Reinforcement Learning Strategies for Energy Management in Low Power IoT

摘要

Energy management in low power IoT is a difficult problem. Modeling the consumption of a sensor node is complicated, they operate in a stochastic environment. They harvest energy in their environment but energy sources present time-varying behavior. It becomes hazardous to predict in advance the energy behavior of our system. In this paper we propose a new approach using both neural networks to estimate the harvesting energy and reinforcement learning algorithms to find the operating parameters to maximize the node's performance while preserving its energy resources.