Intelligent Identification of Electric Meter Failures Based on Convolutional Neural Networks and LSTM

Accurate detection of electric meter failures is of great significance for the efficient operation and troubleshooting of power grids. In response to the high-dimensional, temporal, and diverse fault characteristics of electric meter data, a joint detection model based on Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks is proposed. This model utilizes CNN to extract local spatial features and combines LSTM to capture temporal dependencies, thereby achieving efficient classification of complex fault patterns. Experimental results demonstrate that the proposed model achieves a classification accuracy of 91.7 % on a real electric meter failure dataset, significantly outperforming comparison models such as Support Vector Machines (SVM), Random Forests (RF), and CNN-GRU. The model exhibits high robustness in key indicators such as fault detection rate (93.2%) and false alarm rate (5.2 %). Meanwhile, it maintains strong real-time performance during large-scale data processing, with an average inference time of 0.97 milliseconds per sample and a system throughput of 1030 samples per second. In summary, the CNN-LSTM joint model, leveraging its efficient feature extraction and temporal modeling capabilities for complex characteristics, provides an effective solution for intelligent detection of electric meter failures and demonstrates significant potential for engineering applications.