Comparison of feed forward and cascade neural network for harmonic current estimation in power electronic converter

This paper presents harmonic current estimation using neural network for a power electronic converter. Three types of popular neural architectures namely single hidden layered Feedforward architecture, multi hidden layered Feedforward neural architecture, cascade architecture are considered for investigation. The non-linear load namely diode bridge uncontrolled rectifier with resistive inductive (RL) load is chosen for study. All the three architectures are trained and tested using MATLAB simulation. The performance of three types of neural architectures is compared in terms of accuracy and complexity for harmonic current estimation. The suitable neural architecture is identified for harmonic current estimation. The results obtained are presented.

1. Introduction

Power electronic converters are popularly used for various applications. Such converters introduce harmonics in the grid and makes the current non-sinusoidal in nature [1], [2]. Power electronic converters deal with complex voltage and current waves that are rich in harmonics. Power Quality is often defined as the interaction and coordination of electrical power with electrical equipment. The major power quality issue is harmonics which are mathematical model used to analyse the current drawn by computers, electronic ballasts, arc furnaces, variable frequency drives etc. Hence harmonics estimation in electric power systems is highly essential.