Research on automatic virtual circuit verification technology in smart substation based on improved bidirectional encoder representations from transformers
CHEN Wenhan, ZHAO Zhanglei, GAO Bingwei, CHEN Gang, KONG Fanfang
State Grid Wenzhou Electrical Power Company, Wenzhou, Zhejiang 325000
Abstract：The correctness of virtual circuit connections in smart substations is crucial to the proper operation of the station's relay protection equipment. Traditional virtual circuit verification relies on manual item-by-item comparison, resulting in poor results and a lengthy timeframe. This paper proposes an automatic virtual circuit verification technology for smart substations based on the improved bidirectional encoder representations from transformers (BERT) model. Verification accuracy is first improved through sentence cleaning. Then, the model's understanding of Chinese vocabulary is enhanced by masking Chinese words, replacing the BERT model's single-character masking approach. To further enhance recognition performance, this paper combines the BERT model with a lexical masking model and a Siamese network to obtain an improved BERT model. The virtual terminal description text is divided into two semantically homonymous subsets as input to the improved BERT model. The correctness of the virtual terminal connection is verified by calculating the cosine similarity of the text at both ends of the virtual terminal. Test and application results demonstrate that this method accurately identifies virtual terminal matching and automatically verifies virtual circuits, effectively reducing the workload of traditional virtual circuit verification and improving its accuracy.
2026, Vol. 27 
