Performance Comparison of Neural Network Training Approaches in Indirect Adaptive Control

Abstract

This paper proposes an indirect adaptive control method using neural network (NN) based on a variable learning rates (VLRs) combined with Taylor development (TD) for nonlinear systems. In the proposed control architecture, two neural network blocks are used both as an identifier and a controller. The tracking error and the identification error are used, respectively, to train the neural controller and the neural model. The NN identifier approximates dynamic systems and provides the NN controller with information about the system sensitivity. The gradient-descent method using a developed variable learning rate is mixed with the Taylor development and is applied to train all weights of the NN. The NN TD-VLRs are applied to guarantee the convergence of the proposed control system. The effectiveness of the proposed algorithm applied to an example of nonlinear dynamic systems is demonstrated by simulation experiments. The results of simulation show that applying the mixed proposed method ensures the smallest MSE and the optimal time simulation. Added to that, the neural network controller is insensitive to variations of the system parameters.