Modular multilevel converter (MMC) has been widely used due to its many advantages, and a large number of sub-modules have become potential fault points. Improving the safety and reliability of power equipment is particularly important. To quickly and accurately diagnose open circuit faults in submodules, this paper proposes a fault diagnosis method based on optimized probabilistic neural networks. Firstly, analyze the characteristics of the submodule open circuit fault and select the submodule capacitor voltage as the key diagnostic feature data. Then, based on the characteristics of probabilistic neural networks (PNN) models, the Sparrow Search Algorithm (SSA) is proposed to find the optimal smoothing parameters, and based on this optimized model, a submodule open circuit fault diagnosis strategy is constructed, which can achieve fast and high-precision fault diagnosis without increasing any hardware costs. Finally, the effectiveness of the proposed submodule open circuit fault diagnosis strategy was verified through experiments.