Nonlinear Adaptive Inverse Control Based On RBF Neural Networks

The conventional feedback control tried to heighten the system performance and to cancel disturbance simultaneously through feeding back output and disturbance. It could not get the optimal solutions for control performance and disturbance attenuation at the same time. We can only make a tradeoff between them. However adaptive inverse control can deal with the two problems separately. It doesn't need too much prior knowledge. We can design adaptive inverse controller without knowing plant's mathematic model. In adaptive inverse control systems, the control of plant dynamic response and disturbance-canceling do not interact each other. We can not only obtain good dynamic response, but also make the influence of noise and disturbance to the minimum.Neural network has the characteristic of approaching any nonlinear function, processing data in parallel and distribution, learning and adapting. It is suitable for adaptive inverse control. Radial Basis Function Neural Network is an effective feedforward network. It has high convergence rate and high approaching precision, and can avoid local optima. Systematic analysis and research are made to the various learning methods of RBFNN. The key factor that influences RBFNN's performance is the choice of RBFNN's hidden layer center. First, the usual ways that are employed to choose the number of RBFNN's hidden layer nodes are analyzed and compared. As to RBFNN's weight training, the normal way is RLS which relies on the implicit or explicit computation of the inverse of the input signal's autocorrelation matrix. This not only implies a higher computational cost, but it can lead to instability problems. As a result, we select RPCL-LAF algorithm. It can satisfy the demands of modeling. The ε-filtering algorithm can deal with nonlinear system well and is insensitive to the imprecision of plant model and its delayed inverse model. So we use the ε-filtering algorithm in the controller design. The simulation results show that our control scheme is effective.