To enhance the fault diagnosis of the switch tube of the modular multilevel converter (MMC) submodule, a Sand Cat swarm optimization (SCSO) algorithm is improved. This improved SCSO (ISCSO) algorithm is employed to optimize the fault diagnosis of an extreme learning machine (ELM). Cubic chaotic mapping, a spiral search method, and a sparrow alert mechanism are used to improve the three stages of sand cat search, so as to enhance the convergence speed and search capability of the algorithm. An MMC model is developed on the MATLAB/SIMULINK platform, where the bridge arm circulation is used as the input when a fault occurs in the submodule. By comparing the fault diagnosis performance of ISSO-ELM against ELM optimized by other algorithms, the results show that the proposed method can effectively identify submodule faults. It shows feasibility and superiority in MMC fault diagnosis, offering better fault diagnosis performance.