Novel data-driven open-circuit fault diagnosis method for modular multilevel converter submodules based on optimized deep learning

As the proportion of clean energy continues to increase, low carbon energy systems will be a significant way to achieve the goal of carbon neutrality. Therefore, the reliability of modular multilevel converters (MMCs) is particularly significant. However, conventional open-circuit fault diagnosis (OCFD) methods usually have a limited localization speed or are difficult to achieve in practical engineering. Therefore, a fast and simpled OCFD approach for MMC SMs based on an optimized deep learning is proposed in this article. In this approach, data on the of submodule capacitance voltages are input into a trained WOA-DKELM model without the manually settings. The problems of randomness in the regularization coefficient C and the kernel parameters K can be solved by DKELM with WOA optimization, which has a strong generalization capability and higher prognostic accuracy. The effectiveness of the proposed approach is verified by experiment results. This approach achieves an average identification probability of 0.96 within 20 ms of the fault.