中文摘要：

文章研究了无线传感器网络中存在的多条最短路径路由选择问题。将无线传感器网络看作多Agent系统，采用强化学习理论，提出了一种基于多Agent强化学习的无线传感器网络多路径路由协议MRL-MPRP（Multi-agent ReinforcementLearningbasedMultiple-pathRoutingProtoc01）。该协议综合考虑了所要发送数据的优先级、节点间的链路质量以及节点数据缓冲队列的拥堵情况，为不同优先级的数据选择出当前网络状况下最优的路径进行数据的传输。仿真结果表明了该协议在降低网络平均端一端延时、提升数据包成功投递率方面的有效性。

英文摘要：

In this paper, the optimal route selection problem in the case of several shortest paths with the same hops in wireless sensor networks is considered. A multi-agent reinforcement learning based multiple-path routing protocol（MRL-MPRP） is proposed by regarding wireless sensor network as a multi-agent system. In MRL-MPRP, the sensor node takes the priority of the transmitting data, link quality and congestion of different neighbors into consideration so as to select an optimal route for sending different types of data. The simulation results show that the proposed protocol effectively re- duces the end-to-end delay and increases the packet delivery ratio.