Anomaly Detection System of Power Consumption Data Based on Social Information Verification

Power consumption data has been widely used in many fields. The prediction and analysis of power system data such as power supply and power consumption in power plants plays a significant role in the rational layout and control of energy facilities and improving energy utilization. However, in real application scenarios, due to various reasons such as sensor damage, data transmission failure, memory damage, etc., some paragraphs of power consumption data may have persistent data anomalies. Therefore, how to effectively detect power data anomalies has become a practical and urgent need. In this paper, a spatiotemporal attention-based method SIDM is proposed to detect power consumption data with paragraph anomalies. The Transformer model has excellent performance in capturing the dependencies between variables, and researchers generally use its attention mechanism to capture the mutual influence of data at different times. The SIDM model adds additional spatial attention to its temporal attention to capture the impact of social information on power consumption. By incorporating social information factors such as similar users, related power-consuming equipment, and local climate into the model, verify whether the current data is abnormal. The experimental results on the actual collected power consumption data set show that the effect of the model proposed in this paper is significantly improved compared to the direct prediction without using social information, and compared with other social information fusion detection methods, it has a higher accuracyrate.