Fuzzy Q-Reinforcement Learning-Based Energy Optimization in IoT Network

Abstract

Motivated by the growing environmental concerned (effect of greenhouse gases) coupled with increasing cost of energy, green computing emerges as promising solution to energy-limited IoT network. As IoT network consists of limited low-battery power smart sensors having ability to connect over wireless network for transmission of data, energy harvested from the environment by the sensor node reduces carbon emission and also recharges its battery continuously, and this harvested energy is used by sensors for its working operation that enhances the lifetime of the IoT network. In this paper, a Fuzzy Q-reinforcement learning (FQRL) scheme using fuzzy logic and model-free Q-learning to optimize the energy consumption in perpetual operations of IoT nodes is presented. The optimization of energy consumption is subject to adaptive duty cycle exercised to smart sensors. The learning agent of FQRL updates If-Then rules of fuzzy controllers according to reward received by learning agent through interacting with environment. The learning agent rewards for good action (increasing the firing strength of rule) and punishes (decrease the firing strength of rule) for bad action subject to maintain the energy neutrality condition. Finally, simulation results show the proposed FQRL outperforms in terms of duty cycle and residual energy after perpetual operation. It means presented algorithm FQRL provides smart sensors to achieve better charging status of their battery and suitable for energy harvested IoT networks.