Software-Defined Networking-Based Resilient Proactive Routing in Smart Grids Using Graph Neural Networks and Deep Q-Networks

The enhanced functionality of the smart grid depends on the robust interconnection between its physical and cyber-layer components. Two distinct categories of control data packets exist within smart grids: fixed-scheduling (FS) and event-driven (ED). An intelligent routing strategy is required to satisfy latency requirements for FS and ED packets across various quality-of-service (QoS) levels and must be resilient to failures. Our proposed software-defined routing strategy balances requirements by dynamically adjusting decisions based on packet types. It prioritizes paths with lower latency and higher throughput for ED packets while prioritizing paths with higher redundancy and lower congestion for FS packets. This strategy switches between proactive and resilient modes based on network conditions. First, the proactive routing module (PRM) utilizes a graph neural network (GNN) and a Q-learning (QL) algorithm to fix sub-optimal routes for efficient packet delivery under normal conditions. Second, the resilient routing module (RRM) combines a deep Q-network with GNN to select optimal routes that remain viable even during failures, ensuring continued operation and robustness. Both modules update the queue service rate (QSR) using QL-agent while avoiding congestion. The GNN ensures proactive module selection based on excessive congestion violations indicating failure conditions. Given the efficient performance of the PRM in normal conditions and the resilience of the RRM under failures, the proposed strategy presents a dual-mode routing that minimizes overhead with a high level of resilience. The proposed approach, evaluated using the IEEE 39-bus test system cyber-layer, effectively ensures desired QoS routing regardless of the conditions of the cyber-layer.