Abstract: At present, the node carbon emission factor calculation method based on carbon emission flow theory can effectively realize the fine carbon measurement of electricity. However, the existing methods are mainly based on measurement data or power flow results for post-evaluation calculation, which can not accurately give the power carbon emission factors of different load nodes in the future period, and it isn't easy to provide direct guidance signals for users to implement low-carbon regulation. Because of this situation, this paper proposes a prediction method for the carbon emission factors of power grid user-side nodes based on a variant graph neural network. Considering the coupling relationship between energy flow and carbon emission flow between different nodes in the power grid, based on the topology structure of the power grid, an equivalent transformation method of the adjacency matrix of variant graph neural network considering eliminating passive no-load nodes is proposed. The input-output characteristics of the variant graph neural network for carbon emission factor prediction on the user side are constructed. Based on the trained model, the user-side carbon emission factor can be predicted quickly and accurately only by using the source-load forecast data, without accurate grid parameters, power flow calculation, and carbon emission flow calculation, effectively reducing the calculation cost and improving the application flexibility. Finally, the effectiveness of the proposed algorithm was validated under the IEEE39 and IEEE118 standard test systems and a regional power grid in A city of China Southern Power Grid Company, with an average nodal carbon emission factor prediction error rate of 1.76%, 2.21% and 1.49%, respectively.