Classification Of Power Quality Disturbance Signals Base On Convolutional Neural Network

The high-speed development of modern society increasingly depends on electronic equipment.The equipment failure caused by power quality disturbance(PQD)has become a critical problem that threatens people's life and property safety.The research of power quality has attracted great attention of researchers and industry.The identification and detection of power quality has gradually become the subject of many researchers.Aiming at the problems of power quality disturbance signals,such as various types,complex composition,unclear feature extraction and low classification accuracy,this paper studies the feature extraction,feature selection and recognition classification of PQD signals.(1)In the feature extraction stage,in order to extract more representative features from PQD signals,this paper used S-transform,wavelet transform and Hilbert Huang transform to extract feature vectors.By S-transform,63 features of PQD signal are extracted from the matrix.PQD signal is decomposed by wavelet transform in 5 layers,and 55 features are extracted by 5 layers of detail coefficients.7 features are extracted by Hilbert marginal spectrum of PQD signal.A total of 125 features constituted the original feature set.(2)In the feature selection stage,in order to simplify the feature vector and improve the classification accuracy,this paper adopted the feature selection algorithm based on Relief F and classification and regression tree,and compared their selection results.By calculating the weight value of each feature,the Relief F algorithm selects the feature with the largest weight value,and then determines the feature subset.The classification and regression tree algorithm calculates the Gini importance value of each feature and sorts it to select the feature subset.(3)In the identification and classification stage,For most of the existing power quality disturbance classification algorithms,the data preprocessing and feature selection are required,thus resulting in poor robustness and weak anti-noise performance.This paper proposed an improved one-dimensional convolutional neural network(1D CNN)model and classified the power quality disturbance signals.The feature vectors of the power quality disturbance signals were extracted by three sub-models of the proposed algorithm,and then the feature vectors were combined and input into the BP neural network for classification.Comparing with the 1D CNN and many algorithms,the simulation results show that the proposed algorithm has better robustness and recognition rate,and it can accurately classify the power quality disturbance signal under strong noise environment.