TY - JOUR
T1 - Deep reinforcement learning-based memetic algorithm for solving dynamic distributed green flexible job shop scheduling problem with finite transportation resources
AU - Zhou, Xinxin
AU - Wang, Fuyu
AU - Wu, Bin
AU - Li, Yan
AU - Shen, Nannan
N1 - Publisher Copyright:
© 2025 Elsevier B.V.
PY - 2025/4
Y1 - 2025/4
N2 - To solve the dynamic distributed green flexible job shop scheduling problem with integrated multi-automated guided vehicles (AGVs) transportation (DDGFJSP-MT), a coupled mathematical model is constructed in this study with the objective of minimizing the makespan and total carbon emissions. The complex coupled roles between factories, jobs, machines, and AGVs induced by machine breakdown are explored. Meanwhile, a deep Q-network-based dynamic efficient memetic algorithm (DQN-DEMA) is proposed to solve the problem. First, a four-layer coding is designed to characterize the DDGFJSP-MT, and a novel dynamic decoding technique is developed based on the state variations of the involved subjects and their strong coupling effects following the machine breakdown. Second, an alternating hybrid initialization strategy is employed to improve the quality and diversity of the rescheduled population. Then, several neighborhood search structures are designed based on critical path and bottleneck operation, and DQN is applied to recommend the most suitable local search operator for each elite individual, accelerating the convergence of the rescheduled population and effectively avoiding the waste of algorithmic resources. Finally, performance validation on 20 instances demonstrates that DQN-DEMA obtains the Pareto frontier with higher quality and diversity in 15 instances compared to the six state-of-the-art algorithms.
AB - To solve the dynamic distributed green flexible job shop scheduling problem with integrated multi-automated guided vehicles (AGVs) transportation (DDGFJSP-MT), a coupled mathematical model is constructed in this study with the objective of minimizing the makespan and total carbon emissions. The complex coupled roles between factories, jobs, machines, and AGVs induced by machine breakdown are explored. Meanwhile, a deep Q-network-based dynamic efficient memetic algorithm (DQN-DEMA) is proposed to solve the problem. First, a four-layer coding is designed to characterize the DDGFJSP-MT, and a novel dynamic decoding technique is developed based on the state variations of the involved subjects and their strong coupling effects following the machine breakdown. Second, an alternating hybrid initialization strategy is employed to improve the quality and diversity of the rescheduled population. Then, several neighborhood search structures are designed based on critical path and bottleneck operation, and DQN is applied to recommend the most suitable local search operator for each elite individual, accelerating the convergence of the rescheduled population and effectively avoiding the waste of algorithmic resources. Finally, performance validation on 20 instances demonstrates that DQN-DEMA obtains the Pareto frontier with higher quality and diversity in 15 instances compared to the six state-of-the-art algorithms.
KW - Deep reinforcement learning
KW - Distributed flexible job shop scheduling
KW - Finite AGV transportation
KW - Machine breakdown
KW - Memetic algorithm
UR - http://www.scopus.com/inward/record.url?scp=85218146054&partnerID=8YFLogxK
U2 - 10.1016/j.swevo.2025.101885
DO - 10.1016/j.swevo.2025.101885
M3 - 文章
AN - SCOPUS:85218146054
SN - 2210-6502
VL - 94
JO - Swarm and Evolutionary Computation
JF - Swarm and Evolutionary Computation
M1 - 101885
ER -