Transient Rotor Angle and Voltage Stability Discrimination Based on Deep Convolutional Neural Network with Multiple Inputs

Abstract

Artificial intelligence-based methods have been widely used in power system transient rotor angle or voltage stability analysis. In this study, the deep convolutional neural network with multiple inputs is utilized to train a model for transient rotor angle and voltage stability discrimination. Firstly, the rotor angle and voltage trajectories are placed into the convolutional layers, pooling layers and one fully connected layer respectively. Then, the outputs are combined and put into another fully connected layer to construct the discriminator. The case studies of the 8-machine 36-bus system illustrate the effectiveness of the proposed method. Moreover, the visualized analysis of deep convolutional neural network is provided by the class activation mapping. Based on the visual analysis, the input regions which are important for discrimination can be visualized, which makes the convolutional neural network can be transparent for practical application.