Intelligent Path Selection Algorithm for Tactical Communication Networks Enhanced by Link State Awareness

Yigang ShenXie LeiMing Li^*

• China Telecom Corporation Limited Zhejiang Branch, Hangzhou, China

In tactical communication networks, highly dynamic topologies and frequent data exchanges lead to complex spatiotemporal dependencies in link states. However, existing intelligent routing algorithms often adopt simplistic model architectures and fail to account for the spatiotemporal correlations of link state information, resulting in limited situational awareness and poor adaptability to dynamic network conditions. To address these limitations, an intelligent path selection algorithm—deep reinforcement learning with spatiotemporal-aware link state guidance algorithm (DRLSGA)—is proposed. The algorithm leverages the proximal policy optimization (PPO) framework to construct an intelligent decision-making model, and incorporates a link state feature extraction module. This module employs gated recurrent unit (GRU) and graph attention network (GAT) to capture both long-term temporal dependencies and spatial structural characteristics of sequential link state data, thereby enhancing the model's perception and decision-making capabilities. Furthermore, an attention mechanism is introduced to identify salient features within link state sequences. An optimal routing strategy is then derived through a deep reinforcement learning-based training mechanism. Experiments analysis illustrates that, compared to the existing DRL-ST algorithm, DRLSGA achieves at least a 2.07% reduction in average end-to-end latency and a 1.65% decrease in packet loss rate, while improving average throughput by up to 2.59% under high-traffic conditions. Additionally, the proposed algorithm exhibits superior adaptability to dynamic network topologies.