With the increasing permeability of power electronic equipment in low-voltage distribution lines, fault leakage characteristics have also changed significantly, bringing new challenges to power supply and consumption safety. Aiming at the problem of fault leakage risk identification and early warning caused by line aging and other reasons, this paper proposes a fault leakage risk identification method combining cluster analysis and a Stacked Long Short-Term Memory (LSTM) neural network algorithm. The difference between positive and negative bus currents is used as the original signal for leakage identification. The K-means++ algorithm is used to classify the original signal, and the preliminary screening of fault leakage risk data is carried out. The stacked LSTM neural network is used to identify the abnormal leakage sequence and obtain the predicted value of the leakage data. The abnormal discrimination threshold is further obtained by error calculation. The sudden increase error is calculated and used to characterize the fault leakage. The threshold comparison algorithm is used to finally obtain the statistics of the number of fault leak-age in the leakage data sequence and the time of risk occurrence. Finally, the method is verified using the actual meter measurement data. The experimental results show that the method pro-posed in this paper has a good recognition effect on the risk of fault leakage.