance creates when it appears. The strategy not only can realize pseudolinearity but also can decouple the MIMO system into some independent control loops.(3) Inverse system of the nonlinear non-minimum phase system is instability, so it is difficult to eliminate these defects with conventional linearity control techniques. The neural network inverse control technique based on inverse mapping of system will be failure because the control signals become larger infinitely.The paper can change the non-minimum phase system into the minimum phase system by constructing a pseudo plant, and can consider the time delay system as a generalized minimum phase system. So the scheme this dissertation presents is in common use.