多Agent协作的强化学习模型和算法

Reinforcement Learning Model and Algorithm Based on Multi-agent Cooperation

结合强化学习技术讨论了多Agent协作学习的过程,构造了一个新的多Agent协作学习模型。在这个模型的基础上,提出一个多Agent协作学习算法。算法充分考虑了多Agent共同学习的特点,使得Agent基于对动作长期利益的估计来预测其动作策略,并做出相应的决策,进而达成最优的联合动作策略。最后,通过对猎人-猎物追逐问题的仿真试验验证了该算法的收敛性,表明这种学习算法是一种高效、快速的学习方法。

The multi-agent cooperative learning process based on Reinforcement Learning is addressed and a new multiagent cooperative learning model is proposed. Based on this model, a cooperative learning algorithm is introduced. This algorithm pays fully attention to multl-agent cooperative learning together simultaneity, so it can make each agent predict its action policy based on the estimation on its action＇s long-time reward, At last relevant decisions to be the best associated action policy is made. We conduct a series of empirical evaluation of the algorithm on the hunter-prey problem to validate its astringency. The result shows this algorithm is an efficient and fast method for multi-agent learning.