LEARNING RATE ALGORITHM FOR NEURAL-NET BASED DIRECT ADAPTIVE-CONTROL

An algorithm to determine the learning rate for neural net based direct adaptive control is presented. The proposed learning rate ensures fast convergence of the adaptive controller and at the same time ensures local stability of the control system. The derivation is based upon error linearization in the weight space. The use of the proposed learning rate is demonstrated through application on direct self-tuning control of nonlinear plants. Simulation results are also presented to validate the theoretical findings.