<p>Under the development trend of smart grid, there is a higher requirement for real-time monitoring and rapid traceability of measurement errors of mutual inductor and watt-hour meter. In this article, a set of online monitoring and optimization system of metering error based on edge computing architecture is conceived and implemented, aiming at enhancing the perception ability and abnormal response efficiency of power metering equipment. The system uses three-tier architecture design, integrating data acquisition, edge processing and cloud collaborative analysis mechanism. Furthermore, machine learning algorithms such as support vector machine (SVM), multilayer perceptron (MLP), and long short-term memory network (LSTM) are introduced for error identification and trend prediction, and error traceability among multiple devices is achieved by Bayesian network (BN). The experimental results show that in the classification model, when the sample size is 10,000, the recognition accuracy of MLP reaches 95.7%; In the 24-h error trend prediction of LSTM, the RMSE of the test set is 0.125. In the multi-device scenario, the traceability accuracy of BN can reach up to 96.2%. The research results confirm the stability and practicability of the system under various working conditions, and it has good popularization value. The method proposed in this article provides a new technical path for intelligent operation and maintenance of power metering system.</p>