Abstract: Abstract : The digital twin power grid digitizes the physical power grid through virtual-real mapping. The current virtual real mapping process of power grid facilities, however, faces challenges in terms of limited accuracy in reconstructing static models and poor real-time dynamic response. The traditional manual establishment of static models is inefficient and challenging, making it unsuitable for construction of model in complex power grid. The traditional cloud computing model fails to meet the diverse demands of numerous randomly transmission services during model reconstruction and response processes. Therefore, in this paper we propose a new edge network resource allocation method for virtual-real mapping of digital twin power grid facilities. Firstly, a feature matching method of double threshold nearest neighbor ratio is proposed based on the fusion of Shi Tomasi and Surf algorithms. Additionally, the RANSAC algorithm is further integrated to enhance the accuracy of camera pose estimation, thereby improving the accuracy of the three-dimensional reconstruction model of power grid facilities. The aforementioned foundation serves as the basis for designing utility functions for different transmission delay and accuracy requirements, while also presents a method for subchannel and power resource allocation so as to achieve on-demand allocation of communication resources. In addition, a cloud-edge-end data interaction architecture is built to support the above process. Compared with Shen algorithm and channel average allocation algorithm, our algorithm can improve data transmission timeliness by 9.03% and 31.30% respectively, while ensuring data transmission accuracy. Moreover, the algorithm effectively facilitates the accurate and efficient virtual-real mapping process of power grid facilities.