The verification of energy meter is carried out in labs where the experimental conditions are controlled. However, there are many differences between the actual operation environment and the laboratory environment of energy meter. At present, the research of how to carry out the verification of energy meter in multi-dimensional disturbance environment and value traceability is very limited. In recent years, with the progress in the field of AI and machine learning, a new effective way was provided to develop measurement work in multi-dimensional environment. Firstly, this paper analyzes the factors which affect the accuracy of the energy meter in actual operating environment, and discusses the feasibility of using multi-layer perceptions model to predict the accuracy of energy meter. Next, a prediction model of energy meter accuracy based on multi-layer perceptions neural network algorithm is proposed, using a supervised study to complete the training process. Finally, an energy meter accuracy data is used in simulation experiment to verify and validity the model, filling the gaps in the relevant areas of the domestic.