A CNN-Based Post-Contingency Transient Stability Prediction Using Transfer Learning

One of the main problems in the utilization of machine learning-based classifiers for transient stability prediction is their long training times with the comprehensive large-sized datasets. Using a small-sized dataset to decrease the training time is not reasonable since the dataset should be representative of all types of severe faults. In this paper, a novel methodology based on transfer learning is proposed for real-time post-contingency transient stability prediction to overcome the difficulties about the long training times of these classifiers. In the proposed method, first, a small dataset which contains only the three-phase fault contingencies for various operating points is selected to train a convolutional neural network (CNN) classifier, and then, an additional dataset which involves two-phase-to-ground fault scenarios is used to update the trained CNN using the transfer learning approach instead of retraining the model from ground up. To demonstrate the efficiency of the proposed method, it is applied to the 127-bus test system.