부산대학교 도서관

The construction of a new power system puts forward higher requirements for dispatching human-computer interaction and professional language understanding. Aiming at the problem of lack of effective identification methods for power grid dispatching business intentions, this paper proposes a language intent recognition method for power grid dispatching based on the fusion model of bidirectional encoder representations from transformers (BERT) and text convolutional neural networks (TextCNN). First, the text features of the power grid dispatching language are calculated to generate the dispatching language word vector based on the dynamic word vector pre-trained by BERT. Then, the word vector of the power grid dispatching language is used as the input of TextCNN to implement feature encoding and classify dispatching language intent. Finally, through the verification of the dispatching professional language of a control center, the proposed power grid dispatching intention identification method has stronger identification ability and generalization ability compared with other methods.