Open-Circuit Fault Diagnosis for Grid-tied T-Type Inverters Using an LSTM Autoencoder

Grid-tied T-type inverters are widely used in photovoltaic power (PV), Vehicle-to-Grid (V2G) and other power conversion applications. However, their transistors are fragile and prone to open-circuit faults that can lead to power loss or system crashes. An open-circuit fault diagnosis method for grid-tied T-type converters is proposed based on the long short-term memory (LSTM) autoencoder. At first, the autoencoder extracts latent feature vectors automatically from the three-phase current series, and a low-dimensional feature is obtained. Then, the feature vectors are represented as the multivariate Gaussian distribution, and their distribution features are calculated. Finally, in the stage of online diagnosis, the latent features of real-time signals are extracted by only the encoder and used to measure the latent distance between the signals and the fault categories to realize fault diagnosis. The distance measurement ensures the accuracy of diagnosis instead of directly outputting fault categories by the learning model. Experiment results verify the effectiveness of the proposed method.