ABSTRACT In recent years, smart distribution networks have developed rapidly. However, complex electrical equipment and multisource monitoring data present great challenges for efficient fault detection in distribution networks. Accordingly, this study designs a multistage fault diagnosis framework based on a modified convolutional neural network (MCNN). First, an adaptive multiresolution S‐transform (MST) model is applied to detect the initiation and recovery times of feeder line faults efficiently. Then, feeder line fault waveforms are converted into 2D images on the basis of the results of MST and equal‐interval sampling. Next, a convolutional neural network (CNN) combined with a parallel network is designed as a robust fault classifier. The structure of the classifier model can help enhance accuracy, while the modified activation function can achieve fast convergence. Finally, simulation data obtained from an IEEE model and field data collected from a city power system are used to validate the effectiveness and practicality of the proposed MCNN model. The average 10‐fold cross‐validation results of the fault diagnosis model based on MCNN are better than those of the CNN model in terms of related indicators. Meanwhile, the average 10‐fold cross‐validation accuracy of the proposed classifier based on simulation and field data is 97.3% and 95.6%, respectively.
Xiao et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: