Critical Clearing Time Prediction of Power System Fault Based on Machine Learning

With the expanding of the scale of power grids, the uncertainty and complexity of power system operation are increasing. Critical clearing time (CCT) is the key to transient stability assessment (TSA) in power system operation, security defense, maintenance and other scenarios. The application of machine learning to TSA has become a research hotspot. In this paper, a method of obtaining the CCT by machine learning is proposed. First, time domain simulation under electrical fault is utilized to obtain samples for machine learning. Second, pattern features of transient stability are obtained from the samples, the maximum information coefficient method is used to reduce feature dimension, and the Box-Cox transformation is used to improve data distribution characteristics. Further, based on the XGBoost, Random Forest algorithms, a stacking model with SVR as the meta learner is established to improve the performance of CCT prediction. In the end, this proposed method is verified via the CEPRI 8-machine 36-bus system. The results prove that this method can provide accurate CCT predictions, and it is also robust to load changes.