Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning-Based Approaches for Job Shop Scheduling Problems: A Review

Karima Rihane, Adel Dabah, Abdelhakim AitZai

TL;DR

This survey addresses the Job Shop Scheduling Problem (JSSP) and surveys learning-based approaches, emphasizing neural networks and reinforcement learning as data-driven strategies to complement traditional exact and heuristic methods. It formalizes JSSP, reviews exact, heuristic, and metaheuristic techniques, and catalogs a broad spectrum of learning-based methods, including shallow and deep neural networks, graph-based deep learning, and deep reinforcement learning, often leveraging disjunctive graph representations. Key findings indicate that graph-aware deep RL methods and graph-structured encodings with PPO or DRL show strong potential, outperforming traditional dispatching rules on several benchmarks, yet generalization to unseen instances and scalability to large problems remain significant challenges. The authors highlight future directions such as deeper graph-centric architectures, graph-based DRL, multi-agent coordination, and addressing BJSSP to tackle real-world, large-scale scheduling under complex constraints.

Abstract

Job Shop Scheduling (JSS) is one of the most studied combinatorial optimization problems. It involves scheduling a set of jobs with predefined processing constraints on a set of machines to achieve a desired objective, such as minimizing makespan, tardiness, or flowtime. Since it introduction, JSS has become an attractive research area. Many approaches have been successfully used to address this problem, including exact methods, heuristics, and meta-heuristics. Furthermore, various learning-based approaches have been proposed to solve the JSS problem. However, these approaches are still limited when compared to the more established methods. This paper summarizes and evaluates the most important works in the literature on machine learning approaches for the JSSP. We present models, analyze their benefits and limitations, and propose future research directions.

Learning-Based Approaches for Job Shop Scheduling Problems: A Review

TL;DR

This survey addresses the Job Shop Scheduling Problem (JSSP) and surveys learning-based approaches, emphasizing neural networks and reinforcement learning as data-driven strategies to complement traditional exact and heuristic methods. It formalizes JSSP, reviews exact, heuristic, and metaheuristic techniques, and catalogs a broad spectrum of learning-based methods, including shallow and deep neural networks, graph-based deep learning, and deep reinforcement learning, often leveraging disjunctive graph representations. Key findings indicate that graph-aware deep RL methods and graph-structured encodings with PPO or DRL show strong potential, outperforming traditional dispatching rules on several benchmarks, yet generalization to unseen instances and scalability to large problems remain significant challenges. The authors highlight future directions such as deeper graph-centric architectures, graph-based DRL, multi-agent coordination, and addressing BJSSP to tackle real-world, large-scale scheduling under complex constraints.

Abstract

Job Shop Scheduling (JSS) is one of the most studied combinatorial optimization problems. It involves scheduling a set of jobs with predefined processing constraints on a set of machines to achieve a desired objective, such as minimizing makespan, tardiness, or flowtime. Since it introduction, JSS has become an attractive research area. Many approaches have been successfully used to address this problem, including exact methods, heuristics, and meta-heuristics. Furthermore, various learning-based approaches have been proposed to solve the JSS problem. However, these approaches are still limited when compared to the more established methods. This paper summarizes and evaluates the most important works in the literature on machine learning approaches for the JSSP. We present models, analyze their benefits and limitations, and propose future research directions.
Paper Structure (12 sections, 6 equations, 3 figures, 6 tables, 2 algorithms)

This paper contains 12 sections, 6 equations, 3 figures, 6 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Formulation of the Problem
  4. Job-Shop Scheduling Problem and Resolution Techniques
  5. Machine Learning Approaches for Job Shop Scheduling Problem
  6. Neural Networks
  7. Shallow Neural Networks
  8. Deep Neural Networks
  9. Reinforcement Learning
  10. Deep Reinforcement Learning
  11. Discussion and Future Work
  12. Conclusion

Figures (3)

  • Figure 1: Disjunctive graph representation for JSSP. (a) Visualizes the JSSP instance described in Table \ref{['tab:01']}, where black arrows represent conjunctive arcs (precedence constraints), and dotted lines denote disjunctive arcs grouped into machine-specific cliques, each distinguished by a unique color. (b) Depicts a complete solution where all disjunctive arcs have been assigned directions, resulting in a directed acyclic graph
  • Figure 2: Model of neuron in network.
  • Figure 3: Reinforcement learning one time step.