The ultra high voltage substation engineering is the key to the intelligent upgrading of the power grid, and its digital 3D design is the foundation for achieving intelligent project management. Traditional 3D design methods have low efficiency and urgently require breakthrough technological support. This study proposes a three-dimensional digital design method for ultra high voltage substation engineering based on convolutional neural networks (CNN). By training CNN models, three-dimensional spatial geometric topology information is automatically extracted from two-dimensional construction drawings and related documents, greatly improving the efficiency and accuracy of three-dimensional model construction. At the same time, this method also realizes functions such as automatic calculation of engineering quantities, collision detection, simulation analysis of 3D model data, and realizes the full process digital management of ultra high voltage substation engineering. The research results indicate that the CNN model can automatically generate high-quality 3D digital twin models at a faster speed, laying a solid foundation for subsequent design optimization, simulation analysis, and visualization presentation.