Abstract：The turns of network attack-defense are usually more than once, so it is a good idea to establish the behavior model of network attacker in finite repeated games, and it can help to predict the decision of network attacker more accurately. The behavior of network attackers is analyzed, and attackers are classified as four types according to the rational level, such as long-term myopic, long-term sophisticated, short-term myopic, and short-term sophisticated. Then based on the theories of reinforcement learning, Experience-Weighted Attraction (EWA), Quantal Response Equilibrium(QRE) and so on, four different behavior models are established, and parameter assessment of models is also finished. Finally the experiment result validates the e®ectiveness and accurateness of models.
2015, Vol. 1 
