An End-to-End Framework for Energy-Efficient Cascaded Dual-Shop Collaborative Scheduling With Mating Operations.

Area of Science:

• Operations Research
• Industrial Engineering
• Artificial Intelligence

Background:

• Modern production involves complex, multi-stage processes requiring efficient scheduling.
• Interdependent workshops and integrated main/suborders (mating operations) present unique scheduling challenges.
• Existing scheduling models often overlook collaborative optimization in cascaded dual-shop environments.

Purpose of the Study:

• To formulate and address the energy-efficient cascaded dual-shop collaborative scheduling problem with mating operations (ECDCSP-M).
• To develop a mixed-integer linear programming (MILP) model for accurate problem representation.
• To design an end-to-end graph-based deep reinforcement learning (GDRL) approach for optimized scheduling.

Main Methods:

• Formulation of the ECDCSP-M using mixed-integer linear programming (MILP).
• Development of a graph-based deep reinforcement learning (GDRL) framework.
• Construction of a dual-shop heterogeneous graph and a heterogeneous graph neural network (HGNN) with a three-stage embedding mechanism.

Main Results:

• The proposed GDRL approach effectively models complex interdependencies, including mating operations.
• The method demonstrates strong generalization capabilities across diverse problem complexities.
• Robust solutions were achieved for challenging scheduling scenarios in cascaded dual-shop systems.

Conclusions:

• The developed GDRL approach offers an effective solution for energy-efficient collaborative scheduling in cascaded dual-shop systems.
• The heterogeneous graph construction and HGNN capture intricate system states and relationships.
• This research advances scheduling optimization for complex manufacturing environments with mating operations.