Based the Complex Adaptive System, in this paper, we use the SWARM platform to develop a simulation system which is used to simulate the action of multi-agent group. We utilize the method of reinforcement learning and genetic algorithm to simulate the learning and adaptability ability of the people, and use the Cellular Automation to simulate people's move. We also set up a group system model, all the agents in the model have their own move rules and strategies. Though the agents affect each other and the circumstances, the group system would emerge prominent characters in the macrostructure. Final, we develop our system by using SWARM and JBuilder2005, then simulate it by the given initialized group state, and discuss and analyse the experiment result.