Surrogate-Assisted Genetic Programming With Simplified Models for Automated Design of Dispatching Rules.

Area of Science:

• Operations Research
• Artificial Intelligence
• Machine Learning

Background:

• Automated design of dispatching rules for production systems is a significant research area.
• Genetic programming (GP) is a powerful machine learning technique for this problem, but faces limitations in computational cost, accuracy, and interpretability.
• Existing methods often require substantial computational resources and may produce complex, difficult-to-understand rules.

Purpose of the Study:

• To develop a surrogate-assisted genetic programming (GP) approach to improve the quality of evolved dispatching rules.
• To address the limitations of computational cost, accuracy, and interpretability in automated rule design.
• To introduce new methods for simplifying and visualizing evolved rules to enhance their understandability.

Main Methods:

• Development of a surrogate-assisted genetic programming (GP) algorithm.
• Integration of simplification and visualization techniques for evolved rules.
• Experimental validation of the proposed algorithm against existing literature.

Main Results:

• The surrogate-assisted GP significantly improved the quality of evolved dispatching rules.
• The proposed method demonstrated effectiveness and efficiency, reducing computational costs.
• New simplification and visualization approaches successfully enhanced the interpretability of the evolved rules.

Conclusions:

• The developed surrogate-assisted GP is an effective and efficient approach for automated dispatching rule design.
• The new simplification and visualization methods are crucial for improving the practical application of evolved rules.
• This integrated system shows great potential for advancing automated design in production systems.