To enhance the collaborative command communication performance of helicopter formations in complex low-altitude electromagnetic environments, a closed-loop optimization method targeting link dynamic characteristics is established. By modeling channel instability characteristics in typical low-altitude scenarios, the study designs a link control architecture integrating perception, prediction, decision-making, and scheduling. A QoS prediction module based on gated recurrent units (GRU) senses link evolution trends, while a multi-agent reinforcement learning algorithm jointly optimizes transmission power, routing, and scheduling strategies. A dynamic weighted scheduling mechanism covering multi-service queues is established to enhance transmission timeliness. Results demonstrate that in high-load and complex obstruction areas, the proposed method reduces average end-to-end delay by 22.3%, lowers packet loss rate by 37.5%, and increases system efficiency by 31.6%. It also exhibits lower strategy volatility, excellent generalization performance, and stability, providing methodological support for constructing low-altitude command communication systems with intelligent perception and adaptive capabilities.