Research On Arc Fault Protection Technology Based On Pattern Recognition

With the increase in electricity consumption of residents,the number of electric fire accidents caused by arc faults has increased year after year.At present,arc fault detection devices mainly detect arc faults based on the characteristics of the current waveform.However,the types of loads in the line are continually increasing,and the current waveforms of some loads during operation are very similar to those of other loads when arc faults occur,which causes the arc fault detection device cannot fulfill the protection function correctly.To solve this problem,this paper uses pattern recognition technology to propose three arc fault identification methods based on principal component analysis and support vector machines,transfer learning with convolutional neural networks,and self-organizing feature map networks to identify series arc faults in typical load protection.Firstly,arc faults are simulated by the test setup,and the database of the arc current is established to analyze the main characteristics of the arc fault.Secondly,an arc fault identification method based on principal component analysis and support vector machine is proposed based on the time-domain characteristics of the current.The principal component analysis method is used to reduce the time-domain characteristics of the current,and the first three principal components with the highest contribution rate are selected as the input of the support vector machine.The optimal parameters of SVM are obtained by grid search method and K-fold cross validation method to improve the accuracy of arc fault identification.Thirdly,the time-frequency characteristics of current are analyzed,and an arc fault identification method based on transfer learning with convolutional neural networks is proposed.Convolutional neural networks that have completed pre-training in large data sets are transferred to the recognition of arc fault images.The optimal parameters of the convolutional neural network are obtained by simulation.The results show that the arc faults can be effectively identified using convolutional neural networks.Finally,an arc fault identification method based on self-organizing feature map network is proposed.Through the rules of competitive learning,the network can autonomously extract and learn fault features of current samples,avoiding inaccuracies and one-sidedness caused by artificially selected fault features.The final load classification results and arc fault identification results are obtained according to the detection results and their probability within each sliding window time.The results show that the proposed method can achieve rapid identification of series arc faults and has high accuracy.