Indirect adaptive neurocontrol using localized polynomial networks with CLI cells
F. Liang, H.A. ElMaraghy · 2005
The theoretical issues of the indirect adaptive neurocontrol scheme are clarified. First, the d-step ahead prediction input-output representation of sampled data nonlinear systems is established which converts the system invertibility problem into the existence problem of implicit functions. Then the theoretical solutions of indirect adaptive neurocontrol laws are derived. The localized polynomial networks with the competitive lateral inhibitory (CLI) cells are used to realize the neurocontrollers. Fuzzy controllers are incorporated into the control systems to guarantee their large-extent-stability during training stage. Due to the localized networks and the new adaptation law, the indirect adaptive neurocontrol system output tracking errors are fast convergent. The theory was tested by simulations, which proved the above theory.