在分布式光伏大规模接入的低压配电网中，现有剩余电流装置 （residual current device，RCD）无法区分剩余电流回路中的光伏异常漏电流与触电时的触电电流，容易发生误动作，导致低压配电网的用电安全和供电可靠性存在隐患。针对该问题，提出一种基于支持向量机（support vector machine，SVM）的漏电故障识别方法和基于极致梯度提升（extreme gradient boosting，XGBoost）的触电电流检测方法。首先，基于变分模态分解（variational mode decomposition，VMD）获取不同漏电场景下剩余电流信号的分量，建立故障特征集；然后，将故障特征作为输入，建立基于麻雀搜索算法（sparrow search algorithm， SSA）优化SVM的漏电故障识别模型，对漏电故障类型进行识别；针对发生触电事故时剩余电流无法真实反映触电的情况，建立基于网格搜索与交叉验证（grid search & cross validation，GSCV）优化XGBoost的回归分析模型，实现从剩余电流中精确提取触电电流。算例结果表明：相较标准SVM和核极限学习机（kernel extreme learning machine，KELM）模型，SSA-SVM模型对漏电故障的识别率最高，平均识别准确率达99.28%；基于GSCV优化的XGBoost回归分析模型提取的触电电流值与真实值实现了良好的拟合；所做工作为具备漏电故障识别和触电电流检测功能的新型RCD开发提供了理论依据。

In low-voltage distribution networks with large-scale integration of distributed photovoltaic （PV） systems， existing residual current devices （RCDs） cannot distinguish between abnormal PV leakage currents and electric shock currents in residual current circuits， leading to frequent misoperations. This poses risks to electrical safety and power supply reliability. To solve this problem， this study proposes a leakage fault identification method based on support vector machine （SVM） and an electric shock current detection method based on extreme gradient boosting （XGBoost）. Firstly， variational mode decomposition （VMD） is used to extract components of residual current signals under different leakage scenarios， establishing a fault feature dataset. Then， using these features as input， an SVM model optimized by the sparrow search algorithm （SSA） is developed to identify leakage fault types. For cases where residual currents fail to reflect real electric shock conditions， an XGBoost regression model optimized by grid search and cross-validation （GSCV） is built to accurately extract electric shock currents from residual currents. Test results show that compared to standard SVM and kernel extreme learning machine （KELM） models， the SSA-SVM model achieves the highest leakage fault identification accuracy， with an average of 99.28%. The GSCV-XGBoost model accurately fits the extracted electric shock currents to real values. This work provides a theoretical basis for developing new RCDs with leakage fault identification and electric shock current detection capabilities.