Multi-Agent Reinforcement Learning for Job Shop Scheduling in Dynamic Environments

Sustainability, Год журнала: 2024, Номер 16(8), С. 3234 - 3234

Опубликована: Апрель 12, 2024

In response to the challenges of dynamic adaptability, real-time interactivity, and optimization posed by application existing deep reinforcement learning algorithms in solving complex scheduling problems, this study proposes a novel approach using graph neural networks complete task job shop scheduling. A distributed multi-agent architecture (DMASA) is constructed maximize global rewards, modeling intelligent manufacturing problem as sequential decision represented graphs Graph Embedding–Heterogeneous Neural Network (GE-HetGNN) encode state nodes map them optimal strategy, including machine matching process selection strategies. Finally, an actor–critic architecture-based proximal policy algorithm employed train network optimize decision-making process. Experimental results demonstrate that proposed framework exhibits generalizability, outperforms commonly used rules RL-based methods on benchmarks, shows better stability than single-agent architectures, breaks through instance-size constraint, making it suitable for large-scale problems. We verified feasibility our method specific experimental environment. The research can achieve formal mapping with physical processing workshops, which aligns more closely real-world green issues makes easier subsequent researchers integrate actual environments.

Язык: Английский

---

A discrete event simulator to implement deep reinforcement learning for the dynamic flexible job shop scheduling problem

Simulation Modelling Practice and Theory, Год журнала: 2024, Номер 134, С. 102948 - 102948

Опубликована: Апрель 21, 2024

The job shop scheduling problem, which involves the routing and sequencing of jobs in a context, is relevant subject industrial engineering. Approaches based on Deep Reinforcement Learning (DRL) are very promising for dealing with variability real working conditions due to dynamic events such as arrival new machine failures. Discrete Event Simulation (DES) essential training testing DRL approaches, interaction an intelligent agent production system. Nonetheless, there numerous papers literature techniques, developed solve Dynamic Flexible Job Shop Problem (DFJSP), have been implemented evaluated absence simulation environment. In paper, limitations these techniques highlighted, numerical experiment that demonstrates their ineffectiveness presented. Furthermore, order provide scientific community tool designed be used conjunction agent-based discrete event simulator also

Язык: Английский

---

Real-time scheduling for two-stage assembly flowshop with dynamic job arrivals by deep reinforcement learning

Advanced Engineering Informatics, Год журнала: 2024, Номер 62, С. 102632 - 102632

Опубликована: Июнь 19, 2024

Язык: Английский

---

Multi-agent deep reinforcement learning for dynamic reconfigurable shop scheduling considering batch processing and worker cooperation

Robotics and Computer-Integrated Manufacturing, Год журнала: 2024, Номер 91, С. 102834 - 102834

Опубликована: Июль 18, 2024

Язык: Английский

---

A multi-agent double Deep-Q-network based on state machine and event stream for flexible job shop scheduling problem

Advanced Engineering Informatics, Год журнала: 2023, Номер 58, С. 102230 - 102230

Опубликована: Окт. 1, 2023

Язык: Английский

---