Distributed neurodynamic-based attack-resilient control strategy for virtual power plants with watermarking method

With the increasing proportion of distributed renewable energy in power systems, virtual power plants (VPPs) have rapidly developed as a new architecture for modern power systems. However, the distributed structure of VPPs results in geographically dispersed devices, which increases vulnerability to network attacks. To address these challenges, this study proposes several key strategies. First, a distributed neurodynamic algorithm is proposed to solve the optimal solution of the VPP model. Second, to counter the susceptibility of distributed systems to network attacks, a detection-isolation strategy is developed to defend against external threats. In the detection phase, an improved watermarking method is designed to detect replay attacks. The proposed watermarking method, tailored for distributed systems, operates without incurring additional communication overhead or requiring state estimation for the next time step. Building on the watermarking method, an observation subnetwork is further designed to detect collusion attacks. The establishment of the observation subnetwork allows nodes to mutually monitor each other, facilitating the detection of colluding nodes. In the isolation phase, to mitigate the impact of attacks on the system, an adaptive reputation-based isolation mechanism is introduced. This mechanism gradually marginalizes compromised nodes, ensuring that the system can still achieve optimal scheduling after isolating malicious nodes. Finally, simulations conducted on a nine-node VPP system validate the effectiveness of the proposed strategies.