COOPERATIVE REINFORCEMENT LEARNING BASED THROUGHPUT OPTIMIZATION IN ENERGY HARVESTING-WIRELESS SENSOR NETWORK

摘要

nergy Harvesting-Wireless Sensor Network (EH-WSN) has got increasing attention in recent years. During its actual deployment, we find that the energythat can be harvested from the environment is always continually changing and unpredictable. This paper aims to investigate the energy management approach of EH-WSNunder such circumstance and propose a corresponding dynamic scheme to optimize thenetwork throughput. Here we adopt a Cooperative Reinforcement Learning (CRL) methodfor analysis. Firstly, we model the external environment status, and then the CRL algorithm based on Q-learning starts regulating the EH-node^ duty cycle according to theexternal energy’s variation; meanwhile, the feedback reward takes responsibility for theevaluation of CRL’s regulation. Different from traditional reinforcement learning, CRLfacilitates EH-nodes to share their local knowledge with others periodically. With thisinformation, EH-node chooses which action to take for the current time slot: (i) idling,(ii) sensing, (Hi) calculating, and (iv) transmitting. Experimental results show that theproposed scheme can make EH-node work energy-balanceable, and satisfy the networkthroughput requirement effectively, and it also improves the energy utilization efficiencyobviously in contrast with existing strategies.