Research On Low-voltage Arc Fault Detection Technology Based On Measurement Data Analysis

Electrical fires have always accounted for a high proportion of all types of fires.Among them,a very important factor causing electrical fires is arc fault.The characteristics of arc fault are complex and it can occur under small current situations.It avoided the protection scope of traditional electrical protection devices and buried electrical fire hazards in low-voltage power distribution lines.Therefore,in order to reduce the occurrence of electrical fire,reduce economic and property losses and prevent and control major safety risks,the research of fault arc detection technology is very necessary.Based on the voltage and current recording data collected on the arc physics experiment platform,according to the function orientation and protection scope of arc fault protection products,this paper studies the fault arc detection methods under single load circuit structure and multi-load circuit structure respectively.The main research work of the full text is as follows:(1)The experimental plan is designed considering the load quantity,load type and arc position.The arc experiment is carried out on the platform and the measurement data is collected to form the experimental data set.The current signals in various scenarios are compared and analyzed,and the time domain,frequency domain and image features are extracted to construct the fault arc feature set.Among them,time domain features include zero rest time,frequency domain feature includes current harmonic amplitude,and image feature includes V-I binary images.The statistical results show that the load type,arc location and circuit structure will have different effects on the characteristic value.(2)In the single load circuit structure,a method of fault arc detection and load identification based on harmonic characteristics and random forest is proposed,and the application simplification of the method in practice is discussed.The fault arc detection is regarded as a classification problem.Combined with the classical classification algorithm in machine learning,the detection models are trained to determine whether the scene corresponding to the measurement data contains a fault arc.There are four classification algorithms,such as K-nearest neighbor.They are combined with multiple extracted features to perform performance comparison evaluation on the same data set.The comparison of test results shows the effectiveness of the proposed detection model.With reasonable label setting,fault detection and load identification can be achieved at the same time.(3)A fault arc detection and fault location method based on composite feature and deep learning is proposed under multi load circuit structure,and the influence of circuit structure on detection accuracy is discussed.The harmonic feature and V-I binary image feature with higher accuracy in single features are selected as the basic feature quantity.The full connection neural network and convolution neural network are used for training respectively.Then the output of two network hidden layers is extracted and combined into a feature vector to construct a composite feature.Then,a new neural network is trained with composite features to construct a fault arc identification model.The comparison of detection results shows that the composite feature makes full use of the complementarity of harmonic feature and image feature.Compared with single feature,the fault detection accuracy is improved.Reasonable label setting realizes fault detection and gives the reference information of fault location.