Data-Driven Risk Identification Method for Low-Frequency Oscillations in New Power Systems

With the increasing scale of new energy grid integration, the risk of low-frequency oscillation in the power system has increased, which seriously affect system safety and stability. It is urgent to identify the risk of low-frequency oscillations through steady-state operating features. This article first analyzes the features that affect low-frequency oscillations and constructs a low-frequency oscillation dataset using transient simulations. Secondly, feature selection was performed using the random forest algorithm, and a low-frequency oscillation risk identification model for GA-CNN was proposed. Thirdly, by combining Pearson correlation coefficient and RF algorithm to eliminate redundant features and screen important features, a low-frequency oscillation frequency recognition model based on GBRT was proposed, and hyperparameter optimization was performed using grid search. Finally, the effectiveness of the proposed method was verified by ablation experiments and comparative experiments using low-frequency oscillation datasets under different operating conditions.