Identifying Overlapping Successive Events Using a Shallow Convolutional Neural Network

Abstract

Real-time identification of successive events in power systems is crucial to avoid cascading failures. Existing identification methods are mainly designed for single events and may not accurately identify a subsequent event that occurs when the system is under going the disturbance of a previous event. Since overlapping successive events do not frequently happen in power systems, insufficient multievent data sets exist for training. We develop a data-driven event identification method that can accurately identify the types of overlapping events. Our approach only requires a small number of recorded phasor measurement unit data of single events to train a two-layer convolutional neural network (CNN) classifier offline. We extract the dominant eigenvalues and singular values as features instead of training on time series directly. That reduces the required number of training data sets and enhances the robustness to measurement inaccuracy. In real time, our method first predicts and subtracts the impact of previous events. It then extracts the dominant features from the residual measurements and applies the classifier. We evaluate the method on simulated events in the IEEE 68-bus power system. Our classifier is demonstrated to be more accurate and stable than a direct application of CNN on time series. The robustness of the proposed method to the delay in event detection and noise is validated.