A Wavelet Neural Network-Based NARMA-L2 Feedforward Controller Using Genetic Algorithms to Control Nonlinear Systems

Abstract

As a powerful nonlinear control design strategy, feedback linearization provides viable design tools for a wide range of nonlinear systems. This paper presents an intelligent feedback linearization design using the inverse feedforward control (IFC) scheme to control nonlinear dynamical systems. Particularly, the nonlinear autoregressive moving average (NARMA-L2) network is trained to reproduce the controlled system's forward dynamics. Consequently, the trained NARMA-L2 network can be directly employed in the IFC structure. To enhance the approximation ability of the NARMA-L2 structure, two wavelet neural networks (WNNs) are utilized to constitute the NARMA-L2 controller. Moreover, the RASP1 function was utilized as the mother wavelet function instead of the commonly employed Mexican hat function. To avoid the limitations of the gradient descent (GD) methods, the genetic algorithm has been used as the training method to optimize the NARMA-L2 inverse controller parameters. The simulation results showed that the proposed controller was effective in terms of precise control and robustness against external disturbances. Furthermore, a comparison study with other control structures revealed that the control results of the proposed WNN-based NARMA-L2 controller with the RASP1 function are superior to those of the WNN-based NARMA-L2 with the Mexican hat function, the multilayer perceptron (MLP)-based NARMA-L2 controller, and the PID controller.