Transient Stability Assessment of Power Systems Using Cost-sensitive Deep Learning Approach

This paper presents a deep learning approach with cost-sensitive loss function for transient stability assessment (TSA). Firstly, multiple generator variables after disturbance are utilized as the original inputs. Secondly, the convolutional neural network (CNN) is applied for automatic feature extraction and regression model building for TSA. The generated TSA model can provide more accurate transient stability index than the fully-connected neural network and support vector regression model. Moreover, a cost-sensitive loss function is proposed to reduce the misdetection of instability. Case studies are conducted on the New England 10-machine 39-bus system to verify the effectiveness of the proposed approach.