Learning-Based Genetic Algorithm to Schedule an Extended Flexible Job Shop.

Area of Science:

• Operations Research
• Artificial Intelligence
• Manufacturing Systems Engineering

Background:

• Semiconductor manufacturing presents complex scheduling challenges.
• Existing optimization methods may struggle with efficiency and solution quality for these problems.

Purpose of the Study:

• To develop an efficient and effective optimization algorithm for the extended flexible job-shop scheduling problem in semiconductor manufacturing.
• To improve the balance between computational efficiency and solution quality in scheduling.

Main Methods:

• A learning-based genetic algorithm (LGA) integrating a parallel long short-term memory network-embedded autoencoder.
• Offline unsupervised training of the autoencoder to capture decision variable relationships.
• A co-evolving framework with network-embedded and regular subpopulations for enhanced search capabilities.

Main Results:

• The proposed LGA effectively captures complex decision variable linkages.
• The co-evolving framework balances global and local search, improving optimization ability.
• Numerical experiments demonstrate LGA's superior performance over CPLEX, heuristics, and other methods in finding high-quality solutions within reasonable timeframes.

Conclusions:

• The learning-based genetic algorithm offers a significant advancement for semiconductor manufacturing scheduling.
• The integration of deep learning (LSTM autoencoder) with evolutionary computation provides a powerful optimization tool.
• LGA demonstrates a strong balance between computational efficiency and solution quality for complex scheduling problems.