The purpose of detecting abnormal power consumption is to strike abnormal power consumption and reduce non-technical loss of power. In this paper, an method based on logistic regression algorithm is proposed to identify abnormal power consumption, which includes feature extraction, algorithm construction and test model. Firstly, the related data such as power consumption of power grid are extracted, and the index of power decline trend, line loss index and warning index are extracted from the data set as the evaluation system of abnormal power consumption. Then, the trend index, line loss index and alarm index are normalized and used to calculate the outlier proximity so that part normal users can be distinguished. And then, the rest data will be used to calculate by logic regression. Compared with the logic regression algorithm, the differentiation of these abnormal users is more obvious by using the combine algorithm.