An adaptive proximal policy optimization algorithm based on mission-completion degree is proposed to address the challenge of maintaining large-separation formations for high-speed UAVs in unknown and dynamic environments during manned–unmanned cooperative operations. An observation space that depends solely on relative state information is constructed, and a clipping-coefficient tuning mechanism driven by mission completion degree is devised. A segmented trajectory-truncation method is incorporated to accelerate convergence and enhance control stability. It is demonstrated through simulations that, relative to the baseline proximal policy optimization algorithm, marked improvements in convergence efficiency and tracking accuracy are achieved. Accurate formation is preserved even when the manned aircraft performs random maneuvers or abrupt heading changes, so the effectiveness and robustness of the approach in dynamic combat scenarios are validated, and an efficient solution for manned–unmanned formation keeping is provided.