An Enhanced Adversarial Attacks Method on Power System Based on Model Extraction Algorithm

Artificial intelligence algorithms fit connections between features and problems from a data-driven perspective. Artificial intelligence algorithms perform iterative training of data through neural network, which provides a new thinking dimension for researchers. The researches on the power grid field are no longer limited to modeling and analysis through traditional physical mechanism methods. However, there are security risks on the neural network model established by artificial intelligence algorithms. Attackers apply model extraction attacks to structure a substitute model of target power system model, which supports other attack algorithms to attack the power system and ultimately affects the normal operation of power system. This paper proposes an enhancement method for adversarial attacks on power system based on model extraction algorithm and tests the promoting effect of adversarial sample attack in various scenarios.