Adaptive learning control of affine nonlinear systems using piecewise linearly trained networks

This paper presents an approach of online adaptive learning control for a discrete-time affine nonlinear system with relative degree greater than one. For the system identification and control computation, we use a universal approximation model employing self-organizing and piecewise linear fitting techniques for fast training. The computational load for adaptation of our approach is similar to that of the linear adaptive control. Moreover, the present controller retains the trained control information about nonlinear systems with time varying operating points. As a result, unlike the linear adaptive control, the present control system can quickly adapts itself to any situation similar to the previously trained one. The effectiveness of our method is demonstrated by simulations.