U-neural network-enhanced control of nonlinear dynamic systems

Abstract

Neural networks have been widely used as an approximation for nonlinear dynamic plants in control system design, but they are almost never used as a proper dynamic inverter, particularly with dynamic models available in advance. This study presents a universally feasible U-neural network (U-NN) structure to facilitate the designing control of all dynamic systems modelled with linear/nonlinear polynomial/state space equations. With the presented U-NN, this study proposes a procedure for a model independent control system design, U-control framework/platform. The procedure, against a traditional model based on control system design and a model free/data driven-based design (such as PID control, iterative learning control, and model free control), removes the boundaries of the linear/nonlinear and polynomial/state space model sets, where the model structures are universally treated within the new framework. Furthermore, this study analyses the U-control properties and gives a step-by-step implementation procedure. Several bench test examples demonstrate the effectiveness of the design procedure, as well as the routines in applications.