Font Size: a A A

Feature Extraction And Recognition Of DC Fault ARC

Posted on:2018-05-17Degree:MasterType:Thesis
Country:ChinaCandidate:X DingFull Text:PDF
GTID:2322330512477312Subject:Circuits and Systems
Abstract/Summary:PDF Full Text Request
Because of the characteristics of no periodicity and zero crossing,the DC arc fault detection is more difficult than the AC fault arc.At home and abroad,the feature extraction of DC fault arc is lack of diversification,so the purpose of this paper is to enrich the characteristics of DC fault arc.At the same time give the evaluation criteria of characteristics from the two aspects of feature extraction and recognition.The arc studied in this paper belongs to the hot cathode,gas,series,and DC current bad arcs.And collect current data when normal working and fault arc through experiment,selecting the unstable stage of arc as the basis of feature extraction.Then,17 features are extracted from the time domain,frequency domain,wavelet,spectrum analysis and chaos analysis according to the collected data.In time frequency analysis,4 features of the time domain are extracted from the angle of the change of current,3 frequency domain features are extracted from the statistical point of view,and the wavelet transform is used to extract 3 wavelet features.In spectral analysis,3 power spectrum features are extracted from the amplitude angle,and 3 high order spectral features are extracted from the angle both of amplitude and phase.In the analysis of chaos,we calculate the time delay based on autocorrelation method the embedding dimension based on false nearest neighbor method,which are used to reconstruct the phase space of the arc current,and the maximum Lyapunov exponent is extracted by the method of small data.The feature evaluation index of feature extraction is given,which is the percentage of value magnitude,and the quota value of each feature is calculated,the chaotic feature's value is the maximum.Using support vector machine(SVM)to identify the characteristics of fault arc.The nonlinear support vector machine based on Gauss kernel function is selected as the final classifier.The parameters of SVM are optimized by the grid search method and the 3-Fold cross validation method.The penalty parameter is 0.054409,and the kernel parameter is 1.We trained one classifier with the accuracy of 99.999%,and 17 comparison classifiers,each of the accuracy rate had declined,but the overall still maintained at more than 99.9%.The evaluation standard of recognition level is given,that is to say,the contribution degree to accuracy,and the quota value of each feature is calculated,the chaotic feature's value is the maximum again.The results show that the methods of feature extraction and pattern recognition are correct,the two evaluation criterions can be used to evaluate the characteristics,and the problem of detecting DC arc is a problem of chaos identification.The research in this paper enriches the characteristic types of DC arc fault,at the same time,provides two evaluation criteria which can measure the effectiveness of features.
Keywords/Search Tags:DC fault arc, feature extraction, chaos theory, identification, SVM, evaluating indicator
PDF Full Text Request
Related items