Condition Monitoring of Capacitor Aging in MMC Based on Improved Convolutional Neural Networks

This article introduces a new fault detection method by integrating Convolutional Neural Networks (CNN) and Gated Recursive Units (GRUs) for sub-module capacitors in Modular Multilevel Converters (MMCs). Considering the aging fault characteristics of these capacitors, the proposed method extracts feature from the sub-module capacitor voltages and their associated differential voltage waveforms, which are used as inputs to the proposed model. To improve the selection of hyper-parameters of the CNN-GRU neural network, the Whale Optimization Algorithm (WOA) is utilized, then the aging condition of sub-module capacitors can be evaluated. Simulation results show that the proposed method is valid in monitoring capacitor aging in the MMC sub-module, with an overall accuracy of more than 98.2% in detecting and identifying capacitor aging.