Abstract: The requirements to Agent learning system and the shortages of Agent rule learning system based on genetic algorithm in simulation of complex adaptive system (CAS) is analyzed, and an Agent learning system based on improved immune genetic algorithm is provided. In the algorithm, it uses the field knowledge and expert experience as vaccine to carry on immunity inoculation to the rule in the rules storehouse and the chaos variation model to make a variation on the rules to produce new rules. Finally, with Agent rule learning process in the simulation of interbasin water transfer management complex adaptive system, the algorithm proving and instance analysis is carried on Agent rule learning and evolution mechanism.

Key words: Complex adaptive system; Iimmune learning system; Simulation based on agents; Interbasin water transfer management

摘要： 分析了复杂适应系统(Complex Adaptive System，CAS)仿真中对Agent 学习系统的要求以及基于遗传算法的Agent 规则学习系统的不足，提出了一种基于改进免疫遗传算法的Agent 学习系统，在该算法中将领域知识和经验作为疫苗对规则库中的规则进行免疫接种，并利用混沌变异模型对规则进行变异操作，产生新的规则。最后，结合跨流域调水管理复杂适应系统仿真中的Agent 规则学习过程，对Agent的规则学习和演化机制进行了算法验证及实例分析。

关键词: 复杂适应系统；免疫学习系统；Agent 仿真；跨流域调水管理