Abstract: Q learning requires each state-action transform be visited infinitely, which limits its application when comes to large state-action space. This paper puts forward a black-board-model based multiagents cooperation learning algorithm. Agents cooperate and coordinate by a bull function which is defined in state-action space. By this bull function, agents can find those effective update more quickly and thus avoid those useless updates. Simulation proves the method can speed up the learning process at lower cost.

Key words: multiagents system, cooperation learning, black-board model

摘要： Q学习算法要求智能体无限遍历每个状态-动作转换，因此在涉及状态-动作空间非常大的应用问题时，导致收敛速度非常慢。借助多智能体的合作学习，智能体之间基于黑板模型的方法通过开关函数相互协调合作，可以更快地定位那些有效的状态-动作转换，避免了无效的更新，从而以较小的学习代价加快了Q表的收敛速度。

关键词: 多智能体系统, 合作学习, 黑板模型

CLC Number: 

• TP391