Because of sampling errors and circuit parameter inaccuracies in an actual modular multilevel converter (MMC) model

existing fault diagnosis strategies suffer from poor robustness and difficulty in threshold selection. To solve these problems

this paper proposes a multi-IGBT open circuit fault diagnosis strategy based on capacitor voltage prediction sampling optimization. Here the capacitive voltage

output current and circulating current after segmental error optimization are used as dominant characteristics of fault diagnosis

and a one-dimensional convolutional neural network (1D-CNN) with simple structure and strong adaptability is used as a new fault location scheme to realize high-precision MMC open-circuit fault diagnosis. The proposed method can quickly and accurately locate multi-IGBT faults of the submodule at the initial stage of the capacitor voltage variation without adding additional sensors. Compared with existing diagnosis strategies based on capacitance voltage

this method significantly reduces system parameter errors and enhances the robustness of the fault diagnosis network. Finally

simulations and experiments verify the effectiveness of the proposed fault diagnosis strategy.