关键词: 工业控制系统, 渗透测试, 强化学习, 攻击路径, QLearning

Abstract: Aiming at the deficiencies of existing penetration testing methods, this paper proposes an optimal penetration testing path generation method that combines the characteristics of industrial control systems and reinforcement learning models. Firstly, the typical structure and security threats of the industrial control system and the basic process of the penetration test are analyzed; then the target system and the attacker are modeled based on the reinforcement learning model, and an optimal path generation method for the penetration test based on QLearning is proposed. Finally, the experimental verification is carried out with the petroleum catalytic refining system as the object. The results show that the method can comprehensively consider the differences in testers’ professional skills and target equipment, and generate the optimal path for penetration testing from multiple efficient paths, providing solutions for penetration testing of largescale industrial control systems.

Key words: industrial control system, penetration testing, reinforcement learning, attack path, QLearning