Abstract: At present, the residual current protection devices operating online use the magnitude of the detected residual current effective value as the sole criterion for action. However, this criterion cannot accurately detect the fault leakage current signal, and there are inherent deficiencies and defects in the operating principle, often causing misoperation or refusal to operate.This article fully utilizes intelligent information processing technology and obtains a large amount of original residual current data by constructing a leakage experimental platform; Analyzed and mastered the frequency and time domain characteristics of residual current; Ten features of residual current are extracted, and a low-voltage leakage detection model based on Random forest algorithm is studied and proposed.After testing, the accuracy of the model in this article can reach 99.98%; Finally, the Confusion matrix method is used to compare with the commonly used algorithm, which further verifies the accuracy and feasibility of the low-voltage system leakage detection model based on Random forest algorithm.