Coordination in today’s large power systems is critical to satisfy demand fluctuations and reduce the required generation capacity. In this paper, we employ Reinforcement Learning (RL) techniques to design a control system to enhance coordination among facilities and minimize the power production cost. The controlling agent of the RL observes the the system’s state (e.g., demand volumes), perform actions (e.g., production volumes), and get rewards or punishments as the effect of their actions. Therefore, agent gets trained in the environment, find an optimal policy towards the system’s goal, and establish a smart energy system. The developed control mechanism is tested for different loads (residential, commercial and industrial) and compared against non-smart approaches that exist in the literature.
Paper
Full text
Reinforcement Learning - Based Control Systems For Networked Power Infrastructures
Semantic Scholar · Engineering · 2021
Abstract
Coordination in today’s large power systems is critical to satisfy demand fluctuations and reduce the required generation capacity. In this paper, we employ Reinforcement Learning (RL) techniques to design a control system to enhance coordination among facilities and minimize the power production cost. The controlling agent of the RL observes the the system’s state (e.g., demand volumes), perform actions (e.g., production volumes), and get rewards or punishments as the effect of their actions. Therefore, agent gets trained in the environment, find an optimal policy towards the system’s goal, and establish a smart energy system. The developed control mechanism is tested for different loads (residential, commercial and industrial) and compared against non-smart approaches that exist in the literature.