Online trained neuro-controller with a modified error function

Abstract

A modified error function to improve the performance of a neuro-controller trained online by the backpropagation (BP) algorithm is presented. The performance is significantly improved with the proposed error function compared to that with the traditional error function used in the BP algorithm. Based on this modified error function, the structure of the network can be greatly simplified leading to a very simple neuro-controller. Performance of the simple network based neuro-controller to control a second order system is presented.