Abstract:

Wireless sensor network (WSN) is extensive concerned by academia and industry because of its good performances such as flexible deployment and low cost. But the nodes of wireless sensor network have the great limitation in the aspect of energy, computation, memory size and bandwidth, the complex routing protocols of traditional network can't be applied directly in wireless sensor network, thus a simple and efficient routing protocol became the research focus of wireless sensor network. In order to extend working hours, this paper proposes a routing protocol, Q-WRP, which can balance the wireless sensor network lifetime on the basis of reinforcement learning. The protocol takes account of the factors of residual energy, hop count to sink node, and propagation delay time, allocates Q-value for each node, and finds the optimal routing path according the Q-values of each node at last. Simulation result from NS2 shows that Q-WRP extends the occurrence time of the node that dies firstly, and can balance the wireless sensor network lifetime efficiently.

Key words: wireless sensor network, routing protocol, reinforcement learning