In the embedded system, the energy level of the device can satisfy the energy consumption of the minimum task at all times, which is a necessary condition to maintain the sustainable operation of the energy harvesting embedded system. Aiming at the problem that the energy-harvesting embedded system is unable to ensure the schedulability of tasks in the system due to energy shortage and energy waste, we propose an energy-saving scheduling strategy based on Q-learning. On the premise of meeting the real-time requirements of tasks, the agent can reduce the energy consumption in the process of task execution by reasonably arranging the task scheduling sequence, and ensuring that the remaining energy of the system can maintain the normal execution of the scheduling task. The experimental results show that the q-learning algorithm saves 61.8 % and 55.32 % of the average energy compared with the as late as possible (A LAP) and as soon as possible (ASAP) algorithms, respectively, and the time of system energy is maintained within a reasonable range is on average 21.01 % more stable than that of the ASAP.
Paper
Full text
An Energy-efficient Scheduling based on Q-learning for Energy Harvesting Embedded System
Semantic Scholar · Engineering · 2022
Abstract
In the embedded system, the energy level of the device can satisfy the energy consumption of the minimum task at all times, which is a necessary condition to maintain the sustainable operation of the energy harvesting embedded system. Aiming at the problem that the energy-harvesting embedded system is unable to ensure the schedulability of tasks in the system due to energy shortage and energy waste, we propose an energy-saving scheduling strategy based on Q-learning. On the premise of meeting the real-time requirements of tasks, the agent can reduce the energy consumption in the process of task execution by reasonably arranging the task scheduling sequence, and ensuring that the remaining energy of the system can maintain the normal execution of the scheduling task. The experimental results show that the q-learning algorithm saves 61.8 % and 55.32 % of the average energy compared with the as late as possible (A LAP) and as soon as possible (ASAP) algorithms, respectively, and the time of system energy is maintained within a reasonable range is on average 21.01 % more stable than that of the ASAP.