Reinforcement Learning for Tackling Energy-Saving and Energy-Balance Dilemma of Cluster-Based Routing Protocols in WSNs

Yan Wang, Ke Deng, Yongli Ren, Jeffrey Chan, Guangjian Tian, Christian S. Jensen

Abstract

In Wireless Sensor Networks (WSNs), where sensor batteries cannot be replaced or recharged easily, the limited availability of energy has motivated a variety of cluster-based routing protocols. The unit operation of these protocols is known as a round (aka. a cycle). A round has two phases: 1) sensor clustering for selecting cluster-heads, and 2) routing for transmitting sensed data from sensors to a base-station via cluster-heads in multi-hop. The cluster-based routing protocols face the dilemma of energy-saving and energy-balance in both phases. Simply speaking, we have to trade off energy-saving for energy-balance and vice versa. First, frequent clustering can improve the energy-balance by rotating cluster-heads often among sensors, but the overhead for clustering increases quickly. Second, the multi-hop routing saves the energy of cluster-heads far away from transmitting data directly to base-station, but it deteriorates energy-balance since the cluster-heads near the base-station suffer more energy consumption for transmitting data of their own clusters and data from other cluster-heads as relays. Even though the related issues have been discussed and investigated widely, no learning-based solution has been proposed yet. This study attempts to fill the gap. Instead of proposing a new cluster-based routing protocol, this study adapts reinforcement learning to make two decisions concurrently for the existing protocols: 1) whether to redo sensor clustering for each round and 2) which cluster-heads should be excluded from being relays in multi-hop routing for each round. With several tailored techniques, the reinforcement learning-based solution could significantly boost the existing protocols by improving energy-saving and energy-balance, as evidenced by extensive experiments.
BibTeX
@article{wang2024reinforcement,
  author = {Yan Wang and Ke Deng and Yongli Ren and Jeffrey Chan and Guangjian Tian and Christian S. Jensen},
  title = {{Reinforcement Learning for Tackling Energy-Saving and Energy-Balance Dilemma of Cluster-Based Routing Protocols in WSNs}},
  journal = {IEEE Access},
  year = {2024},
  doi = {10.1109/access.2024.3468377},
}

Loading related content...

Figure