Uncertain Interruptibility Multiobjective Flexible Job Shop via Deep Reinforcement Learning Based on Heterogeneous...

Area of Science:

• Operations Research
• Artificial Intelligence

Background:

• Flexible job shop problem (FJSP) research often overlooks practical constraints.
• Realistic constraints include employee working hours and non-interruptible operations.

Purpose of the Study:

• To develop an improved deep reinforcement learning (DRL) approach for the flexible job shop problem (FJSP).
• To address realistic constraints such as employee working hours and non-interruptible operations.
• To concurrently optimize makespan, total costs, and total lateness.

Main Methods:

• Utilizes end-to-end multidecision-intelligent body proximal policy optimization (m-PPO).
• Embeds a heterogeneous graph self-attention neural network (HGAN) for feature extraction.
• Employs rule-driven job decision agents and data-driven operation-machine (O-M) pair decision agents.

Main Results:

• The HGAN model efficiently extracts features from heterogeneous graphs.
• Agents incorporate problem-specific knowledge for decision-making.
• Network-generated, self-updated weight parameters optimize multiple objectives concurrently.

Conclusions:

• The proposed DRL framework effectively handles realistic constraints in FJSP.
• The method demonstrates superior performance in optimizing makespan, costs, and lateness.
• Numerical experiments validate the effectiveness of the improved DRL approach.