Abnormal Data Detection of IoT Intelligent Energy Meter Based on D-S Evidence Theory

In IoT smart energy meters, abnormal data features can only be extracted through current changes, resulting in low detection accuracy. Therefore, a method for detecting abnormal data in IoT smart energy meters based on D-S evidence theory is proposed. Based on the D-S evidence theory, massive, complex, and uncertain intelligent energy meter data is processed, constructed in a non empty finite field, and a reliability allocation function is defined to quantify the degree of support of evidence for operational status, achieving data noise filtering. Input the data samples into the twin encoder network to obtain low dimensional latent variables, and extract abnormal data features by calculating current, voltage change rate, etc. Use features to set local reachable density and local anomaly factor for each data point. When the local anomaly factor is greater than 1, it is considered an abnormal data point, and when it is less than 1, it is not. The experimental results show that the design method has a maximum recall rate of 0.92 and a minimum recall rate of 0.85 for anomaly detection, with an accuracy rate of over 98% in detecting abnormal voltage fluctuations, demonstrating its higher accuracy.