Recognition Method of Voltage Sag Sources Based on Deep Learning Models' Fusion

The recognition of the voltage sag sources is the basis for formulating a voltage sag governance plan and clarifying the responsibility for the accident. The causes of voltage sag can be divided into single voltage sag sources and composite voltage sag sources, the complexity of grid equipment and the regionalization of power consumption patterns present new challenges to the traditional recognition method of voltage sag sources based on physical characteristics. This paper proposed a voltage sag sources' recognition method based on model fusion, using the convolutional neural network (CNN) in the deep learning algorithm to obtain the timing characteristics and spatial characteristics of voltage sag signals, the deep confidence network (DBN) was used to replace the full-connected layers in CNN for purifying high-dimensional features and acting as a classifier, in order to enhance the network's multi-tag classification capabilities. Utilize simulated and noise-added data to iteratively train and test the network repeatedly, high recognition accuracy and anti-noise performance of the fusion model have been verified. Compared with the traditional voltage sag sources' recognition method, the generated model has good generalization ability and can be effectively applied in practical projects. © 2019 Chin. Soc. for Elec. Eng.