Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LLM-Upgraded Graph Reinforcement Learning for Carbon-Aware Job Scheduling in Smart Manufacturing

Zhiying Yang, Fang Liu, Wei Zhang, Xin Lou, Malcolm Yoke Hean Low, Boon Ping Gan

TL;DR

Luca introduces an LLM-upgraded graph deep RL framework for carbon-aware FJSP, fusing GNN-derived structural embeddings with LLM-derived contextual embeddings via a gated fusion module. The policy, trained with PPO, optimizes a dual objective of makespan and carbon emissions, guided by state and feedback prompts that enhance the LLM's situational awareness. Extensive experiments on synthetic and public datasets show Luca outperforms carbon-aware baselines and demonstrates robustness across different λ settings and emission-rate variations. The work highlights the practical potential of combining semantic reasoning with graph structure for sustainable, real-time manufacturing scheduling.

Abstract

This paper presents \textsc{Luca}, a \underline{l}arge language model (LLM)-\underline{u}pgraded graph reinforcement learning framework for \underline{c}arbon-\underline{a}ware flexible job shop scheduling. \textsc{Luca} addresses the challenges of dynamic and sustainable scheduling in smart manufacturing systems by integrating a graph neural network and an LLM, guided by a carefully designed in-house prompting strategy, to produce a fused embedding that captures both structural characteristics and contextual semantics of the latest scheduling state. This expressive embedding is then processed by a deep reinforcement learning policy network, which generates real-time scheduling decisions optimized for both makespan and carbon emission objectives. To support sustainability goals, \textsc{Luca} incorporates a dual-objective reward function that encourages both energy efficiency and scheduling timeliness. Experimental results on both synthetic and public datasets demonstrate that \textsc{Luca} consistently outperforms comparison algorithms. For instance, on the synthetic dataset, it achieves an average of 4.1\% and up to 12.2\% lower makespan compared to the best-performing comparison algorithm while maintaining the same emission level. On public datasets, additional gains are observed for both makespan and emission. These results demonstrate that \textsc{Luca} is effective and practical for carbon-aware scheduling in smart manufacturing.

LLM-Upgraded Graph Reinforcement Learning for Carbon-Aware Job Scheduling in Smart Manufacturing

TL;DR

Luca introduces an LLM-upgraded graph deep RL framework for carbon-aware FJSP, fusing GNN-derived structural embeddings with LLM-derived contextual embeddings via a gated fusion module. The policy, trained with PPO, optimizes a dual objective of makespan and carbon emissions, guided by state and feedback prompts that enhance the LLM's situational awareness. Extensive experiments on synthetic and public datasets show Luca outperforms carbon-aware baselines and demonstrates robustness across different λ settings and emission-rate variations. The work highlights the practical potential of combining semantic reasoning with graph structure for sustainable, real-time manufacturing scheduling.

Abstract

This paper presents \textsc{Luca}, a \underline{l}arge language model (LLM)-\underline{u}pgraded graph reinforcement learning framework for \underline{c}arbon-\underline{a}ware flexible job shop scheduling. \textsc{Luca} addresses the challenges of dynamic and sustainable scheduling in smart manufacturing systems by integrating a graph neural network and an LLM, guided by a carefully designed in-house prompting strategy, to produce a fused embedding that captures both structural characteristics and contextual semantics of the latest scheduling state. This expressive embedding is then processed by a deep reinforcement learning policy network, which generates real-time scheduling decisions optimized for both makespan and carbon emission objectives. To support sustainability goals, \textsc{Luca} incorporates a dual-objective reward function that encourages both energy efficiency and scheduling timeliness. Experimental results on both synthetic and public datasets demonstrate that \textsc{Luca} consistently outperforms comparison algorithms. For instance, on the synthetic dataset, it achieves an average of 4.1\% and up to 12.2\% lower makespan compared to the best-performing comparison algorithm while maintaining the same emission level. On public datasets, additional gains are observed for both makespan and emission. These results demonstrate that \textsc{Luca} is effective and practical for carbon-aware scheduling in smart manufacturing.

Paper Structure

This paper contains 26 sections, 7 equations, 2 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Problem Formulation: Carbon-Aware FJSP
  3. Luca: Methodology
  4. System Architecture Overview
  5. LLM-based Embedding
  6. State Prompt
  7. Feedback Prompt
  8. LLM
  9. GNN-based Embedding
  10. Fusion of Embeddings
  11. Carbon-Aware Policy
  12. Policy Formulation and Rewards
  13. Validation for Generalization
  14. Training Procedure
  15. Experimental Study
  16. ...and 11 more sections

Figures (2)

  • Figure 1: System architecture of Luca for carbon-aware FJSP. The framework leverages GNN and LLM to produce a structural and contextual embedding of a scheduling state, and RL to learn policies that jointly optimize makespan and carbon emissions.
  • Figure 2: Multi-objective performance comparison between Luca and Drl-c across various values of the balance parameter $\lambda$. Luca yields Pareto-dominant schedules.