A neuro-controller with guarantied stability

Abstract

This paper introduces an MLP-based neuro-controller with a learning algorithm that guarantees the stability of a class of closed loop neural network control systems. The underlying control system represents a special class of non-linear systems. The neuro-controller indeed represents a direct adaptive controller and assumes that all the states are measurable. In the design, no additional controllers or robustifying terms are needed. Neural network weight matrices are adapted online with no initial offline training.