Abstract: In response to the problem of low detection accuracy in real-time detection of electric energy meter faults in practice, a real-time detection method for electric energy meter faults under multi-source data integration is proposed. This article introduces the graph theory minimum weight point coverage strategy to optimize the sensor deployment scheme, deploying current/voltage sensors, temperature sensors, vibration sensors, and light sensors at key nodes, and constructing a multi-source perception network to comprehensively collect operational status data of electric energy meters. Secondly, this article adopts a consistency judgment method based on probability distance to integrate and process multi-source heterogeneous data, improving the quality and reliability of data integration. On this basis, this article constructs a convolutional neural network (CNN) fault detection model, which automatically learns fault features from multi-source data to achieve real-time recognition of fault types. The experimental results show that the proposed method has an average recall rate and F1 score of over 95% for seven typical faults, demonstrating good engineering application value and real-time performance.