Nowadays, the backwardness of the power automation management in our country causes the loss of a lot of energy. In order to improve the situation, an anti-stealing mathematical model is introduced in this paper. Firstly, ten factors are selected to build the indictor evaluation system, data mining is used to process lots of the electricity data. Then, a mathematical model based on BP neural network is built for analyzing the customer consumption behavior. With the model, the suspicion coefficient of electricity stealing can be calculated, and the credit rating of power consumer is also classified. In this paper, some typical companies are selected to verify the electricity anti-stealing model, and come to a conclusion that it provides a feasible idea for the electricity stealing problem.