The Research On Methods Of Fault Diagnosis And Feature Extraction In Cascaded PV Inverter

As an important component of photovoltaic power generation,photovoltaic inverters are mainly responsible for the role of electrical energy conversion,and are the main equipment for converting photovoltaic DC to AC.Among them,cascaded inverters gradually occupy an important position in new-generation photovoltaic inverters due to their many advantages such as multi-level,modularity,easy expansion,and large capacity and so on.Statistical results show that during long-term operation of the inverter,parametric and structural faults are easily occurred by the internal and external environment.Once the fault occurs,it will seriously endanger the personal safety of the operators and the power system.Therefore,the fault diagnosis of the inverter is of great significance for maintaining the safe and stable operation of the entire system.Based on this,a cascaded photovoltaic inverter is taken as an object,and its parametric and structural faults are studied from feature extraction and fault diagnosis to provide theoretical support for improving the reliability and stability of photovoltaic systems.For the parametric fault caused by the degradation of the bus capacitance,starting from feature extraction and fault diagnosis,mainly including:1)A parametric fault feature extraction method based on variational mode decomposition and hilbert-huang transform marginal spectrum(VMD-HHT marginal spectrum)is proposed;Meanwhile,an approximate entropy algorithm is used to determine the decomposition mode number(K)in the VMD process.The research shows that compared with empirical mode decomposition(EMD)and fast fourier transform(FFT),the proposed method can effectively implement the feature extraction of parametric faults.2)Propose parametric fault diagnosis methods of fast wavelet auto-encoders(FWAE)and deep wavelet extreme learning machine(DWELM);Meanwhile,in order to optimize the hyper-parameters of the proposed DWELM algorithm,an improved symbiotic organisms search(ISOS)algorithm was proposed for the hyperparameter optimization of DWELM.Finally,both the proposed DWELM and VMD-HHT marginal spectrum are together used in the parameter identification of the equivalent series resistance(ESR)of the bus capacitor,and the research shows that the proposed method can effectively improve the accuracy of parameter identification.For structural faults caused by IGBT open circuit,starting from feature extraction and fault diagnosis,mainly including:3)A structural fault feature extraction method based on deep auto-encoders(DAE)is proposed.Research shows that compared with wavelet decomposition(WT)and FFT,DAE is used for adaptive feature extraction of the original output voltage signal,which has thecharacteristics of high degree of automation and strong anti-noise ability,and can effectively implement the feature extraction of structural faults.4)Propose a manifold learning dimensionality reduction method called the ensemble principle component and neighborhood preserving(EPCNP).This method combines the advantages of principal component analysis(PCA)and neighborhood preserving embedding(NPE)to maximize the global and local structural characteristics of original data,which is more conducive to extracting low-dimensional effective information in the original data set.Finally,the proposed EPCNP method is combined with DAE to form a multi-scale EPCNP for structural fault diagnosis.The research results show that the multi-scale EPCNP can achieve fast and accurate fault diagnosis while maintaining good anti-noise performance.