Spare parts support plays an import role during wartime. In order to meet the requirements of Precision Support, spare parts support must be planned deliberatively prewar and be executed flexibly to deal with various uncertainties. Based on the similarity between the wartime spares support system and the multi agent system, Agent based modeling and simulation methods are adopted to investigate the dynamic coordinate mechanism during the wartime. Groups in the multi-agent system's structure are described on the bases of the relationship between the Agents. To decide how to supply spares dynamically during the wartime, the new multi agent reinforcement learning method is designed and presented. A simulation example is illustrated in the end and the simulation result shows that the method is effective.