Internal Model Control Using a Self-Recurrent Wavelet Neural Network Trained by an Artificial Immune Technique for Nonlinear Systems


This paper presents a Self-Recurrent Wavelet Neural Network (SRWNN)-based Internal Model Control (IMC) for nonlinear systems. As the internal model, a Nonlinear Autoregressive Moving Average (NARMA-L2) is employed for obtaining a forward system model. Then, this model is directly used to formulate the control law. The proposed SRWNN-based IMC is an enhanced version of a previously published Wavelet Neural Network (WNN)-based IMC scheme. Particularly, the enhancement was attained by considering three modifications, which include the use of an initialization phase for the parameters of the wavelon layer, the utilization of self-feedback connections in the wavelon layer, and the exploitation of RASP1 as the mother wavelet function. The modified Micro Artificial Immune System (modified Micro-AIS) is employed as the training method. From the simulation results, the efficiency of the suggested methodology have been proved concerning control precision and disturbance rejection ability. Moreover, the superiority of the SRWNN over the WNN and the Multilayer Perceptron (MLP) as the IMC controllers has been confirmed from a comparative study. Furthermore, the modified Micro-AIS has accomplished better results compared to the Genetic Algorithm (GA) concerning control precision.