Datadriven abnormal electricity usage detection generally classifies or clusters user electricity usage behavior according to the set characteristic index items to identify electricity usage anomalies. Affected by the diversity of electricity usage behaviors in different sectors, there is often a problem of high false alarm rate in practical applications. Utilizing the characteristics of similar electricity usage behaviors of users in similar subindustry, this paper proposes a method that refines characteristic index items to detect abnormal electricity usage based on industryspecific electricity usage characteristics. Based on the actual load data of the users of the street lamp, firstly the load composition and electricity behavior characteristics of the users of the street lamp are analyzed. Then characteristic index items that can accurately describe the electricity consumption behavior of street lamp users is established. Thereafter, on this basis, according to the analysis of the statistical characteristics of the characteristic index items, the bimodal histogram method is used to set the judgment threshold of abnormal street lamp users. The abnormal street lamp users whose characteristic index items exceed the threshold in the actual power grid are identified. At last the effectiveness of proposed method is tested. Moreover, general law of the regional distribution of users with abnormal electricity consumption is also analyzed and summarized.