With advances in surveying technologies and the increasing demand for high-precision 3D spatial information, traditional methods fail to meet the modeling requirements of modern power grid infrastructure. This paper proposes an innovative modeling approach, pioneering the integration of Self-Transformer, Point Transformer, and Cross-Transformer deep learning models for the fusion of high-precision point cloud and spatial vector data, tailored to the complex spatial structures and multi-scale characteristics of power grids. The Self-Transformer captures global context to model long-range dependencies among distant components, the Point Transformer extracts fine-grained local geometric features for intricate structures, and the Cross-Transformer fuses multi-scale features to ensure precise segmentation of both large and small components. A customized data fusion strategy, leveraging an improved Iterative Closest Point algorithm and semantic labeling, generates geometrically accurate and semantically enriched 3D models. Experimental results demonstrate significant improvements in model accuracy, quality, and semantic representation, providing robust support for digital twins, real-time monitoring, and predictive maintenance. This approach lays the foundation for intelligent power grid management and efficient operations, showcasing potential in critical infrastructure modeling.