Fault Diagnosis and Location Approach of Modular Multilevel Converter Based on Dual One-Dimensional Convolutional Neural Network

Abstract

One dimensional convolutional neural network (1-D CNN) has a high identification rate in the application of MMC fault diagnosis and location. Due to the large number of (SMs) in MMC and a large amount of redundant data in feature extraction, 1-D CNN based on the SMs capacitor voltages has the disadvantages of large amount of calculation and low accuracy in the process of classification and recognition. In this paper, a dual 1-D CNN fault diagnosis and recognition method based on the sub units (SUs) voltage and the bridge arm current is proposed. The number of features is reduced from 6N to N+6, and the fault diagnosis and recognition of the sub unit voltages and the bridge arm currents are carried out respectively. Finally, the characteristics obtained from the sampling of sub unit voltages and the bridge arm currents are classified and identified. The simulated results based on an MMC model with four SMs per arm show that the presented fault diagnosis and location method based on the dual 1-D CNN has a higher accuracy, which verifies the propositions of this paper.