Dynamic power management utilizing reinforcement learning with fuzzy reward for energy harvesting wireless sensor nodes

This paper considers a scenario that wireless sensor node is powered by harvesting energy with the characteristics of ambiguity and uncertainty. Reinforcement learning with fuzzy reward (RLFR) is used in this study for the dynamic power management of such energy harvesting wireless sensor nodes. By interacting with the given environment, the RLFR adjusts the duty-cycle in data sensing task according to the variable incoming energy related signals. The outcomes of these interactions are evaluated by fuzzy reward that express how well the duty-cycle adjustments in satisfying given requirement of energy neutrality. Simulation results show that the RLFR not only satisfies the sensing requirement in maintaining energy neutrality, but it also achieves better energy utilization in terms of residual battery energy in comparing with another existing dynamic power management method.