Reinforcement Learning-based Routing Mechanism For Multi-sink Wireless Sensor Networks

In WSNs, sensor nodes have the constrained energy, computer skills and communication ability. It is important and difficult to design for WSNs routing protocol that how to cooperate and play its role in the overall between sensor nodes and how to prolong network lifetime. In this paper, routing algorithms of WSNs is as the main study object. This dissertation analyze some typical WSN routing algorithm home and abroad, discuss the advantages of multi-sink WSNs in dispersing the network energy consuming, and propose two kinds of Reinforcement Learning-based multi-sink routing mechanisms.It can be seen from the existed WSN routing algorithms that they always consider sectional elements, such as hops, distance, location, et al. In this paper, Q-learning is exploited, remaining energy, communication overhead, distance and hops are considered synthetically, the sensor node with the most reward will be selected as the next hop. It is named as Q-learning-based Routing Protocol for Multi-sink WSNs. In addition, another mechanism called Temporal Differences-based LEACH Protocol for Multi-sink WSNs is proposed, aiming at resolving the problem of consuming energy too largely, node failure too early due to random property of LEACH while selecting cluster heads. The mechanism increase largely the probability of selecting the sensor node with more remaining energy as cluster head, make the cluster head choice without random.It can be seen from the simulation results that the above two routing mechanisms balance the energy consuming of sensor nodes in some degree, prolong the network lifetime. In the multiple sink WSNs surroundings, the network lifetime become longer, the network performance better.