Secure Transmission Scheme for Smart Grid CPS with Digital Twin Architecture Based on Multi-Agent DRL

The CPS technology based on the digital twin provides a dynamic and accurate digital model and real-time control of equipment for the construction of the smart grid. However, integrating physical and information sides makes the power grid CPS system vulnerable to attacks from the network and physical sensors, which brings new challenges to communication security. From the perspective of collaborative, secure communication, this paper proposes a secure transmission scheme for power grid CPS based on physical layer security and deep reinforcement learning. Grid transmission nodes select idle trusted devices as relay transmission nodes. Signals are transmitted between nodes through narrow directional beams, which alleviates blocking effects and prevents eavesdroppers from obtaining good eavesdropping capacity. We use the multi-agent dual-dueling deep Q-network algorithm (D3QN) to optimize the selection of cooperative nodes, maximizing the total secrecy ability as the optimization objective. The simulation experiment proves that the proposed scheme can effectively improve the secrecy ability of CPS nodes.