In view of the frequent occurrence of electricity arrears, how to use scientific methods and technical means to predict the arrears risk of power customers and reduce business risk is an urgent problem for the Power Grid Corp. Taking high-voltage customers from a certain areaas an example, this paper analyzes the factors affecting the recovery of electricity tariff, and extracts therisk features from multiple scales such as arrears frequency and duration, and then establishes a model based on logistic regression algorithm to predict the risk of user arrears. The model′s evaluation results show that the indicators such as the prediction accuracy, precision and recall rate are still relatively accurate when user information is not comprehensive enough. The model issensitive to the identification of risk users, and can be used to guide the power grid corp to formulate arrears risk management policiesand improving their management capabilities.