Abstract: The large-scale integration of new energy into the grid reduces the inertia level of the power grid and increases the risk of transient frequency instability. However, existing frequency prediction methods cannot simultaneously consider the speed and accuracy of online frequency prediction for new energy grids. This article proposes an online prediction method for transient frequency response based on wide area measurement system (WAMS) technology, which integrates WAMS data model with physical model. This method first constructs a frequency dynamic response analysis model for the frequency modulation characteristics of new energy, and uses WAMS real-time data to obtain key characteristics of frequency dynamic response; Then, based on the Transformer model, the key features of frequency response are embedded as physical knowledge into the input feature quantities of the prediction model, and used for model training to achieve rapid prediction of transient frequency response by integrating WAMS data with the physical model; Finally, the speed and accuracy of the proposed transient frequency online prediction method were verified through the analysis of an improved IEEE39 node system example.