Communication Messages Based Fusion Awareness and Anomalies Diagnosis for Metering Anomalies in Power System

With the penetration of source–grid–load integration and the promotion of power market, higher requirements are put forward for the reliability of power metering and timely operation and maintenance. Therefore, this paper proposes a lightweight measurement anomaly diagnosis model suitable for heterogeneous message information, which integrates the message text semantics and time characteristics, and makes full use of the operation status information in the measurement message. Firstly, TextCNN is used to extract semantic representation from message fields, and GRU-D is used for robust modeling of time series data with missing values and irregular sampling. The two modes are adaptively integrated through the gated fusion mechanism, and the BiLSTM is introduced to capture the global dependency across time steps to achieve multi class fault classification. The experimental results show that the proposed method is obviously superior to multiple baseline models on the fault data set of power metering equipment, and has better diagnostic ability for typical fault types such as communication interruption and time anomaly.