Severe disturbance may make the frequency fall below allowable value and make power system unable to maintain a steady frequency. In this paper, a predictive method for  frequency nadir is proposed based on convolutional neural network (CNN). The measured operation data before and immediately after the disturbance is used as the input of CNN, with the frequency nadir predictive value as the output. The CNN input tensoris are constructed on a 2-D plane that is able to reflect spatial distribution characteritics of nodes operation data. The electrical distance is used to describe the spatial correlation of power system nodes, and the t-SNE dimensionality reduction algorithm is presented to map the high-dimensional distance information of nodes to the 2-D plane. The case study results show that the proposed method can predict the frequency nadir of  center of inertia after the disturbance accurately .

frequency nadir,convolutional neural network,deep learning,dynamic frequency prediction,power system