The development of power grids increases management demands for power information data, making health management of intelligent electricity meter critically important. This study constructs a dual-channel architecture intelligent electricity meter fault diagnosis model based on data characteristics. The model employs multi-head attention mechanisms and bidirectional gated recurrent units to extract and process features from high and low-frequency collected data. Comparing experiments between dual-channel and single-channel models demonstrates that the dual-channel fault diagnosis model achieves superior performance in accuracy, weighted F1 score, and weighted recall rate. Additionally, the model outperforms other fusion architecture models by achieving higher classification accuracy for handling 11 fault types. This research provides an effective method for intelligent electricity meter fault diagnosis and enhances health management capabilities of power system equipment.