Research On Self-organization For Wireless Sensor Networks Based On Reinforcement Learning

In WSN, Sensor nodes have constrained energy, computer skills and communication ability. It is important and difficult to design for wireless sensor network self-organization that how to collaborate between the nodes and to play its role in the overall and comprehensive. Meanwhile, Because of the characteristics of the sensor node limited energy, how to extend the network lifetime is also another point for a self-organizing wireless sensor network design. In this paper, self-organization of wireless sensor networks as the main object of study. The author's dissertation analyzes several typical researches of self-organization for wireless sensor networks based on reinforcement learning and proposes two researches of self-organization for wireless sensor networks based on reinforcement learning.It can be seen from the existing method of self-organization, self-organizing wireless sensor networks is implemented by routing protocol or topology control. This dissertation proposes QLSOP (Q-learning of self-organization policy ) and TD-MST (temporal differences minimum spanning tree self-organization policy ). QLSOP makes full use of the Q-Learning algorithm for dynamic adaptation, considering the distance, hops, communication energy consumption, the remaining energy, so that the node can use the value function to find the optimal path and balance the energy consumption and the residual energy. TD-MST calculate the ratio of energy consumption and residual energy using node energy prediction. TD-MST construct the minimum spanning tree using the ratio as a weight and sink nodes as root. It is noteworthy that the use of energy predicted TD algorithm can avoid the residual energy of nodes exchanging information and reduce the energy consumption of the control signaling.It can be seen from the simulation results, the above two methods of self-organizing can control the average delay of network in a lower level, while extending the network life. This research is mainly applied to small-scale scenarios which requires a higher performance of self-organization , but also helps to explore new methods of self-organization.