Reward Function Evaluation in a Reinforcement Learning Approach for Energy Management
Reward Function Evaluation in a Reinforcement Learning Approach for Energy Management
复制标题
能源管理强化学习方法中的奖励函数评估
DOI：
--
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
J. Diguet
中科院分区：
文献类型：
--
作者：
Yohann Rioual;Y. Moullec;J. Laurent;Muhidul Islam Khan;J. Diguet
在过去的十年中，由于需要处理更多的资料、传输更多的数据和更长的作业周期，城市无线网络的能源需求有所增加。另一方面，电池技术的进步还不够快，无法满足这些需求。因此，小型化能量收集技术被越来越多地用于补充wban电池。然而，由于在节点运行过程中收获的能量变化很大，这给系统带来了不确定性。研究表明，强化学习算法可以用于管理节点中的能量，因为它们能够在不确定的情况下做出决策。但这些算法的效率取决于它们的奖励函数。在本文中，我们探讨了不同的奖励函数，并试图找出最合适的变量，以在这些函数中使用，以获得期望行为。四种不同奖励函数的实验结果说明了其选择如何影响节点的能量消耗。
In the past decade, the energy needs in WBANs have increased due to more information to be processed, more data to be transmitted and longer operational periods. On the other hand, battery technologies have not improved fast enough to cope with these needs. Thus, miniaturized energy harvesting technologies are increasingly used to complement the batteries in WBANs. However, this brings uncertainties in the system since the harvested energy varies a lot during the node operation. It has been shown that reinforcement learning algorithms can be used to manage the energy in the nodes since they are able to make decisions under uncertainty. But the efficiency of these algorithms depends on their reward function. In this paper we explore different reward functions and seek to identify the most suitable variables to use in such functions to obtain the expected behavior. Experimental results with four different reward functions illustrate how the choice thereof impacts the energy consumption of the nodes.