Abstract: Game theory, an important part of artificial intelligent technique, has been applied on network defense very well. Static model has been used widely in most of the previous studies. However, some work shows such model cannot follow the evolving of the strategies of attackers. In this paper, an active defense model based on dynamic game theory of non-cooperative and complete information has been given, that is, the attack-defense game tree has been generated by adding some virtual nodes on the original attack-defense graph. Based on the attack-defense game tree, the best defense strategies are achieved under current network environment through resolving the Nash equilibrium in different situations. Besides, for the scenarios with complete information and incomplete information, two algorithms have been proposed respectively. The analysis and experimental results show that the complexity of the algorithms can be guaranteed not worse than other similar works. Moreover, not only for scenario with complete information, but also in incomplete cases, the sensible results can be found. With the comparison of mixed strategy Nash equilibrium generated by static game and described in a probabilistic form, results given by the sub-game perfect Nash equilibrium are more easily to be understood and operated. Network research based on game theory should have a good application in the future network security product.