Skip to Main Content
This paper proposed a power quality disturbances classification system based on wavelet transforms and novel probabilistic neural network (PNN). Wavelet transform is utilized to extract feature vectors for various power quality disturbances based on multi-resolution analysis. The decomposition signal is divided into 5 equal length bins in each level. Root mean square (RMS) value of the wavelet coefficients that fall within that bin is regarded as a dimension of feature vectors. These feature vectors are applied to a probabilistic neural network for training and testing. Evolutionary algorithm is used to in this paper as a multivariate optimization scheme for finding multiple sigma values in estimation of probabilistic density function. One of the major virtue of PNN stems from its modular architecture design, then it can be easily extended adapt to a changing environment by appropriate chromosomes and generations. We examined that different decomposition levels of wavelet transform are concerned with the classifier accuracy, and the performance of classification is minor distinction with different wavelet families under the condition of same decomposition level.