A recursive convolutional neural network based on DeBERTa and adaptive pooling for calculating semantic similarity of power grid data

At present, research on duplicate data detection in power grid measurement data usually only uses traditional network models or detection methods, and rarely incorporates the concept of machine learning to study. This article starts from the recursive convolutional neural network and adds the high-performance pertained language model DeBERTa to establish a duplicate data detection model, and analyzes it. In the pooling layer module, an adaptive pooling method is used to replace the traditional max pooling module, and an attention layer is creatively added to enhance the focus on key data information, assign weights to different feature vectors, and then perform duplicate detection of data information. Finally, the effectiveness of the model proposed in this paper is validated through case study.