The electricity consumption collected by a power user electric energy data acquisition system in the electricity market environment becomes an important basis for market settlement. Real-time identification of abnormal power data can not only improve the quality of power data, but also provide a reference for detection of faults in the collection terminal and identification of abnormal power consumption behavior. In order to solve the problems of existing anomaly recognition methods, a method based on probability prediction is proposed. First, on the basis of analyzing the types and characteristics of electric power anomalies, a detection method for offline training probability prediction models and online identification of abnormal data is proposed. Then, a structured power consumption model based on the state space model is proposed to model the user power consumption rules. The Variational Bayesian Inference is used to train the model in order to realize the probability prediction of power consumption. Finally, it uses the prediction standard score to measure the difference between the measured electricity data and the electricity probability prediction result, so as to realize the online identification of abnormal data. Real electricity data are used for verification and comparison with other methods to verify the practicability and effectiveness of this method. This work is supported by the Science and Technology Project of State Grid Corporation of China (No. 1100-201919158A-0-0-00).