Abstract: The process of DDoS attack-defense game in new situation is different now,so the payoff value cannot be quantified effectively and the game strategy cannot be adjusted dynamically to maximize the payoff using existing methods.In response to this problem,a DDoS attack-defense game model based on Q-learning was designed,and at the same time an algorithm was proposed on the basis of the model.Firstly,the payoff of the attacker and defender was calculated with the network entropy quantitative assessment method.Secondly,the single DDoS attack stage was studied using matrix game method.Finally,the model algorithm was proposed by introducing the Q-learning method into the game process,with which the strategies are adjusted dynamically according to the learning outcomes to maximize the payoff.The result of verification testing shows that the defender can achieve a higher payoff when adopting the model algorithm,thus the algorithm turns out to be practicable and effective.

Key words: DDoS attack-defense,Matrix game,Q-learning,Network entropy,Nash equilibrium