Open circuit fault prediction method of energy storage converter based on MFCC characteristics

Converter open circuit fault prediction plays an important role in improving the intelligent operation and maintenance of energy storage systems. Consider the existing methods for non-intrusive identification of the IGBT open circuit faults in power conversion system (PCS), signal feature extraction occurs difficulties, explosion of data dimensions, and instability of the threshold judgment interval. An open-circuit fault prediction method for energy storage converters based on mel frequency cepstral coefficient (MFCC) features is proposed to support the normal operation and maintenance of PCS. First of all, to communicate the three-phase current on the side is the input signal, and the MFCC fault characteristic data set is constructed by analyzing the energy distribution and envelope characteristics of the signal spectrum in different frequency intervals, and then combined with the nuclear principal component analysis to realize the nonlinear fault under charging and discharging conditions Feature dimensionality reduction screening; secondly, taking the low-dimensional fault feature set as input, build a fault state prediction model based on Bayesian optimization algorithm and one-dimensional convolutional neural network (CNN-1D); Finally, taking the simulation data of different fault states under different working conditions as examples, it is verified that the proposed method has good robustness and accuracy even in a complex noise environment, and has better advantages compared with the existing methods.