Learning-based energy-aware mobile charging under stochastic energy dynamics in rechargeable sensor networks
TL;DR
The proposed study presents a learning-driven MC framework, which explicitly represents the charging procedure as a sequence of tour-by-tour decision-making problems with energy and time constraints, and demonstrates the efficacy of the proposed approach in significantly prolonging the lifespan of RWSNs by improving energy transmission, reducing charging delays, and mortality rate compared to non-adaptive baselines.
Abstract
Rechargeable wireless sensor networks (RWSNs) operate under stochastic energy dynamics, fluctuating communication links, time-varying loads, and variable traffic requirements. These coupling aspects make mobile charging (MC) a sequential decision-making problem, according to which suboptimal selection of targets or routes can propagate into energy starvation and a high mortality rate. However, a lot of recent research strategies rely on non-adaptive heuristics which limit their ability to adjust to the evolving network conditions. The proposed study presents a learning-driven MC framework, which explicitly represents the charging procedure as a sequence of tour-by-tour decision-making problems with energy and time constraints. The proposed network model consists of three synergistic steps: (i) energy-aware k-mean cluster was employed with the optimal selection of centroids to partition the network area into clusters to optimize spatial compactness and energy balance, which restricts the decision space for online scheduling. Furthermore, (ii) a non-stationary multi-armed bandit formulation to learn the charging strategies for scheduling of the sensor nodes restricted in decision space, which involves the real-time decision-making to strike a balance between exploration of the variant space and exploitation of the optimal actions for each sensor node, which help tour-dependent charging utility by selecting service sets Si. (iii) A priority-weighted ant colony optimization for traveling salesman problem is applied to identify spatial distance for routing path and urgency-based heuristics constraints by leveraging the energy-aware pheromone deposition mechanism. As a result, the optimization of routing is adapted on the scheduling information, which is generated from the evolving energy state. The simulation results validate the efficacy of the proposed approach in significantly prolonging the lifespan of RWSNs by improving energy transmission, reducing charging delays, and mortality rate compared to non-adaptive baselines.