Power Quality Disturbances Classification Based On Wavelet Transform And Artificial Neural Networks

In order to analyze and study the phenomenon of power quality, find the cause of power quality issues and take appropriate solutions, it is very significant to classify power quality disturbances correctly. First this paper introduces the issue of power quality, analyzes the existing classification methods of power quality disturbances in depth. Then, wavelet transform (WT), artificial neural networks (ANN), particle swarm optimization (PSO) algorithm, principal component analysis (PCA) and kernel principal component analysis (KPCA) are used to classify power quality disturbances. The work of this paper mainly includes two aspects:(1)A method classifying power quality disturbances is presented based on wavelet transform and PSO-BP neural network.For the shortcomings of gradient descent algorithm of traditional BP algorithm, such as slow convergence and being into local minima easily, a new algorithm is proposed based on PSO algorithm to improve neural network BP algorithm, that is, PSO-BP algorithm. Then a large number of power quality disturbances waveform data are got using PSCAD/EMTDC simulation software and applied to wavelet transform and multi-solution analysis. At last the signal power variations at each scale are extracted as feature vectors and input PSO-BP network to classify power quality disturbances. Simulation results show that this method has the rapider convergence speed and meet the training requirements more easily than the BP algorithm.(2)Another method classifying power quality disturbances is also presented based on PCA+KPCA of wavelet coefficients feature and probabilistic neural networks (PNN).For the shortcomings of the method that used directly each scale wavelet coefficients to classify power quality disturbances, such as the large input volume and slow learning velocity, a new method to extract disturbances feature is presented based on PCA+KPCA of wavelet coefficients feature. First the mathematical models of disturbances are established to get many waveform data. The waveforms are applied to wavelet transform and multi-solution analysis to get the wavelet coefficients of each scale. Then PCA+KPCA is applied to reduce wavelet coefficients feature vectors dimension. At last they are put into PNN to classify. Simulation results show that this method not only achieves the purpose of reducing the input quantity but also improves the speed of feature extraction and has favorable classification precision.In this paper, the above mentioned two methods of power quality disturbances classification make improvements respectively in both neural network classification and feature extraction for the limitations of previous methods. Simulation results verify the feasibility and effectiveness of the two methods in power quality disturbance classification.