Aiming at the problem that machine learning methods are difficult to achieve high-precision online diagnosis for open-circuit fault (OCF) of power electronic converters

we put forward a fast and effective online diagnosis method for OCF of power electronic converters. Firstly

a lightweight layer is introduced into a conventional convolutional neural network in the proposed method to perform denoising

compression

and multi-scale feature extraction from the output current of a power electronic converter. Secondly

an attention mechanism is used to adaptively adjust the weights of the physical loss function in the physical information network in order to construct a high-precision fault diagnosis model. Wherein

the physical loss function is constructed by introducing the total harmonic distortion physical quantity. Finally

an online diagnosis platform is developed to validate the proposed method for the four-level converter. The results demonstrate that the proposed method achieves a diagnostic accuracy higher than 99.26%

a maximum diagnostic error below 5%

and an online diagnostic time of approximately 13 ms.