DC Arc Fault Detection Method Based On Machine Learning

DC arc fault occurs frequently in the DC power supply system.Once the DC arc fault occurs,high temperature will be produced with spark spatter.If the fault is not found in time and the power is cut off,it will cause fire easily.The DC arc fault detection device can detect the DC arc fault in time and cut off the power supply,which can effectively avoid the fire.Therefore,reliable DC arc fault detection technology is of great significance for ensuring the safe operation of the DC equipment.This paper studies DC arc fault detection technology and proposes a DC arc fault detection method based on support vector machine(SVM).The DC arc fault feature extracted from the current sampling data forms a training sample.Through supervised learning,an SVM classifier with DC arc fault detection function is trained.The trained SVM classifier can predict the DC arc fault of DC power supply system based on the characteristic of DC arc fault sampled from the input current data.The paper extracts and analyzes the characteristic of DC arc fault current in different dimensions,and analyzes the current variation in the time domain,and get the time-domain characteristics of DC arc fault.The frequency domain characteristics of the DC arc fault are obtained by analyzing the change of the energy value in the frequency domain of the current by Fourier transform.Aiming at the disadvantages of Fourier transform in the analysis of non-stationary signals such as DC arc fault,the Hilbert-Huang transform is used to extract the accurate feature in time-frequency domain.The fractal dimension based on Hilbert-Huang transform is used as the time-frequency characteristic of DC arc fault detection.The DC arc fault is regarded as a standard chaotic motion.The maximum Lyapunov exponent which can characterize the degree of system chaos is extracted by Wolf method and the maximum Lyapunov exponent of current is taken as the chaotic feature of DC arc fault detection.This paper proposes a DC arc fault detection method based on SVM and its implementation process,and selects the nonlinear support vector machine based on Radial basis function as the SVM classifier.The SVM classifier is trained by constructing training samples from the eigenvalues of the DC arc fault extracted from the original current sampling data.In the training,K cross validation and heuristic genetic algorithm are used to optimize the SVM penalty parameter C and kernel parameter g.In this paper,a trained SVM classifier is built to test data classification performance(including detection accuracy and error judgement rate)by using DC arc fault characteristic value extracted from original current sampling data,and to test the effect of DC arc fault detection.The accuracy of the classifier is more than 99.63%,and the error judgement rate is only 0.096%.The experiment also compares the contribution of different characteristics to the detection accuracy.The experimental results show that the SVM-based DC arc fault detection method has achieved good performance in both detection accuracy and error judgement rate.