An Analog Parallel Neural Network Circuit for Auto-Detection of Partial Discharges in Power Cables

Abstract

In order to realize an anto-detection system of a single pulse of partial discharge in power cables, a neural network (NN) system with very fast computation (no more than I ms) are desired. In this paper, we present a trial manufactured auto-detection NN system using an analog parallel circuit. The feature of this system is adoption of a Random Weight Change learning algorithm, which is suitable for the analog processing without smooth sigmoid function. Experimental results of a manufactured circuit are also investigated.