The invention discloses a power risk customer screening method based on a machine learning model, and belongs to the technical field of power. At present, in power customer management, customers withabnormal electricity consumption are always important concerns by enterprises, but the power enterprises always lack effective means to identify the customers with abnormal electricity consumption, and in the past, the efficiency is low and the customers with abnormal electricity consumption cannot be directly screened from mass customers by mainly observing the electricity consumption load curveof a single customer. The method comprises the following steps: step 1, acquiring electricity consumption data of a client; 2, designing a power outlier user identification index system; step 3, establishing a power outlier user identification model based on a local abnormal factor algorithm LOF; and step 4, applying the model to realize automatic screening of the power outlier customers. Based ona machine learning algorithm, abnormal electricity consumption customers can be automatically identified, safe operation of a power grid is guaranteed, economic loss of the power grid is avoided, andthe power grid management level is improved.