Black-box model identification using neural networks and adaptive control for fast time-varying nonlinear systems

A fast and flexible adaptive self-tuning control (STC) is proposed in this paper for nonlinear, fast time-varying and multi-input multi-output (MIMO) systems using a novel output and error recurrent neural networks (OERNN). The key point of this research for nonlinear control is to develop a fast tracker with a flexible adaptive control scheme which does not require previous knowledge about the plant to be controlled, but black-box model. Hence its algorithms have a flexibility for diverse plant applications. In order to carry out this research goal, system identification has successfully been achieved based on a recurrent neural network model, and nonlinear quadratic (NQ) optimal law has also been derived and tested to the fast tracking problem for a revolute 3-DOF robotic manipulator.