Multi pulse rectifier classification using scale selection wavelet & probabilistic neural network

Abstract

Three phase multi pulse rectifier classification using scale selection wavelet and probabilistic neural network is presented in this paper. The scale selection wavelet selectively perform continuous wavelet transform on the desired scales, which are determined by the scale frequency relationship to precisely locate each harmonic center frequency for harmonic analysis. Thus, the continuous wavelet transform selectively transform only the 16 characteristic harmonic frequencies of interest from 2nd to 25th order, which are required for three phase multi pulse rectifier classification. The 16 characteristic harmonic frequencies energy are used as the input vector to the probabilistic neural network to classify 5 types of three phase multi pulse rectifier including 3, 6, 12, 18 and 24 pulse converter. Various sets of harmonic distortion signals are used to evaluate the performance of these wavelet and neural network based classification system. The results show excellent performance in terms of high accuracy in classifying harmonic distortion caused by three phase multi pulse rectifier. These harmonic classification information serves as guideline to develop and optimize mitigation solution to reduce harmonic disturbance and resonance problem in the industry facility.