Fault diagnosis of shipboard medium-voltage DC power system based on machine learning

Abstract

This study proposed a fault diagnosis method of a shipboard medium-voltage DC (MVDC) power system based on Noise-Assisted Multivariate Empirical Mode Decomposition (NA-MEMD) and Multilevel Iterative – LightGBM (MI-LightGBM), which overcomes the limitations of the existing fault diagnosis methods in this regard, such as relies heavily on the relay or slow training process. MI-LightGBM is proposed to solve the problem of unbalanced training samples caused by the difficulty in obtaining fault samples in practical engineering. First, NA-MEMD was adopted to pre-process the voltage signals, which were decomposed into a set of components called Intrinsic Mode Functions (IMFs) according to the local characteristic time scales of the original signals. The energy moment of each order IMF was calculated as fault feature vector to train the MI-LightGBM model, which led to the development of a high-precision fault classifier. A model of a shipboard MVDC power system was established using the AppSIM Real-Time Simulator. Simulations were performed on earth fault and short-circuit fault at the generator output and DC cable. Compared with the existing fault diagnosis methods, the proposed method is simple to use and save more than half of the training time while maintaining high diagnostic performance, which is more suitable for engineering applications.