Abstract: The air combat formation confrontation method based on war game research mainly uses rules or operation research and other means, which has some defects, such as unreasonable hypothesis, inaccurate modeling, poor adaptability and so on. Reinforcement learning algorithm can independently learn and organize countermeasure strategies according to combat data to deal with complex battlefield conditions, but the existing reinforcement learning has high requirements for combat data. When the action space is too large, the convergence of the algorithm is slow, and has higher requirements for the simulation platform. In view of the above problems, an intelligent decision-making method for air combat formation confrontation integrating knowledge data and reinforcement learning is proposed. The input of the decision-making method is the battlefield fusion situation. The hierarchical decision-making framework is used to control the operator to select and execute the task, and the upper layer includes an action selector driven by expert knowledge. The lower layer includes the bullet avoidance action actuator, reconnaissance action actuator and strike action actuator controlled by reinforcement learning algorithm. Finally, experiments based on typical combat scenarios are carried out to verify the feasibility and practicability of the proposed method, and the experiments show that the method has the advantages of accurate modeling and efficient training, etc.