A Power System Transient Stability Assessment Model Based on Stacked Denoising Autoencoder

Since the grid data is not always "perfect", there are often cases where data is missing or contains noise. In order to solve the above problems and further improve the generalization ability of the power system transient stability assessment model, we introduce Denoising Autoencoder (DAE) structure and propose a model based on Stacked Denoising Autoencoder (SDAE). Firstly, construct the original input features that can reflect the transient stability characteristics of the power system. And then use the input features layer-by-layer unsupervised learning to get the encoder of DAE. Stacking all the trained encoders constitutes a SDAE as an advanced feature extraction model. Next, using the advanced features and labels has a supervised training classifier to obtain a complete transient stability assessment model. The model proposed in this paper has been tested on 10-machine New-England Power System. The simulation results show that the proposed model is robust to the "imperfection" of the input data, and the generalization ability is stronger.