Dual deep reinforcement learning agents-based integrated order acceptance and scheduling of mass individualized prototyping

• Published in: Journal of cleaner production 2023-11, Vol.427, p.139249, Article 139249
• Main Authors: Leng, Jiewu, Guo, Jiwei, Zhang, Hu, Xu, Kailin, Qiao, Yan, Zheng, Pai, Shen, Weiming
• Format: Article
• Language: English
• Subjects: artificial intelligence; cleaning; Deep reinforcement learning; design; Dual deep reinforcement learning agents; exhibitions; income; Mass individualized prototyping; methodology; Order acceptance decision; Virtual production scheduling
• ISSN: 0959-6526; 1879-1786
• DOI: 10.1016/j.jclepro.2023.139249

Coordinating order acceptance decisions with production scheduling to maximize revenue is challenging for Mass Individualized Prototyping (MIP) service providers. This paper presents a dual deep reinforcement learning agents-based (DDRLA) integrated order acceptance and scheduling (IOAS) for improving revenue. Firstly, a deep reinforcement learning-based virtual production scheduling (VPS) agent together with 8 state features and 11 action rules is designed. The VPS agent quickly and virtually reschedules a dynamically-arriving accepted order to evaluate the overall impact of accepting this order, including consumed capacity and increased revenue. Then, a deep reinforcement learning-based order acceptance decision (OAD) agent is designed. Based on the information guidance resulting from an interaction with the VPS agent, the OAD agent selectively accepts orders to maximize long-term gains, as well as to improve system resilience in the presence of a high ratio of urgent orders. The experiment results show that the proposed DDRLA method has better performance, compared with other IOAS approaches.