GA-HPO PPO: A Hybrid Algorithm for Dynamic Flexible Job Shop Scheduling

Area of Science:

• Operations Research
• Artificial Intelligence
• Manufacturing Systems Engineering

Background:

• The Job Shop Scheduling Problem (JSP) is NP-hard, with extensions like the Dynamic Flexible Job Shop Scheduling Problem (DFJSP) posing significant real-world challenges.
• Stochastic task arrivals, varied deadlines, and task types in DFJSP make traditional optimization and rule-based methods insufficient.

Purpose of the Study:

• To develop a novel hybrid algorithm, GA-HPO PPO, to address the complexities of the Dynamic Flexible Job Shop Scheduling Problem.
• To enhance learning efficiency and scheduling performance for dynamic manufacturing environments.

Main Methods:

• A hybrid algorithm integrating genetic-algorithm-based hyperparameter optimization (GA-HPO) with proximal policy optimization (PPO).
• Training on four datasets and evaluation across ten benchmark DFJSP datasets.
• Comparative analysis against Double Deep Q-Network (DDQN), standard PPO, and rule-based heuristics.

Main Results:

• GA-HPO PPO significantly reduced overdue tasks (18.5 in 100-task, 197 in 1000-task scenarios) while maintaining high machine utilization (67%, 28%).
• The algorithm achieved competitive makespan values (108-114, 506-510 time units).
• Demonstrated 25% faster convergence and 30% lower performance variance compared to standard PPO, indicating robustness.

Conclusions:

• GA-HPO PPO offers an effective and scalable solution for the DFJSP, outperforming existing methods.
• The hybrid approach enhances dynamic scheduling optimization in practical manufacturing settings.
• The model exhibits strong generalization capabilities across diverse scheduling conditions.