Energy Efficiency in Distribution Systems Based on Task Scheduling using Reinforcement Learning and Actor-Critic Method

Energy consumption in data centers and systems is increasing rapidly, which is a fundamental issue in the present age. An important advantage of distribution systems is cost savings because they do not require the initial installation and commissioning of resources and are scalable and flexible, but Load balance and scheduling are a challenge in distribution systems. This paper presents a method for scheduling tasks on dynamically available resources and the system uses continuous learning for best performance. In the proposed method, the Actor-Critic is used to improve decision making in reinforcement learning to extract the rules of distribution and use them in reinforcement learning to improve and facilitate energy efficiency goals. The proposed method was compared with the method presented in the same work in terms of "Completion time of all tasks " and "energy consumption" criteria. In th e evaluations, the energy consumption of the proposed method was more appropriate than the compared method. In environments where queue length is formed and resources and requests change rapidly, this energy consumption increases slightly due to the increasing number of scenarios and continuous learning. In general, the proposed method is suitable for stable environments, low changes or more balanced time intervals Because the learning process takes time.