An anomaly Detection Method for Electricity Consumption Data Based on CNN-BiLSTM-Attention

Abstract

As the complexity and uncertainty of smart distribution networks increase, data security issues in smart meters have become a pressing challenge, such as false data injection and electricity theft. To ensure fairness, safety, and overall economic efficiency of distribution networks, it is essential to accurately detect abnormal electricity consumption. However, traditional methods relying on on-site inspections by grid personnel suffer from low efficiency and high costs in detecting user anomalies. This paper proposes an electricity consumption data anomaly detection method based on CNN-BiLSTM-Attention. CNN is utilized to extract data features, while BiLSTM and attention mechanisms capture contextual information in sequence data. Furthermore, experiments conducted on data extracted from smart meters demonstrate that the proposed model outperforms other models in anomaly detection, with accuracy, recall, and F1-Score all exceeding 91%. These results validate the effectiveness and feasibility of the proposed method, providing an efficient solution for user anomaly detection in national power grids.