Research on Heterogeneous Network Coupling Edge Addition Strategy based on Data-driven Method

Cyber-Physical Networks, such as coupled powercommunication and transportation systems are highly vulnerable to cascading failures. Existing reinforcement strategies either critical nodes protection or edges addition, neglecting loadcapacity limits and spatial costs. We propose an edges addition strategy based on mGNN-PPO, which embeds a multi-layer Graph Neural Network within Proximal Policy Optimization to learn coupling-edge augmentation strategies under geographic distance and node capacity constraints. We utilize multi-layer GNNs to extract node and global features from multi-layer networks, taking into account node degree, load, capacity, and geographical location information of nodes. An Actor Network is trained to predict constrained edge addition actions and a Critic Network is trained to evaluate state values. Experiments were conducted on two heterogeneously coupled networks with 1500 nodes. Experiment results show that compared to existing strategies, the proposed strategy effectively enhances the robustness of the network under random attacks and reduces the cost of adding edges.