Study On Feature Extraction And Classification For Electrical Equipment Partial Discharge Signal

Partial discharge(PD)is closely related to the internal insulation of electrical equipment and the breakdown of isolator.The mechanism and location of different types of PD are different.And the types of PD affect the insulating condition in varied degrees.So it is of great significance to accurately and rapidly identify the type of partial discharge in electrical equipment.Based on the analysis of the characteristics of PD,feature extraction and classification method of PD signal are proposed.The main contributions of this paper are as follows:A new method for partial discharge feature extraction based on Variational Mode Decomposition(VMD)and Multi-scale permutation entropy(MPE)was proposed to solve the poor stability and low recognition rate problems.Four typical PD signals produced by discharge models in the laboratory were firstly decomposed into several band-limited intrinsic mode functions(BLIMFs)with different frequency-bands by VMD.Its MPE is then used to produce original characteristic quantities.Meanwhile,the original feature vector dimension was reduced according to max-relevance and min-redundancy criteria(mRMR).Support vector machine(SVM)classifier was employed to classify the PD signal.The complexity and uncertainty of PD signal in difference frequency band were described more effectively by the feature extracted by the proposed method.The proposed method is more robust and its recognition rate is higher.A Variable Predictive Model based Class Discriminate method(VPMCD)is constructed to classify PD signals.In this paper,37 types of statistical characteristics and 9 types of time-frequency characteristics were extracted.PD patterns were classified by Variable Predictive Model based Class Discriminate method(VPMCD).Experimental results demonstrate that VPMCD algorithm gains a more recognition rate and a better computational efficiency than BP Neural Network and SVM.Aiming at solving the performance degradation problem caused by small number of PD samples,an improved Variable Predictive Model based Class Discriminate method(VPMCD)was proposed.Orthogonal complete basis and moving least square method were applied to building and solving the prediction model,which increased the accuracy of model.Experimental result shows that the improved algorithm gains a higher recognition rate than BP Neural Network,SVM and VPMCD with small samples.