Multi-Dimensional Resources Management with GNN for Adaptive Routing Optimization

The primary challenges in routing optimization include adapting to dynamic environments with frequent node and link changes. Handling the computational complexity of large-scale networks and balancing communication resources in multi-objective optimization are also key difficulties. Traditional methods focus on optimizing a single dimension, which limits their effectiveness in dynamic network environments. They often fail to capture the full network state, making it difficult to adapt to changes in topology or traffic. This lack of a comprehensive view leads to poor resource balance and suboptimal performance as network conditions change. To address these challenges, we propose an Adaptive Routing algorithm with joint optimization of Multi-dimensional network Resources (AR-MRs) based on graph neural networks (GNNs). This algorithm optimizes multiple network resources simultaneously and effectively tackles the issues of incomplete resource consideration and insufficient balance prevalent in existing routing methods. As a result, overall network performance and reliability are enhanced. Additionally, we innovatively design a resource-adaptive module based on GNN. By leveraging GNN, complex relationships between network links and nodes are captured. This module thoroughly analyzes network states and dynamically adjusts resource allocation. Balanced optimization across multiple resource dimensions is thereby ensured. The effectiveness of the algorithm was validated through deployment in a simulation environment. Simulation results indicate that, compared to existing solutions, this approach significantly reduces end-to-end communication delay, decreases bit error rate, and enhances packet transmission efficiency.