Research On Feature Extraction And Recognition Method Of Series Arc Fault In Mine Electrical Connector

Electrical connectors are often used to realize electrical connection between cables and cables and between cables and electrical equipment in mine power supply and distribution system.The loose fault of electrical connector occurs frequently due to adverse mine environment such as moisture,narrow space and frequency landing operation of electromechanical device.In loose fault state,the contact resistance of the electrical connector increases,and the contact temperature rises,then spark discharge occurs and evolves into series arc fault.It will cause electric power interruption and even lead to electric fire and other safety accident.In series arc fault state,the impedance increases and current decreases because the time-varying arc resistanceis in series with load.The existing over-current protection device can'tprotect series arc fault.Therefore,it is significant to study the detection method of series arc fault.A three-phase series arc fault test platform was developed and lots of series arc fault tests were carried out with three-phase motor load and inverter load.Take A-phase current signal for an example,the current waveform features were analyzed.When a series fault arc occurs in any phase of the circuit,the A-phase current waveform has arc fault characteristics.And the phase of the occurrence of the arc fault characteristic is related to the line where the fault arc occurs.A kind of arc fault detection and line selection method based on A-phase current signal was proposed.The main work of the paper is as follows:(1)A two-stage block singular value decomposition(TBSVD)algorithm was proposed.When arcfault occurred in different circuit lines,the main difference of A-phase current is the amplitude and phase of fault features.It is difficult to distinguish arc fault occurred in different circuit lines by using global features of the matrix extraxted with conventional singular value decomposition(SVD)algorithm.The constructed feature matrix was divided intoseveral small blocks by using TBSVD.And the singular value of each small block was calculated and put into the singular value matrix sequentially.Then the singular value of the singular value matrix was obtained and used as arc fault feature.TBSVD can better describe the local features of the matrix,and the obtained feature vector has fewer dimensions.So it is convenient to identify arc fault.(2)The arc fault feature was extracted by using TBSVD algorithm.It is crucial to construct arc fault feature matrix with one-dimensional circuit signal.The A-phase current signal was processed by using wavelet threshold de-noising,piecewise aggregate approximation(PAA)and first order differential preprocessing method.Then the arc fault feature matrix was constructed by using four kinds of method.There are improved attractor track matrix(IATM),fractional Fourier transform(FRFT),Stransform(ST),Wigner-Ville distribution(WVD)respectively.And the arc fault feature was extracted by using TBSVD algorithm.(3)The parameters of support vector machine(SVM)were optimized by using particle swarm optimization combined with grid search(PSO-GS)algorithm.The PSO-GS algorithm combines the advantages of both PSO and GS.In the early stage,the location near the optimal solution was found by using PSO,and the optimization interval of GS was determined according to the calculation results of PSO.The global optimal solution was finally obtained by searching within the optimization interval.(4)The accuracy of arc fault detection and line selection was tested by using optimized SVM fault identification model and four kinds of fault feature extraction methods.There are IATM-TBSVD,FRFT-TBSVD,ST-TBSVD and WVD-TBSVD respectively.The accuracy of arc fault detection is higher than 85%for each method.However,the accuracy of line selection is higher than 85%only for FRFT-TBSVD method.The FRFT contains both time-domain and frequency-domain features,so it is better than other three kinds of method.(5)The singlephase multi-load circuit series arc fault experiments were carried out with six kinds of loads,and the main circuit current waveform was analyzed.When arc fault occurred in one branch,the characteristic of the arc fault was weakened due to the influence of the normal working current of other branches.The arc fault characteristics was extracted by analyzing the current pretreatment signal of main circuit with the FRFT-TBSVD algorithm.And the accuracy of arc fault detection and line selection was tested by usingoptimized SVM fault identification model.The accuracy of botharc fault detection and line selectionis higher than90%in singlephase multi-load circuits.So the validity of the algorithm in singlephase multi-load circuits was verified.