1.School of Electrical Engineering, Beijing Jiaotong University, Beijing 100044, China;2.Central China Branch of State Grid Corporation of China, Wuhan 430077, China

With the proposal of the “carbon emission peak and carbon neutrality” goal and the large-scale grid connection of renewable energy clusters, the problem of power system frequency safety is highlighted again. Therefore, by referring to the deep learning method, a two-stage multi-level early warning model for power system frequency safety based on improved residual network is proposed. Firstly, the multi-level fine division for power system frequency safety is carried out, and the multi-level early warning model for power system frequency safety is proposed and constructed. In the first stage, the proposed model uses the classification evaluator based on the improved residual network to evaluate whether the disturbed frequency will exceed early warning value for the safety, and gives the early warning level. In the second stage, the regression predictor is used to further give the risk degree of early warning samples. Finally, the improved IEEE 10-machine 39-bus system and Illinois 200-bus system are taken as examples to test the model. The results show that the model has high early warning accuracy for the safety, which is not only better than other shallow learning methods and deep learning models, but also can accurately forecast the frequency risk degree, and has good robustness and anti-noise ability.