Electricity distribution system is at the forefront of major changes in terms of electrification, especially in mobility and heating. A large number of distributed energy resources are being connected to the distribution network, intensive electrification of mobility is taking place, all while household consumption continues to rise. All these new elements in the network are causing problems with power quality. Traditionally, in order to provide sufficient quality of electricity supply, network reinforcements were made. Our goal is to explore possible benefits of using deep reinforcement learning algorithms in order to mitigate over/under voltage problems compared to traditional network reinforcement approach. This work focuses on two state-of-the-art architectures of deep reinforcement learning. In order to find efficient control strategies, we implemented a deep reinforcement learning agent that, based on the training dataset of actual historical electricity consumption and production data, learns voltage control strategy in an actual low voltage distribution network. Learned models were also tested on an independent test dataset.
|
