Digital twin technology, as an important digitization tool, has become a new approach for power system transformation and upgrading. In this paper, a new power system modeling method and framework driven by data-physics fusion is proposed with digital twin technology as the core. Specific research is carried out for the power system frequency response prediction problem, the improved SFR model is used as the physical model, and the two-branch network model (TBNAM) based on attention mechanism is selected as the data-driven model, which realizes the construction of the data-physical fusion model TBNAM-ISFR. The simulation results show that the TBNAM-ISFR model can be effectively applied to the electromechanical transient simulation of the power system driven by the data-physical fusion modeling, and the average absolute errors of the power angle and voltage magnitude under each scenario do not exceed 1.40% and 0.22%, respectively, with good generalization ability to meet the computational requirements of various scenarios. Compared with other comparative models, the TBNAM-ISFR model has the smallest frequency response prediction error value in different power systems, while the prediction time is only 0.21ms, which can simultaneously meet the requirements of accuracy and timeliness in online applications.
