부산대학교 도서관

With the increasing complexity of power grids and the enhanced functionality of electrical equipment, particularly transformers, which are core components of power systems, any malfunction can significantly impact the safe and stable operation of the grid. Traditional fault diagnosis methods mainly rely on experience, which not only lacks the necessary explanatory power but also responds slowly to fault changes, making it difficult to solve problems timely and accurately. To address these issues, this paper establishes a knowledge base focused on transformer faults and proposes a named entity recognition method. This method, based on the ALBERT-BiLSTM-CRF algorithm, efficiently extracts key entities from unstructured data. Achieving an F1 score of 0.9517 in processing transformer fault text data, the simulation results not only validate the efficiency of our method but also demonstrate its superiority over other traditional deep learning algorithms.