Intelligent Fault Diagnosis for Overhead Lines with Covered Conductors: Using Large Language Model

Fault diagnosis of partial discharge (PD) is crucial for the protection of overhead lines with covered conductors. Facing the challenge of identifying PDs that may have diverse fault patterns from background noise interferences, a novel intelligent fault diagnosis utilizing the large language model (LLM) is developed. To effectively apply LLM to PD diagnosis, the domain knowledge-based prompts are designed by incorporating the specific domain information, PD detection task description, and measurement data information. To further improve the capability of LLM reasoning antenna signals, a signal reprogramming method is adopted to align the modalities of the measured signals and natural language. Finally, an output projection is constructed to identify PD by taking in the features learned from the LLM, whose backbone model remains intact during the learning process. Experimental results validate the efficiency and effectiveness of the developed method.