Q-learning Based Adaptive Zone Partition for Load Balancing in Multi-Sink Wireless Sensor Networks

In many researches on load balancing in MultiSink WSNs, sensors usually choose the nearest sink as destination for sending data. If all sensors in this area all follow the nearest-sink strategy, sensors around nearest sink called hotspot will exhaust energy early and this sink is isolated from network. In this paper, we propose a load balancing scheme for multi-sink WSNs. A mobile anchor with directional antenna is introduced to adaptively partition the network into several zones so the traffic load in the region can be assigned to the sink. Besides, to adapt to different data traffic pattern, we apply machine learning to mobile anchor and implement a Q-learning agent. Through interactions with environment, the agent can discovery a near-optimal control policy for movement of mobile anchor and achieve minimization of residual energy’s variance among sinks, which prevent the early isolation of sink and prolong the network lifetime.