Abstract: The phenomenon of stealing electricity destroys the order of social power supply and consumption,and even hinders the development of new distribution system construction when it is serious. In order to identify the behavior of stealing electricity more accurately,an intelligent recognition model of stealing electricity based on the combination of extreme learning machine（ELM）and support vector machine（SVM）is proposed. By using the big data of electric energy measurement,the data status indicators of power theft users are analyzed,and the evaluation system of power theft indicators is built. The index evaluation system is used to train the intelligent identification model of stealing electricity,and the ELM-SVM prediction model is used to identify the users of stealing electricity. This method effectively integrates the advantages of ELM algorithm and SVM algorithm. The example shows that the recognition accuracy of the recognition model can reach 97.8%,which shows the feasibility of ELM-SVM combined method. The highprecision,high-efficient prediction and recognition of users’ stealing behavior are realized.