Based on the deep learning theory

a novel method for sub-modular (SM) open-circuit fault diagnosis of modular five-level inverter (MFLI) is presented based on the stacked sparse auto-encoder (SSAE). The SM open-circuit fault detection and location problem of MFLI is converted into a classification problem. Firstly

the capacitor voltage signals of all SMs in the MFLI circuit are combined into a 24-channel signal. Then

by moving window along the 24-channel signal with the sliding window

a set of signal segments are acquired which are flattened into vectors and used as SSAE's input subsequently to realize the unsupervised feature learning layer by layer. The deep feature with concise expression of original fault dataset is established and connected to the Softmax classifier to output the fault diagnostic result. In addition

in order to enhance the anti-noise performance of the proposed method

the SSAE is trained by adding Gauss noise to improve the robustness of feature expression. The results show that the proposed fault diagnosis method has the high robustness and versatility with the average accuracy of 98.09% and the average fault diagnosis time of 31.47 ms.