- Titel
- YoloRL: simplifying dynamic scheduling through efficient action selection based on multi-agent reinforcement learning
- Zitierfähige Url:
- https://nbn-resolving.org/urn:nbn:de:bsz:l189-qucosa2-957990
- Konferenz
- 21. AALE-Konferenz. Dresden, 12. März - 14. März 2025
- Quellenangabe
- Tagungsband AALE 2025
; menschzentrierte Automation im digitalen Zeitalter
Herausgeber: Hochschule für Technik, Wirtschaft und Kultur Leipzig
Erscheinungsort: Leipzig
Erscheinungsjahr: 2025
Seiten: 127-137
ISBN: 978-3-910103-05-4 - Erstveröffentlichung
- 2025
- DOI
- https://doi.org/10.33968/2025.14
- Abstract (EN)
- In modern manufacturing environments, it is essential to be able to react autonomously and dynamically to unpredictable events in an automated manner in order to schedule production in a cost-effective manner. One of the prerequisites for the development of this technology is the progressive integration of cyberphysical systems into industrial sectors. Data generated by the industry constitutes the basis for operative and strategic decision-making in this context. Collecting these data in real time, transforming it if necessary, and analyzing it in order to ensure time-critical decision-making is a major challenge. This paper presents a novel approach that simplifies dynamic scheduling through efficient action selection. YoloRL, the method presented in this paper, which is based on reinforcement learning, which allows for a reduction in the complexity of the training process in a substantial way. For the purpose of identifying promising action sequences, YoloRL does not take into consideration all of the state information of an episode; it only takes into account the initial state. As a result, training complexity is significantly reduced while at the same time robust and adaptive control can be achieved. This study improves the manufacturing system’s performance by minimizing the overall completion time (for any given order). Experimental results indicate that the proposed method results in a faster generalization of the domain knowledge and provides for a powerful policy that is both efficient and reliable in dynamic environments. With YoloRL, overall completion time is reduced by a moderate but quantifiable amount compared with the traditional approach. In accordance with our experimental results, the proposed methodology has the ability to accelerate and stabilize the training process. Thus, a reliable and generalizable policy network is established, which can nevertheless respond dynamically to unforeseen events and changing environmental conditions due to its adaptability. The policy ...
- Freie Schlagwörter (DE)
- Multi-Agent Reinforcement Learning, Dynamic Scheduling Problem, Flexible Job Shop, YOLO
- Herausgeber (Institution)
- Hochschule für Technik, Wirtschaft und Kultur Leipzig, Leipzig
- Version / Begutachtungsstatus
- publizierte Version / Verlagsversion
- URN Qucosa
- urn:nbn:de:bsz:l189-qucosa2-957990
- Veröffentlichungsdatum Qucosa
- 19.02.2025
- Dokumenttyp
- Konferenzbeitrag
- Sprache des Dokumentes
- Englisch
- Lizenz / Rechtehinweis
CC BY 4.0