Large Language Models for Power System Applications: A Comprehensive Literature Survey

This comprehensive literature review examines the emerging applications of Large Language Models (LLMs) in power system engineering. Through a systematic analysis of recent research published between 2020 and 2025, we explore how LLMs are being integrated into various aspects of power system operations, planning, and management. The review covers key application areas including fault diagnosis, load forecasting, cybersecurity, control and optimization, system planning, simulation, and knowledge management. Our findings indicate that while LLMs show promising potential in enhancing power system operations through their advanced natural language processing and reasoning capabilities, significant challenges remain in their practical implementation. These challenges include limited domain-specific training data, concerns about reliability and safety in critical infrastructure, and the need for enhanced explainability. The review also highlights emerging trends such as the development of power system-specific LLMs and hybrid approaches combining LLMs with traditional power engineering methods. We identify crucial research directions for advancing the field, including the development of specialized architectures, improved security frameworks, and enhanced integration with existing power system tools. This survey provides power system researchers and practitioners with a comprehensive overview of the current state of LLM applications in the field and outlines future pathways for research and development.

💡 Research Summary

This comprehensive literature survey provides a detailed examination of the emerging integration of Large Language Models (LLMs) into power system engineering, covering research published between 2020 and 2025. The paper begins by contextualizing the need for advanced tools in modern power grids, which are growing increasingly complex due to renewable energy integration and smart grid technologies. It identifies LLMs, with their profound natural language processing, reasoning, and knowledge retrieval capabilities, as a promising technological avenue to address these challenges.

The paper first outlines the fundamentals of LLMs, explaining the core Transformer architecture, self-attention mechanisms, and key concepts like tokenization and embedding. It notes the relevance of models like GPT-4 and Llama, and highlights the potential utility of few-shot and zero-shot learning in a domain where labeled data can be scarce.

The core of the survey is a systematic exploration of LLM applications across critical power system domains:

1. Fault Diagnosis & Anomaly Detection: Studies show LLMs, combined with advanced prompt engineering, can improve the accuracy and explainability of fault diagnosis. Hybrid models using optimized LLMs have also demonstrated success in time-series forecasting for fault prediction, such as forecasting leakage current in insulators.
2. Load Forecasting & Demand Response: Research incorporates NLP-processed external data (e.g., news sentiment) into forecasting models like LSTMs to improve accuracy. Transformer-based architectures are also being specifically adapted for electrical load time-series data.
3. Cybersecurity: The paper presents a dual perspective. It analyzes significant security threats introduced by LLMs, including privacy invasion, data poisoning, and semantic attacks. Conversely, it explores LLMs’ potential for enhancing security by analyzing network traffic and logs for threat detection.
4. Control & Optimization: Several frameworks are proposed, such as SafePowerGraph-LLM, which uses graph and tabular representations to solve Optimal Power Flow (OPF) problems. Other work includes the development of GAIA, a power dispatch-specific LLM, and demonstrations of LLM agents for real-time system control.
5. System Planning & Scheduling: LLMs are investigated for scenario generation in grid planning and for user-centric scheduling, where agent architectures convert user voice requests into resource allocation vectors.
6. Simulation & Modeling: A modular framework is proposed to equip LLMs with the ability to perform power system simulations using unfamiliar software tools, positioning them as potential research assistants.
7. Knowledge Management & Decision Support: LLMs are identified as valuable tools for extracting information from vast technical documentation, providing intuitive data presentations to operators, and building domain-specific terminologies for intelligent design systems.

Despite the promising potential, the survey clearly identifies major challenges hindering practical implementation. These include a scarcity of high-quality, domain-specific training data; fundamental concerns about the reliability and safety of using LLMs in safety-critical, real-time control loops; and the “black-box” nature of LLMs, which conflicts with the need for explainability in engineering decisions.

The paper concludes by highlighting emerging trends, such as the development of power system-specific LLMs and hybrid approaches that combine LLMs with traditional power engineering methods. It outlines crucial future research directions, emphasizing the need for specialized architectures, robust security frameworks, and seamless integration pathways with existing power system software tools. This survey serves as a foundational reference for researchers and practitioners, mapping the current landscape and charting a course for responsible and effective innovation in applying LLMs to power systems.