An actor-critic framework based on deep reinforcement learning for addressing flexible job shop scheduling problems.

Area of Science:

• Operations Research
• Artificial Intelligence
• Manufacturing Systems Engineering

Background:

• Industry 4.0 drives manufacturing towards customization and flexibility.
• Evolving market demands necessitate advanced solutions for flexible job shop scheduling problems (FJSPs).
• Traditional scheduling methods face challenges in dynamic and complex manufacturing environments.

Purpose of the Study:

• To propose a novel reinforcement learning (RL) approach for addressing flexible job shop scheduling problems (FJSPs).
• To develop an actor-critic architecture integrating value-based and policy-based RL methods for optimal policy generation.
• To enhance manufacturing flexibility and efficiency in response to Industry 4.0 demands.

Main Methods:

• Utilized a hybrid actor-critic reinforcement learning architecture.
• Formulated the Markov decision process with a comprehensive feature set and eight action sets inspired by scheduling rules.
• Designed reward functions to minimize job completion times and ensure constraint adherence.

Main Results:

• The proposed RL framework demonstrated superior performance compared to heuristic, RL, and intelligent algorithms on standard FJSP benchmarks.
• The method achieved significant improvements in adaptability and efficiency, especially with large-scale datasets.
• Consistently outperformed traditional scheduling approaches in simulations.

Conclusions:

• The developed reinforcement learning approach offers an effective solution for complex flexible job shop scheduling problems.
• The actor-critic architecture provides an optimal policy for dynamic manufacturing environments.
• The findings highlight the potential of RL in advancing Industry 4.0 manufacturing capabilities.