Research on the Design of Multimodal Knowledge Graph for Power Grid Infrastructure Construction

This paper systematically studies and designs a multimodal knowledge graph system suitable for power grid infrastructure construction, aiming to optimize information integration and decision support in the construction process. The system obtains real-time information of the construction site through high-precision images and sensor data, and uses the named entity recognition (NER) algorithm enhanced by the attention mechanism to accurately extract key entities and relationships related to construction from multi-source data such as images, texts, and sensors. In the system simulation stage, this paper verifies it through a large amount of real data, focusing on evaluating the performance of the NER algorithm and the effectiveness of the attention mechanism. The results show that compared with the traditional algorithm, the NER algorithm using the attention mechanism has a significant improvement in recognition accuracy, especially when processing complex scene data, its accuracy has increased by more than 15%. In addition, the multimodal knowledge graph performs well in showing the correlation between different modal data, providing a more intuitive and operational decision support tool for construction site managers.