Adaptive Direct Inverse Control Scheme Utilizing a Global Best Artificial Bee Colony to Control Nonlinear Systems

Abstract

This paper presents an adaptive direct inverse control (ADIC) scheme which employs a self-recurrent wavelet neural network (SRWNN) structure as a feedforward controller. This intelligent control approach is used to control nonlinear dynamical systems. As an effective optimization technique, the artificial bee colony (ABC) algorithm is employed to find the optimal settings for the SRWNN parameters. In particular, a modified version of the ABC algorithm, which is called the global best ABC (GB-ABC) algorithm, is proposed in this work to train the SRWNN controller. This algorithm has shown better optimization results compared to other optimization techniques. Moreover, a simplified yet efficient online training method is used to enhance the control performance of the SRWNN-based ADIC scheme. Unlike gradient-based methods which require an additional neural network to train the inverse controller, the online training method used in this work adapts the controller parameters directly without the need for an extra forward neural model. To show the effectiveness of the proposed control approach, several nonlinear dynamical systems are considered. Specifically, by conducting several evaluation tests, the ability of the SRWNN-based ADIC to control each of the considered systems is evaluated in terms of control accuracy, generalization ability, and robustness to external disturbances. The results of all these tests have clearly indicated the efficiency of the control scheme. Furthermore, from a comparative study with other related controllers, the SRWNN has demonstrated its superiority in terms of achieving better control accuracy and demanding shorter training time.