Adaptive control of discrete systems using neural networks

Etxebarria (1994) has presented a simple adaptive control scheme for discrete systems using two linear two-layered neural networks. Specifically, one of these networks is used to learn online the dynamics of the unknown plant using the Widrow-Hoff delta rule. The other network uses this learning to adjust its connection weights and to generate the control signal. Etxebarria has proven that the resulting closed-loop system is globally stable and has shown that the controlled output tracks the reference signal asymptotically. The author states that there is room for improving the performance of the adaptive neural control scheme of Etxebarria, in particular, its transient performance. In this correspondence, a method is proposed to enhance the transient performance of the above control scheme by replacing the output y of the unknown plant by a linear combination of y and its derivative y˙. Furthermore, the proposed method does not change the structure of both the neural estimator and controller and, as such, the increase in overall computation is minimal. Two examples, based on simulation and experimental studies, are presented to demonstrate the effectiveness of the proposed method