Risk assessment of cyber-physical distribution networks based on graph convolutional networks

The rise of distributed renewable energy has increased the complexity of grid, the vulnerabilities and security issues of Cyber-Physical Distribution Systems (CPDS), where physical and information systems are deeply integrated, have also become more prevalent. There is an urgent need for an effective risk prediction and assessment methodology for these systems This paper introduces a framework to evaluate the risk of resource disconnection from cyber-attacks, considering their effects on physical equipment. Utilizing Graph Convolutional Networks (GCNs) and Long Short-Term Memory (LSTM) networks, the framework models network topological structure and extracts system electrical characteristics, respectively. By employing a neural netw ork trained on offline simulation data, the framework decreases real-time computational demands for assessing risks on the physical side. The IEEE-33 bus system and IEEE-118 bus system are used for validation, demonstrating the model's accuracy and efficiency in providing reliable assessment outcomes. This approach offers a significant step forward in enhancing the security and resilience of CPDS against cyber threats.