False Data Injection Attack Detection Method Based on Long Time Series Prediction

False Data Injection Attack (FDIA) is a typical network attack in power systems, which interferes with the state estimation (SE) process by manipulating power data to influence decision analysis in power systems, thereby affecting the normal operation of the Smart Grid. This paper presents a power FDIA detection method based on long time-series prediction. The method employs an improved Informer model built upon the Transformer architecture, optimizing the model structure and introducing novel attention mechanisms to enhance computational efficiency, speeding up model training and data prediction. Simulation experiments on the IEEE-14 node system are conducted, comparing the proposed method with detection methods utilizing other deep learning algorithms such as Transformer. The results validate the effectiveness of the proposed approach, accurately detecting tampered attack data and preventing losses caused by erroneous state estimation in power systems.