In this paper, we examine the internet of things system which is dedicated for smart cities, smart factory, and connected cars, etc. To support such systems in wide area with low power consumption, energy harvesting technology without wired charging infrastructure is one of the important issues for longevity of networks. In consideration of the fact that the position and amount of energy charged for each device might be unbalanced according to the distribution of nodes and energy sources, the problem of maximizing the minimum throughput among all nodes becomes a NP-hard challenging issue. To overcome this complexity, we propose a machine learning based relaying topology algorithm with a novel backward-pass rate assessment method to present proper learning direction and an iterative balancing time slot allocation algorithm which can utilize the node with sufficient energy as the relay. To validate the proposed scheme, we conducted simulations on the system model we established, thus confirm that the proposed scheme is stable and superior to conventional schemes.

翻译：本文针对专用于智慧城市、智能工厂及车联网等场景的物联网系统展开研究。为在广域范围内以低功耗支持此类系统，无需有线充电基础设施的能量收集技术成为保障网络持久性的关键课题之一。考虑到各设备充电位置与能量分布可能因节点及能量源的分布差异而存在不均衡性，最大化全网节点最小吞吐量的优化问题成为NP难挑战。为突破该复杂度瓶颈，我们提出一种基于机器学习的中继拓扑算法，创新性地引入反向传播速率评估方法以提供合理的训练方向，并设计了一种迭代式均衡时隙分配算法，可将能量充足的节点用作中继。通过在自建系统模型上的仿真验证，证实所提方案具有稳定性且优于传统方案。