Channel Big Data Drives Deep Reinforcement Learning for 6G Air-Space-Ground Network Resource Scheduling

6G integrated air-space-ground networks, incorporating multiple nodes such as satellites, drones, and ground base stations, face challenges including heterogeneous network coordination, dynamic and time-varying channels, and resource supply-demand imbalances. Traditional resource scheduling methods are ill-suited to these complex scenarios. Existing scheduling strategies lack the ability to deeply mine channel big data, making it difficult to accurately capture the evolution patterns of channel states. Furthermore, deep reinforcement learning models suffer from slow convergence and poor generalization in multi-objective optimization. This paper is structured as follows: First, it constructs a channel big data acquisition and preprocessing system to extract core features such as channel quality, user needs, and network topology. Second, it designs a deep reinforcement learning scheduling model that integrates an attention mechanism, introducing a multi-objective optimization function to balance throughput, latency, and energy consumption. Finally, it verifies the effectiveness of the model through a simulation platform. Experimental results show that the proposed DRL algorithm outperforms traditional algorithms in all indicators, with a resource utilization rate of 89%, which is 22 percentage points higher than the RR algorithm and 13 percentage points higher than the Greedy algorithm. This provides technical support for efficient resource scheduling in 6G air-space-ground networks.