Abstract: Line loss management is one of the key goals of the power supply chain, and it is also the basis for promoting quality and efficiency improvements, enhancing the economic efficiency of the power grid. The power consumption characteristics and normal line loss characteristics formed by power grid lines, stations and users have their own power attribute characteristics, which comprehensively reflects their different power consumption characteristics and line loss characteristics like a human fingerprint. Based on the massive user data collected by the electricity consumption information collection system and the daily updated meteorological temperature data, a power fingerprint rolling prediction and diagnosis model based on a fixed window vector autoregressive temperature model is proposed, which intelligently analyzes the abnormal probability of lines and stations, and improves the precise management level of the State Grid. Experiments verify the feasibility and effectiveness of the proposed method.