CNN with squeeze and excitation attention module for power system transient stability assessment

An approach based on Convolutional Neural Networks (CNNs) with a squeeze and excitation attention mechanism (SEAM) is investigated in this paper to assess the transient stability of a sample power system. In general, attention mechanisms in CNNs are intended to assist the selective focus of the model to increase their performance capabilities for different applications. In particular, the incorporation of the SEAM is considered here as a mean to automatically and explicitly model the importance of channel interdependencies in a given set of feature maps, which is accomplished by attention weights that modulate the influence of each channel. By collecting representative input-output examples from extensive simulations of the electric grid of Baja California Sur (BCS) in Mexico under different operational points, the response of the proposed approach to assess the transient stability of the grid is investigated on a sample dataset. Simulation results demonstrate that the CNN model with SEAM is able to provide enhanced learning performance and superior prediction response, as compared to the same CNN structure with no SEAM.