电工技术学报2025，Vol.40Issue(11)：3476-3485,10.DOI:10.19595/j.cnki.1000-6753.tces.241626
基于故障动态特性与轻量级梯度提升机的低压零火触电检测方法
Detection Method of Low-Voltage Neutral-Live Electric Shock Based on Fault Dynamic Characteristics and Light Gradient Boosting Machine
摘要
Abstract
Current electric shock detection methods are primarily designed to address faults between the live wire and the ground wire,mainly relying on monitoring changes in residual current to identify issues.However,in the case of a neutral-to-live electric shock fault,the fault circuit often does not cause a significant change in the residual current.This presents a considerable challenge for existing detection methods when it comes to identifying neutral-to-live electric shock incidents. To address the aforementioned issues,a low-voltage neutral-to-live electric shock faults detection method based on dynamic fault characteristics and a light gradient boosting machine has been proposed.Firstly,a 1:1 prototype experimental platform for a low-voltage distribution network was established in a real system.Under various operating scenarios involving multiple household loads,experiments reproducing live neutral shock faults were conducted alongside control experiments using a sliding resistor to replace the electrically shocked body.A substantial amount of experimental samples representing both fault and normal operating states was collected,creating a comprehensive database.Secondly,the complexity of neutral-to-live electric shock faults is assessed based on the interference of load current on fault current.A fault circuit electrical equivalent model is established by considering the dynamic resistance and breakdown arcs at the dual contact points of the neutral-to-live shock,in conjunction with biological dynamic impedance.The impact of fault current on the main circuit current is analyzed.Finally,features of the main circuit current are extracted from the perspective of magnitude and high-frequency components,and the temporal changes of individual features before and after the occurrence of faults are compared.Given the difficulty in clearly distinguishing between fault and non-fault states based on individual features alone,along with the fact that these features exhibit varying sensitivity to both states,a multidimensional representation of the system state is employed.Following an ensemble computational approach,a lightweight gradient boosting machine model is developed,leveraging its uni-directional gradient sampling method and ensemble operation mechanism to accurately classify the two states. The proposed method was evaluated on a test dataset consisting of 50 666 samples,achieving an overall accuracy of 96.82%.Specifically,the identification accuracy for 35 831 normal samples was 97.50%,while the accuracy for 14 835 neutral-to-live electric shock faults was 95.17%.The test results indicated that the proposed method could accurately distinguish neutral-to-live electric shock faults from normal operating conditions,including those in the control group with the sliding rheostat added,even when the fault information was significantly obscured by high load currents.Compared to existing methods,the proposed approach shows an advantage in accurately detecting low-voltage neutral-to-live electric shock faults. The following conclusions can be drawn from the analysis:(1)By incorporating the time-varying impedance of biological tissues,variations in contact resistance,and breakdown arcs,the dynamic characteristics of faults were examined,revealing two effects of neutral-to-live electric shock faults on the main circuit current:changes in current magnitude and variations in high-frequency components.These findings served as the basis for constructing feature vectors.(2)The contribution of individual features to distinguishing between neutral-to-live electric shock faults and normal operating conditions is limited,resulting in significant inter-class ambiguity that can easily disrupt the sample fitting performance of traditional pattern recognition models.However,if features can exhibit a certain degree of sensitivity across different classes,the combination of multidimensional features can facilitate comprehensive discrimination.(3)Due to its inherent resilience to disturbances,the ensemble model can effectively mitigate interference caused by inter-class ambiguity and demonstrate strong generalization capabilities.关键词
零火触电/低压配电网/轻量级梯度提升机/主线路电流Key words
Neutral-live electric shock/low-voltage distribution network/light gradient boosting machine/main line current分类
信息技术与安全科学引用本文复制引用
仝海昕,曾祥君,喻锟,周泽华..基于故障动态特性与轻量级梯度提升机的低压零火触电检测方法[J].电工技术学报,2025,40(11):3476-3485,10.基金项目
国家自然科学基金重点项目(52037001)和湖南省研究生科研创新项目(CX20220858)资助. （52037001）
