In the field of the electric power industry, improving the entity relation model can increase the efficiency of knowledge graph completion, build a smart grid, and provide strong support for relevant regulatory decisions. Therefore, we propose a deep learning-based entity relation discovery model called BERT-SRR for the electric power domain. This model includes constructing an electric power domain knowledge base using professional domain knowledge and adding entities and their relationships to it. It utilizes a BERT model trained incrementally to extract Chinese text features and conducts sequence labeling learning with entity and relationship information from the knowledge graph. Furthermore, combining a stacked convolutional neural network and a student reordering network enables high-accuracy knowledge graph completion and prediction of entity relationships. The comparative experimental results demonstrate that the proposed model outperforms the benchmark model, showing a significant improvement in F1 score ranging from 3.66% to 9.78%.