Design and implementation of a neural-network-controlled UPS inverter

Abstract

A low-cost analog neural network control scheme for the inverters of uninterruptible power supplies (UPS) is proposed to achieve low total harmonic distortion (THD) output voltage and good dynamic response. Such a scheme is based on a learning control law from representative example patterns obtained from two simulation models. One is a multiple-feedback-loop controller for linear loads, and the other is a novel, idealized load-current-feedback controller specially designed for nonlinear loads. Example patterns for various loading conditions are used in the offline training of a selected neural network. When the training is completed, the neural network is used to control the UPS inverter online. A simple analog hardware is built to implement the proposed neural network controller; an optimized PI controller is built as well. Experimental results show that the proposed neural network-controlled inverter achieves lower THD and better dynamic response than the PI-controlled inverter does.