Deep Learning-Based Text Entity Recognition Method for Distribution Network Operation and Maintenance

Abstract

The distribution network has accumulated a large amount of text data in long-term operation and maintenance, dispatching and other operations, and a large amount of entity data of the distribution network is contained in these text data. However, due to the large amount of text data and most of the texts are proper nouns, there is no one entity recognition algorithm suitable for this field, leading to a series of defects caused by poor entity recognition in the process of constructing knowledge graphs. To this end, this paper proposes a new deep learning-based entity recognition method (D-BERT+Bi-LSTM+GAM+CRF) for distribution networks. The method uses the distribution network operation and maintenance scheduling and other protocol documents to train the model. Text entity recognition for distribution network operation and maintenance dispatching operation work orders in real scenarios. Experiments showed that the F1_score, Precision, and recall of this scheme reached 81.8%,80.4%, and 83.2%, respectively, which improved the F1_score by 1.3%-26.6% compared with the conventional scheme. It effectively guides the construction of knowledge graphs.