Abstract:With the rapid development of the aviation industry, air traffic volume continues to rise, airspace density has increased significantly, and flight separation standards have been reduced, making aircraft-following behavior increasingly prominent in high-density traffic flows. Aircraft-following modeling has become a research hotspot in the field of air traffic control. In existing models, the acceleration of the leading aircraft is rarely considered, and there is a lack of integration of stability criteria into reinforcement learning-driven aircraft operation simulations. To address these issues, this study focuses on the terminal approach phase, where aircraft operate at high density. Based on the ground vehicle-following theory, an aircraft intelligent driving model (AIDM) is developed by considering both ground-to-air communication delays and the acceleration of the leading aircraft. Stability is analyzed by the perturbation method and the direct transfer function method, and the critical stability condition is derived. The model’s local stability (between a leading and a following aircraft) and asymptotic stability (within a fleet) are examined. Finally, local stability simulations, asymptotic stability simulations, and terminal-area aircraft-following scenario simulations are conducted. The derived critical stability criterion is embedded into the reward function and combined with a classical reinforcement learning algorithm to explore autonomous aircraft operation and enhance operational stability. Simulation results show that both the local and asymptotic stability domains of the AIDM are significantly expanded. Analysis of the unstable domain shows that increasing the sensitivity coefficient to the leading aircraft’s acceleration from 0 to 0.3 increases the stability domain of the model by 8.5% compared to the original IDM model. Moreover, reducing the communication delay by 20 s increases the stability domain by 17.65%. In scenario simulations, under identical disturbances, the AIDM requires 21.9% less time to regain stability than the original IDM model. The reinforcement learning algorithm incorporating the stability-aware reward mechanism achieves significantly higher average scores than traditional reward schemes, effectively improving aircraft operational stability. These findings confirm the mechanism’s effectiveness in agent learning and its potential for supporting autonomous aircraft operations.