As the core component of the photovoltaic inverter, insulated gate bipolar transistor (IGBT) will not only affect the stable operation of photovoltaic inverter system in case of an open circuit fault occurrence, but also damage the system equipment as well. In view of a reduction of the number of sensors and fuse multi-scale features, a new fault diagnosis method has thus been proposed for the photovoltaic inverter based on empirical mode decomposition (EMD) and two-dimensional convolution neural network (2D-CNN). The proposed method uses EMD to extract the intrinsic mode function component of current signal and the original signal so as to form two-dimensional feature data, with the data subsequently input into the 2D-CNN model for training, thus finally realizing the open circuit fault diagnosis of IGBT. Experimental results show that this method helps to improve the accuracy of fault diagnosis, characterized with a better performance of both effectiveness and robustness in a noisy environment.