Optimization of Airline Scheduling Using Reinforcement Learning with PPO Algorithm

The Airline Scheduling Problem (ASP) has significant economic and operational value in air trans portation management. However, its complexity and dynamics make traditional mixed integer linear programming (MILP) and heuristic algorithms face challenges such as computational complexity and insufficient adaptability in large-scale and real-time scenarios. In recent years, reinforcement learning (RL) methods, especially the Proximal Policy Optimization (PPO) algorithm, have attracted attention due to their adaptive ability and convergence efficiency. However, how to effectively apply PPO to ASP still requires solving problems such as state modeling, feature expression and convergence stability. This paper applies and validates a set of reinforcement learning frameworks based on PPO, which achieves efficient solution of ASP through domain-specific state modeling and feature normalization. The experimental results show that this method is significantly superior to baseline algorithms such as Q-learning and standard PPO in terms of convergence speed, solution stability and scheduling quality, and demonstrates the potential to compete with traditional operations research methods. The research verified the practical application value of reinforcement learning in the field of complex dynamic scheduling, providing strong support for intelligent decision-making in scenarios such as air transportation and emergency logistics.