Adaptive inverse control for nonlinear systems based on RBF neural network

An adaptive inverse controller for nonlinear systems is designed by using RBF neural network. The controller consists of a RBF identification network and a RBF control network. An optimization algorithm is proposed for redundant number of hidden units of familiar RBF neural network, and the approach combines the rival penalized competitive learning (RPCL) and the improved regularized least squares (IRLS) to provide an efficient procedure for constructing a minimal RBF neural network that generalizes very well. The RPCL adjusts centers, while the IRLS estimates the connection weights. The effectiveness of the proposed controller is illustrated through a simulated application to a nonlinear system.