In the context of advanced smart grids and metering infrastructures, the task of analyzing and managing power metering data has gained paramount importance. Confronted with the complexities of numerous device interactions and data exchanges, the effective and accurate identification of anomalies is crucial. Traditional methods, often limited by manual threshold settings or statistical techniques, fall short in addressing the intricacy of large-scale data. Addressing this gap, our paper introduces an innovative autoencoder-based model for anomaly detection in gateway energy metering data. Our approach was rigorously tested on a real-world dataset, targeting anomaly detection in gateway power metering equipment. The results underscore the effectiveness of our model, which achieved a high accuracy of 0.93, a notable recall rate of 0.88, and an impressive F1 score of 0.91. These metrics not only demonstrate the robustness and reliability of our model in handling complex anomaly detection tasks but also mark a significant advancement in smart grid data analysis techniques.