- Titel
- Beyond Monte Carlo: leveraging temporal difference learning for superior performance in dynamic resource allocation
- Zitierfähige Url:
- https://nbn-resolving.org/urn:nbn:de:bsz:l189-qucosa2-957852
- 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: 113-126
ISBN: 978-3-910103-05-4 - Erstveröffentlichung
- 2025
- DOI
- https://doi.org/10.33968/2025.13
- Abstract (EN)
- The application of reinforcement learning to dynamic industrial scheduling has gained increasing attention due to its capability to optimize complex manufacturing processes. With the advent of Industry 4.0 and the rise of smart manufacturing, new challenges arise that require innovative approaches, particularly in environments where there is a high degree of variability and uncertainty. Previous research has demonstrated that reinforcement learning, in particular Monte Carlo methods, is highly effective in optimizing resource allocation in job-shop scheduling scenarios. Even though Monte Carlo methods are effective where reward functions are clear and retrospective, real-world manufacturing systems often require more dynamic decision-making capabilities in real-time, for which temporaldifference methods are more appropriate. Despite the effectiveness of reinforcement learning in this area, there is a gap in understanding how different reward functions affect the learning process. In this study, we systematically examined multiple reward functions within a temporal difference system, applying a sensitivity analysis to assess their effects during the training and evaluation phases. Our results demonstrated that the overall performance of the production line improved despite the inherent complexity and challenges posed by temporal difference methods. Our findings demonstrate the effectiveness of multi-agent reinforcement learning for improving manufacturing efficiency, and provide implications for future research on scalable, real-time industrial scheduling.
- Freie Schlagwörter (EN)
- Multi-Agent Reinforcement Learning, Proximal Policy Optimization, Dynamic Resource Allocation, Dynamic Scheduling, Flexible Job Shop
- Herausgeber (Institution)
- Hochschule für Technik, Wirtschaft und Kultur Leipzig, Leipzig
- Version / Begutachtungsstatus
- publizierte Version / Verlagsversion
- URN Qucosa
- urn:nbn:de:bsz:l189-qucosa2-957852
- Veröffentlichungsdatum Qucosa
- 19.02.2025
- Dokumenttyp
- Konferenzbeitrag
- Sprache des Dokumentes
- Englisch
- Lizenz / Rechtehinweis
CC BY 4.0