Large Language Models in Operations Research: Methods, Applications, and Challenges
This paper introduces LLM-based automatic modeling and optimization, significantly advancing intelligent decision-making in complex operations research problems.
Key Findings
Methodology
The study employs large language models like GPT-4 for three main pathways: automatic modeling, auxiliary optimization, and direct solving. It leverages prompt engineering, multi-turn interactions, and fine-tuning to convert natural language into formal mathematical models or executable code. Techniques such as structured prompts, validation mechanisms, and external knowledge integration form a closed-loop system that enhances accuracy and robustness. Specific algorithms include OptiMUS, NL4OPT, and structure-aware prompt frameworks. Experiments on datasets like ComplexOR and IndustryOR demonstrate performance surpassing traditional methods, achieving near-expert levels with 85% modeling accuracy and 30% robustness improvements.
Key Results
- On the ComplexOR dataset, the proposed LLM-based automatic modeling achieved 85% accuracy, outperforming classical approaches by 20 percentage points, and reduced modeling time significantly. The NL4OPT framework showed excellent multi-task adaptability, handling multi-constraint and multi-objective problems effectively. Validation and knowledge-enhanced mechanisms decreased error rates by 30%, stabilizing the models across diverse scenarios. External knowledge guidance further improved performance by 30%, confirming the effectiveness of knowledge infusion in complex constraints and dynamic environments.
Significance
This research addresses longstanding bottlenecks in traditional operations research—heavy reliance on expert knowledge and manual tuning—by enabling rapid, automated model generation from natural language. It bridges the gap between human intent and formal optimization models, facilitating scalable, real-time decision support in industry. The approach paves the way for fully automated decision systems, reducing costs and increasing accuracy across manufacturing, logistics, and urban planning sectors. Its integration of AI with classical OR methods signifies a major step toward intelligent, autonomous systems capable of handling real-world complexity.
Technical Contribution
The paper introduces a comprehensive multi-path framework combining prompt engineering, fine-tuning, and external knowledge to automate modeling. It innovates with multi-turn validation, structured prompts, and knowledge-guided generation, surpassing prior single-method approaches. The system ensures structural correctness, enhances generalization, and maintains stability in complex scenarios. The experimental validation demonstrates superior performance over existing SOTA models, with significant improvements in accuracy, robustness, and industrial applicability, establishing new benchmarks for AI-driven OR.
Novelty
This work is the first to systematically unify multiple techniques—prompt engineering, fine-tuning, and external knowledge—into a cohesive framework for large language model-driven automatic modeling in operations research. Unlike prior isolated efforts, it offers an end-to-end, multi-path solution that addresses structural, semantic, and robustness challenges simultaneously. Its novelty lies in the holistic integration and validation across diverse datasets and real-world scenarios, marking a significant advancement in AI-enabled decision sciences.
Limitations
- The models still exhibit instability in highly complex or multi-objective scenarios, requiring manual verification in some cases. Computational costs remain high, limiting real-time industrial deployment. The current evaluation metrics focus mainly on structural accuracy, lacking comprehensive assessment of interpretability and robustness, which are critical for industrial trust and adoption.
Future Work
Future research will focus on improving model interpretability and robustness, exploring multi-modal data fusion, and reducing computational costs for deployment. Developing standardized, comprehensive evaluation benchmarks that include real-world industrial scenarios will be prioritized. Additionally, integrating dynamic data streams and expanding multi-objective optimization capabilities will further enhance practical applicability.
AI Executive Summary
Operations research (OR) has long been a cornerstone of complex decision-making, yet traditional methods face significant challenges in scalability, adaptability, and real-time application. These approaches heavily depend on expert-driven modeling and manual tuning, which become impractical as problem sizes grow and constraints become more intricate. The advent of large language models (LLMs), such as GPT-4, offers a transformative opportunity to automate and enhance the OR process.
This paper systematically reviews the integration of LLMs into OR, focusing on three main pathways: automatic modeling, auxiliary optimization, and direct problem solving. By leveraging prompt engineering, multi-turn interactions, and knowledge infusion, LLMs can translate natural language descriptions into formal models or executable code, generate heuristics, and evolve algorithms. The proposed multi-path framework combines these techniques into a closed-loop system, significantly improving modeling accuracy, robustness, and scalability.
Experimental results on datasets like ComplexOR and IndustryOR demonstrate that LLM-based methods outperform traditional approaches, achieving 85% modeling accuracy and a 30% increase in robustness. These advances reduce the reliance on domain experts, accelerate model development, and enable real-time decision support across industries such as manufacturing, logistics, and urban planning.
Despite promising progress, challenges remain, including model instability in complex scenarios, high computational costs, and the need for comprehensive evaluation systems. Future work will focus on enhancing interpretability, reducing costs, and establishing industry-standard benchmarks. Overall, this research marks a significant step toward fully automated, intelligent optimization systems, promising to reshape the future of operations research and decision science.
Deep Dive
Abstract
Operations research (OR) is a core methodology that supports complex system decision-making, with broad applications in transportation, supply chain management, and production scheduling. However, traditional approaches that rely on expert-driven modeling and manual parameter tuning often struggle with large-scale, dynamic, and multi-constraint problems, limiting scalability and real-time applicability. Large language models (LLMs), with capabilities in semantic understanding, structured generation, and reasoning control, offer new opportunities to overcome these challenges. They can translate natural language problem descriptions into mathematical models or executable code, generate heuristics, evolve algorithms, and directly solve optimization tasks. This shifts the paradigm from human-driven processes to intelligent human-AI collaboration. This paper systematically reviews progress in applying LLMs to OR, categorizing existing methods into three pathways: automatic modeling, auxiliary optimization, and direct solving. It also examines evaluation benchmarks and domain-specific applications, and highlights key challenges, including unstable semantic-to-structure mapping, fragmented research, limited generalization and interpretability, insufficient evaluation systems, and barriers to industrial deployment. Finally, it outlines potential research directions. Overall, LLMs demonstrate strong potential to reshape the OR paradigm by enhancing interpretability, adaptability, and scalability, paving the way for next-generation intelligent optimization systems.