Adaptive Energy Management by Reinforcement Learning in Cluster-based Solar powered WSNs

Abstract

Power management in wireless sensor networks (WSNs) is crucial due to the limited battery energy. Energy harvesting wireless sensor networks (EH-WSNs) can extract energy from environmental sources to break this limitation. Consistent with the nodes’ energy collection and workload, adaptive energy management strategies are used to improve energy utilization, thereby making the system perform efficiently and operate sustainably. Solar-powered sensor nodes in the network are grouped into clusters. Data collected by cluster members are sent to the cluster head and finally transmitted to the base station. The goal of the whole network is to maintain an energy neutrality state and to maximize the effective data throughput in the network. This paper proposes an adaptive power management method based on reinforcement learning (RL) algorithms for the solar-powered sensor networks. RL methods are applied for adjusting the transmission rate for each node to keep harvested energy more balanced among the whole clustered network. The Q-learning and the latest proximal policy optimization (PPO) methods are applied to maximize transmitting sensing data over the whole network. Experiments show that both methods are suitable for performing energy management by better adjusting the duty cycle of each node to achieve the overall application goal of maximizing information transmission.