In view of the large number of virtual terminals to be connected in smart substations, traditional connection methods often suffer from low identification efficiency, large verification workload, and inaccurate connection results. To address these issues, an automatic virtual terminal connection method based on the sentence-masked language model as correction bidirectional encoder representations from transformer (Sentence-MacBERT) model is proposed. First, the key information required for the automatic virtual terminal connection is extracted and preprocessed. Second, the Sentence-MacBERT virtual terminal automatic connection model is constructed and trained to obtain the optimal model. Finally, the preprocessed short addresses and Chinese descriptions are separately input into the model to obtain the comprehensive sentence embeddings, and cosine similarity matching is carried out to complete the automatic connection of virtual terminals in smart substations. The results show that, compared with traditional virtual terminal automatic connection methods, the proposed method achieves higher connection efficiency, with an accuracy of 94.38%, enabling accurate virtual terminal connection.