在ＭＤＰ中，单Ａｇｅｎｔ可以通过强化学习来寻找问题的最优解．但在多Ａｇｅｎｔ系统中，ＭＤＰ模型不再适用．同样极小极大Ｑ算法只能解决采用零和对策模型的ＭＡＳ学习问题．文中采用非零和Ｍａｒｋｏｖ对策作为多Ａｇｅｎｔ系统学习框架，并提出元对策强化学习的学习模型和元对策Ｑ算法．理论证明元对策Ｑ算法收敛在非零和Ｍａｒｋｏｖ对策的元对策最优解． In MDP, single agent can find the optimal solution to the problem through reinforcement learning. However, in multi-agent systems, the MDP model no longer applies. The same minimum and maximum Q algorithm can only solve MAS learning problem using zero and countermeasure model. In this paper, the non-zero-sum Markov game is adopted as the learning framework of multi-agent system, and the learning model of meta-game reinforcement learning and the meta-solution Q algorithm are proposed. It is proved theoretically that Q - Metaphor Q - Convergence converges to the optimal solution to the meta - game of non - zero sum Markov games.