Research On The Method Of Soft Fault Feature Extraction For PV Inverter

Solar energy is one of the most promising new energy sources since the 21 st century.Photovoltaic power generation has become a way to efficiently use solar energy.Photovoltaic inverters are important components in photovoltaic power generation systems.Once a fault occurs,the stability of system will be reduced or the system even stop working.The fault feature extraction of photovoltaic inverter is the key to improve the stability of photovoltaic power generation system.The soft fault caused by the degradation of the components and parameters is not obvious and the distinguishability is poor.The traditional feature extraction method is difficult to achieve the expected results.The study of the soft fault feature extraction method of photovoltaic inverter has practical value.The specific research content of this paper is as follows:This article selects a representative three-level midpoint clamped(NPC)inverter as the experimental object.Firstly,based on an in-depth analysis of its working principle,the types of capacitive soft faults are classified,and then a circuit model is established through Matlab/Simulink.Simulation results show that the voltage signal between the neutral points of the three-phase bridge arm corresponding to different soft fault types is used as the fault signal,and two soft fault feature extraction methods are proposed:Soft fault feature extraction based on VMD wavelet energy,when the component parameters degenerate,the frequency band energy distribution of the fault signal will fluctuate greatly.The signal frequency band feature contains fault information,and the VMD wavelet energy of the fault signal can be a fault feature.Firstly,the permutation entropy algorithm is used to optimize the number of modal components of the VMD,and the wavelet energy of the modal component of the fault signal is extracted to construct the fault eigenvectors.Then,the dimension vectors of the eigenvectors are reduced.Finally,the classification of soft faults is achieved by the support vector machine.Compared with the traditional wavelet and EMD wavelet feature extraction methods,the VMD wavelet feature extraction method has high accuracy and high speed,and it has prominent advantages in soft fault diagnosis.Soft fault feature extraction based on parameter identification,the capacitance value is the most direct indicator of the health of the capacitor.The capacitance value can be used as a soft fault feature.The statistical parameters of the signal can well reflect the energy distribution and state fluctuation of the signal time domain.Partial dimension coefficients and dimensionless coefficients are preferred as statistical parameters of the fault signal and a parameter vector is constructed.The parameter vector is used as the input of the ELM.Taking the capacitance value as the output of the ELM,the number of hidden-layer neurons in the optimal ELM is determined by analyzing the prediction error and the running time of the algorithm.The trained ELM network is used for parameter identification to acquire the capacitance value.Experiments show that the capacitance value compared with the BP algorithm,the proposed method is more applicable to the soft fault feature extraction of the inverter.