Open Access

REVIEW

Applications of Large Language Model in HVDC Systems: Concepts, Development, and Perspectives

Xing Wen¹, Huan Chen¹, Ning Wang^1,*, Yu Song¹, Zhuqiao Qiao², Bin Zhang¹
1 China Southern Power Grid Extra High Voltage Power Transmission Company, Guangzhou, 510663, China
2 China Southern Power Grid Extra High Voltage Power Transmission Company Kunming Bureau, Kunming, 650217, China
* Corresponding Author: Ning Wang. Email:

Energy Engineering https://doi.org/10.32604/ee.2025.073567

Received 21 September 2025; Accepted 16 December 2025; Published online 05 January 2026

Abstract

High voltage direct current (HVDC) systems play a pivotal role in long-distance, high-capacity, and cross-regional power transmission. However, their complex structure, wide-ranging impact of faults, and stringent safety requirements pose significant challenges to operational stability. Conventional model-based and data-driven methods for tasks such as text classification, fault diagnosis, and operation and maintenance support suffer from limited scalability and interpretability. Recent advances in large language model (LLM) provide new opportunities to address these issues. This paper provides a systematic review of LLM applications in HVDC systems. Firstly, it introduces the core architecture and training mechanisms of LLM. Subsequently, a comprehensive evaluation framework is established for LLM’s application in HVDC systems across four dimensions: accuracy, generalization capability, interpretability, and complexity. Secondly, this paper examines the applications of LLM in three key domains: text classification, fault diagnosis, and question answering. It further extends the discussion to additional application areas while evaluating the performance of various methodological approaches. Specifically, in text classification, LLM enhance the semantic interpretation of operation and maintenance reports. In fault diagnosis, LLM achieve accurate fault identification through multi-source data fusion and reasoning. In question answering applications, LLM provide intelligent decision support through retrieval-augmented and multimodal frameworks. Finally, from the perspectives of real-time performance, interpretability, and applicability, this paper presents three key conclusions and discusses two critical aspects. It further outlines three forward-looking research directions focusing on multimodal data fusion and real-time decision-making. The goal is to provide an up-to-date and comprehensive reference for researchers exploring the integration of LLM into HVDC system analysis, operation, and management.

---

Keywords

Large language model; high voltage direct current; text classification; fault diagnosis; question answering

---