Deep Reinforcement Learning for Smart Grid Protection Against Coordinated Multistage Transmission Line Attacks

With the increase of connectivity in power grid, a cascading failure may be triggered by the failure of a transmission line, which can lead to substantial economic losses and serious negative social impacts. Therefore, it is very important to identify the critical lines under various types of attacks that may initiate a cascading failure and deploy defense resources to protect them. Since coordinated multistage line attacks can lead to larger negative impacts compared with a single-stage attack or a multistage attack without coordination, this paper intends to identify the critical lines under coordinated multistage attacks that may initiate a cascading failure and deploy limited defense resources optimally. To this end, we first formulate a total generation loss maximization problem with the consideration of multiple attackers and multiple stages. Due to the large size of solution space, it is very challenging to solve the formulated problem. To overcome the challenge, we reformulate the problem as a Markov game and design its components, e.g., state, action, and reward. Next, we propose a scalable algorithm to solve the Markov game based on multi-agent deep reinforcement learning and prioritized experience replay, which can determine the optimal attacking line sequences. Then, we design a defense strategy to decide the optimal defense line set. Extensive simulation results show the effectiveness of the proposed algorithm and the designed defense strategy.