A Reinforcement Learning-Based Detection Method for False Data Injection Attack in Distributed Smart Grid

Abstract

False data injection attack(FDIA) is a traditional attack for the smart grid. There are many methods for the detection of the FDIA, but few of them can send the attack alarm successfully without an attack model. In this paper, we propose a reinforcement learning-based FDIA detection method for the distributed smart grid. The detection problem is formulated as a partially observable Markov decision process(POMDP) problem, and the observation of the POMDP can be obtained from the estimation of state and attack which come from the Kalman filter. By using the Sarsa algorithm, we can get a Q-table through online training. Finally, we use the IEEE-118 bus power system to evaluate the performance of our detector, and numerical results show the accurate response for the FDIA.