Dynamic job shop scheduling under multiple order disturbances using deep reinforcement learning.

Area of Science:

• Operations Research
• Artificial Intelligence
• Manufacturing Systems Engineering

Background:

• Dynamic job shop scheduling is complex due to multiple order disturbances.
• Existing methods struggle with real-time adjustments and multiobjective optimization.

Purpose of the Study:

• To develop a novel approach for dynamic job shop scheduling using multiagent deep reinforcement learning.
• To minimize total tardiness and makespan in manufacturing environments.

Main Methods:

• Utilized Independent Proximal Policy Optimization (IPPO), a multiagent deep reinforcement learning algorithm.
• Represented system states using a five-channel two-dimensional image.
• Designed a reward function focused on reducing total tardiness and makespan.

Main Results:

• The IPPO-based approach demonstrated superior performance compared to traditional deep reinforcement learning algorithms and dispatching rules across 72 scenarios.
• Achieved significant improvements in minimizing total tardiness and makespan.
• Showcased strong optimization and exploration capabilities.

Conclusions:

• The proposed IPPO method offers a promising and effective solution for complex, multiobjective scheduling in dynamic manufacturing environments.
• Highlights the potential of multiagent deep reinforcement learning for addressing real-world scheduling challenges.