A hybrid Protection Scheme based on Deep Reinforcement Learning

Kordowich G, Jaworski M, Lorz T, Scheibe C, Jäger J (2022)

Publication Type: Conference contribution

Publication year: 2022

Publisher: IEEE

Event location: Novi Sad

DOI: 10.1109/ISGT-Europe54678.2022.9960539

Abstract

The growth of distributed renewable energy re-sources pushes classic protection schemes to their limits. An automated workflow is necessary to handle the volatility of future power systems characterized by constantly changing network states. Therefore a hybrid, deep reinforcement learning based protection scheme is proposed in this paper, which combines a single agent training for central protection with a multi agent training for a decentralized fallback.For this purpose, a training framework using a combination of PowerFactory and Python-Scripts connected via a Shared Memory is developed. The framework is used to train agents using Deep Recurrent Q-Learning. First a central protection is trained, which can then be imitated by decentral agents using only local measurements. Tests of the setup show a fast and selective response, which proves that deep reinforcement learning can indeed automatically generate a protection scheme fulfilling high requirements of dependability and security.

Authors with CRIS profile

How to cite

APA:

Kordowich, G., Jaworski, M., Lorz, T., Scheibe, C., & Jäger, J. (2022). A hybrid Protection Scheme based on Deep Reinforcement Learning. In Proceedings of the 2022 IEEE PES Innovative Smart Grid Technologies Conference Europe (ISGT-Europe). Novi Sad, RS: IEEE.

MLA:

Kordowich, Georg, et al. "A hybrid Protection Scheme based on Deep Reinforcement Learning." Proceedings of the 2022 IEEE PES Innovative Smart Grid Technologies Conference Europe (ISGT-Europe), Novi Sad IEEE, 2022.

BibTeX: Download