Abstract: The tactical-level wargame simulates the dynamic process of war in a random way, which can provide a decision-making background and test environment closer to real war for military intelligent decision-making technology. A behavior decision-making framework with complementary knowledge and data is proposed for the command and control of wargame multi-agent. The framework solves the shortcomings of fixed behavior pattern and weak migration ability of traditional decision-making based on knowledge reasoning to a certain extent, and also solves the requirement of data mining algorithm for a large number of human high-quality replays of the soft decision-making algorithm based on wargame data mining. It integrates knowledge, data and learning method to form a decision-making algorithm, the macro actions are processed by the knowledge reasoning based decision-making algorithm and the mirco actions are processed by the data mining-based soft decision-making algorithm. The ability of decision-making model is improved by iterative learning with self-confrontation replay data. Based on this framework, a wargame AI is designed and realized, which has achieved good results in a national intelligent wargame competition, and reflects the characteristics of high flexibility and good generalization.