Research on Attack Techniques for Cyber-Range Simulations based on Deep Reinforcement Learning

Key Points

• The research aims to analyze the efficacy of deep reinforcement learning algorithms in simulating cyber-attack techniques.
• Created a cyber-battle simulation environment mimicking real attack procedures.
• Applied DQN, PPO, and SAC algorithms to compare learning performance of attack agents.
• Evaluated the effectiveness of defense techniques based on derived attack patterns.
• DQN and SAC produced an optimal sequence of 9 targets, matching the goal.
• PPO generated over 30 attack sequences, demonstrating 3.3 times lower optimization efficiency.
• Provided quantitative comparisons of reinforcement learning method performances for future attack pattern analysis and defense validation.