Abstract: Reinforcement learning has been widely applied in many single or multi agents systems as an important machine learning method.However，the learning quality of the reinforcement learning is greatly influenced by the used algorithms and their parameters，different algorithms or a little change of their parameters can lead to large differences in learning qualities.But it is difficult to determine the best algorithm and the optimal parameter when the model of the environment is unavailable.In order to avoid the influence of the parameters，a fused Sarsa learning system based on multi-agent is proposed to deal with the reinforcement learning environment as a multi-agent environment.And a maze experiment show the feasibility and validity of the algorithm.

Key words: reinforcement learning, multi-agent, fusion, Sarsa（λ） algorithm

摘要： 强化学习作为一种重要的机器学习方法，已经被广泛应用于许多单智能体和多智能体系统。强化学习的性能受所使用的学习算法及其参数的影响很大，不同的学习算法或者参数很小的变化都可能导致学习性能很大的变化。当环境模型未知时，确定最好的算法和最优的参数是困难的。为了避免参数的影响，提出了一种基于多Agent的融合Sarsa（λ）学习系统，它把强化学习环境当作多智能体环境来处理。最后用迷宫实验仿真，结果验证了该方法的可行性和有效性。

关键词: 强化学习, 多智能体, 融合, Sarsa（λ）算法