Návrh pokročilých řídicích algoritmů pro IoT senzory napájené termoelektrickým generátorem

Abstract

Alternative power sources became a big challenge in powering devices deployed in Internet of Things (IoT) networks due to effort to reduce using of batteries, focus on sustainability and reduce regular maintenance. Thermoelectric generators (TEGs) are solid state energy harvesters which can convert thermal energy into electrical energy in a reliable and renewable manner. Based on the state of the art, this dissertation thesis presents a comprehensive review focused on machine learning approaches used in TEG-powered IoT sensors for available energy prediction or energy management. The dissertation thesis also brings an overview of application areas of TEG-powered IoT devices which obtain the temperature difference from various heat sources such as an environment, biological structures, machines or technologies. The presented dissertation thesis objectives at new applications with a energy prediction and management based on machine learning methods where supervised algorithms could allow better estimation of incoming energy and unsupervised and semi-supervised approaches lead to adaptive and dynamic operation. The device’s controllers are based on a reinforcement learning approaches, such as a Q-learning (QL) and a Double Q-learning (DQL) method. The model hardware incorporates a TEG energy harvesting subsystem with a DC/DC converter, a load module with a microcontroller, and a LoRaWAN communications interface. The model is controlled according to adaptive measurements and transmission periods. The controllers reward policy evaluate the level of charge available to the device. Using four years of historical soil temperature data in an experimental simulation of several controllers configurations, the QL and DQL controller demonstrated correct operation, high reliability, flexibility, and high efficiency in the use of the harvested energy.