Evolving Scheduling Heuristics via Genetic Programming With Feature Selection in Dynamic Flexible Job-Shop Scheduling

Area of Science:

• Operations Research
• Artificial Intelligence
• Computer Science

Background:

• Dynamic flexible job-shop scheduling (DFJSS) is a complex optimization problem.
• Genetic programming hyperheuristics (GPHH) are used for evolving scheduling heuristics.
• Feature selection is crucial for GPHH performance but underexplored in DFJSS.

Purpose of the Study:

• To develop a novel GPHH framework with integrated feature selection for DFJSS.
• To automatically evolve scheduling heuristics using only relevant features.
• To enhance the interpretability and efficiency of GPHH for DFJSS.

Main Methods:

• A two-stage GPHH framework incorporating feature selection.
• Individual adaptation strategies to guide feature selection.
• Evolution of scheduling heuristics based on selected features for DFJSS.

Main Results:

• Achieved more interpretable scheduling heuristics with fewer features and smaller sizes.
• Demonstrated comparable heuristic quality to existing methods.
• Significantly reduced training time for heuristic evolution.

Conclusions:

• The proposed feature selection approach enhances GPHH for DFJSS.
• The method leads to more efficient and understandable scheduling solutions.
• This work addresses a critical gap in applying GPHH to dynamic scheduling problems.