Two-Layer Defense Architecture for False Data Injection Attack in Grid CPS

The destructive problem, which is in grid cyber-physical system, is generated by false data injection attack. A two-layer defense architecture based on random forest and long short-term memory is proposed. Firstly, random forest algorithm is utilized to implement multi-label localization layer for false data injection attack, and outputs the location label of attacked node. Then, the grey wolf algorithm is adopted to optimize the long short-term memory network, achieving the state data prediction layer. After eliminating state data corresponding to the attacked node, historical state data is utilized to predict. Finally, through the simulation analysis, the effectiveness and feasibility of the proposed method are demonstrated by experiments: For the attack location layer, the detection accuracy of the proposed is about 4% higher than that of the comparison in IEEE14 system. For the state prediction layer, the root-mean-square error of the proposed is decreased by 2.43% and 1.08% in node 2, and by 3.03% and 1.66% in node 5, which is compared to the BP and LSTM, respectively.