A Method of Frequency Features Prediction of Post-disturbance Power System Based on XGBoost Algorithm

As the frequency stability problem of modern power systems is prominent, rapidly and accurately predicting the frequency features of post-disturbance power system is of great significance for ensuring the stable operation of the power system. In order to achieve rapid prediction of frequency stability problems, this paper proposes a method for predicting frequency features based on XGBoost (Extreme Gradient Boosting) algorithm of post-disturbance power system. First, the maximum rate-of-change of frequency (RoCoF), frequency nadir $(f_{nadir})$, and quasi-steady state frequency $(f_{\mathrm{s}s})$ are used as output indicators to characterize the frequency features. Using the data before and after the disturbance to construct an initial feature set for frequency prediction. And through the Pearson correlation coefficient method for key feature screening. Then, an advanced machine learning method, the XGBoost algorithm, is introduced to establish nonlinear mapping relationships between input features and frequency feature indicators. Simulation analysis of the proposed method is performed in the New England 10-machine 39-bus system. The validity and superiority of the proposed method is verified by comparing it with two shallow machine learning algorithms, BP (Back Propagation) neural network and SVR (Support Vector Regression), and one deep learning algorithm, CNN (Convolutional Neural Networks).