Autonomous Penetration Test Path Planning Based on OIPPO
Yan Huang, Qiang Yang, Bo Yu, Mengqi Zhang · 2024
Intelligent penetration testing combines deep reinforcement learning to plan attack paths and simulate complex attack scenarios to identify system weaknesses. Compared with traditional penetration testing, it can learn and optimize strategies independently and simulate more complex attack paths and attacker behaviors.However, the existing methods have problems such as gradient explosion, gradient disappearance and suboptimal solution in the training process. In order to solve these problems, this paper introduces proximal policy optimization (PPO) algorithm, and proposes an improved PPO algorithm which combines orthogonal initialization and policy entropy control (OIPPO).The algorithm combines the advantages of orthogonal initialization and PPO algorithm, and improves the performance and efficiency of the algorithm through policy entropy control. The experimental results show that the convergence speed of OIPPO algorithm is better than that of traditional PPO and its improved algorithms in various network scenarios, and it shows better effectiveness, scalability and compatibility when the network scale is expanded.