Data-Driven Transient Frequency Stability Modeling and Analysis Based on ResNet and SHAP

In recent years, China has been actively constructing new-type power systems to optimize the nation-wide energy structure and improve energy efficiency. As a result, ultra-high voltage(UHV) transmission has become increasingly popular, and the proportion of renewable energy in power systems has gradually increased. These phenomena have resulted in unprecedentedly complicated power grids with hybrid AC/DC interconnection structures, increased system uncertainty, and difficulty in maintaining a balance between power generation and load. In this context, system transient frequency stability is greatly challenged. To address this challenge, this paper proposes a method to model and analyze frequency stability characteristics of receiving-end power grids based on interpretable deep learning. First, the PSASP simulation software is employed to set up a series of representative operating conditions and transient faults to obtain sample data. Next, based on the simulated sample data, a residual convolutional neural network (ResNet) is utilized to predict the stability of frequency after a fault occurs in the power grid under the influence of AC-DC hybrid structure. Finally, the network prediction results are analyzed using the shapley additive explanations method (SHAP). The proposed method is verified with a realistic receiving-end system, which verifies its effectiveness and provides important guidance for follow-up frequency stability control measures.