Multi-agent Attention Allocation Method for Optimizing Spectrum Resources of High-Speed Railway Vehicle-to-Ground Communication in Millimeter Wave

Millimeter-wave vehicle-to-ground communication is a key technology for high-capacity and low-latency services in intelligent high-speed railway systems, but high mobility causes time-varying channels, Doppler shifts, path loss, and frequent handovers. To improve spectrum utilization and anti-interference capability, this study proposes a multi-agent attention allocation method for spectrum-resource optimization in millimeter-wave high-speed railway communication. A vehicle-to-ground communication model is constructed with ground millimeter-wave base stations, high-speed train terminals, service links, and differentiated traffic requirements. The spectrum allocation problem is formulated as a nonconvex mixed-integer programming problem that maximizes throughput and spectrum efficiency while minimizing delay under power, channel, and service-priority constraints. Each communication link is modeled as an agent responsible for spectrum sensing and resource allocation, and an attention mechanism dynamically assigns cooperation weights according to channel quality, service priority, and resource demand. A PPO-based reinforcement learning algorithm solves the dynamic allocation problem. Simulations at 28 GHz and 350 km/h show that the proposed method increases throughput to 28.6 Gbps, reduces average delay to 3.2 ms, improves spectrum utilization to 82.3%, and maintains stronger anti-interference performance than game-theory, single-agent, and fixed-weight multi-agent baselines. The method is directly relevant to electromagnetic-wave propagation, millimeter-wave networking, and railway communication optimization.