Abstract:

False data injection attacks pose a severe threat that cannot be overlooked during the development of new cyber-physical power systems. These attacks can tamper with power grid data to create false grid states， mislead operators into making incorrect operational decisions， and consequently disrupt the stable operation of the power system. Moreover， existing defense methods are incapable of addressing attacks involving complex data types or pinpointing abnormal states. Therefore， a multi-scenario AC false data injection attack strategy was proposed， and an attack model better aligning with actual power grid environments and exhibiting strong stealthiness was constructed. On this basis， a defense mechanism based on a multivariate detection model was designed， effectively integrating the advantages of three detectors： extreme learning machine， extreme gradient boosting， and light gradient boosting machine. Using multi-scenario attack cases as training data， an efficient attack detection model capable of pinpointing abnormal states was formed. Both the attack and defense models were simulated in IEEE 14-bus and IEEE 57-bus systems. The experimental results verified the effectiveness， stealthiness， and diversity of the attacks， as well as the real-time performance and accuracy of the detection mechanism.

Key words: cyber-physical power system, electric power system information security, false data injection attack, multivariate detection model, extreme learning machine, extreme gradient boosting, light gradient boosting machine