False data injection attack detection in dynamic power grid: A recurrent neural network-based method

Abstract

The smart grid greatly facilitates the transmission of power and information by integrating precise measurement technology and efficient decision support systems. However, deep integration of cyber and physical information entails multiple challenges to grid operation. False data injection attacks can directly interfere with the results of state estimation, which can cause the grid regulator to make wrong decisions and thus poses a huge threat to the stability and security of grid operation. To address this issue, we propose a detection approach against false data injection attacks for dynamic state estimation. The Kalman filter is used to dynamically estimate the state values from IEEE standard bus systems. A long short-term memory (LSTM) network is utilized to extract the sequential observations from states at multiple time steps. In addition, we transform the attack detection problem into supervised learning problem and propose a deep neural network-based detection approach to identify attacks. We evaluate the effectiveness of the proposed detection approach in multiple IEEE standard bus systems. The simulation results demonstrate that the proposed detection approach outperforms benchmarks in improving the detection accuracy of malicious attacks.