| Modular multilevel converter(MMC)has been widely used in power electronic transformers as the input stage to realize power conversion because of its good expansibility,easy redundancy,low switching loss and low harmonics.MMC integrates a large number of submodules,every submodule is a potential failure device,which increases the failure probability of MMC.How to diagnose submodule faults through efficient fault diagnosis strategies to improve the safety and reliability of system is a problem that needs to be solved urgently.This dissertation takes the MMC which has half-bridge structure submodule as the research object,in order to improve the fault diagnosis reliability and rapidity of submodule,the parametric faults diagnosis techniques and structural faults diagnosis techniques are deeply studied.The works in this dissertation are as follows:(1)A simulation model of MMC is EstablishedFirst,the basic working principle and topological structure of MMC are analyzed,and the commonly used modulation strategies and voltage equalization methods are introduced.Then,according to the characteristics of the MMC in the power electronic transformer,the carrier phase-shifted sinusoidal pulse width modulation strategy and the double closed-loop control method are adopted to establish the simulation model of the MMC.The model is the basis of fault analysis and diagnosis.(2)An efficient parametric fault diagnosis method is proposedFirst,in view of the complex calculation and poor anti-interference of the existing parametric fault feature extraction methods,a feature extraction method based on matching pursuit is proposed.This method is used to extract the frequency domain feature vector from the bridge arm current and the submodule capacitor voltage.Secondly,according to the timing characteristics of the submodule,a parameter identification method based on the long short-term memory(LSTM)network is proposed,and the extracted fault feature vector is input to the LSTM network to predict the capacitance parameters.The research results show that the parametric fault diagnosis method which combine the proposed matching pursuit algorithm with the LSTM network can realize high diagnostic accuracy.(3)A structural fault feature extraction method based on deep wavelet siamese network is proposedAiming at the problem that it is difficult to extract effective features from small sample dataset in existing feature extraction methods,the advantages of deep wavelet neural networks and Siamese networks are combined,and deep wavelet Siamese network(DWSN)is proposed for extract fault feature.The research results show that the feature extraction method based on DWSN can adaptively extract the fault features,so that the feature vectors of different faults have a better degree of separation.(4)A separate diagnosis method for structural faults based on machine learning is ProposedAiming at the problem of large data collection and poor robustness in the submodule structural fault diagnosis method based on machine learning,a separate diagnosis method based on machine learning is proposed,which divides the fault diagnosis into fault detection and fault location.In the fault detection stage,the OSELM-VPMCD is proposed to detect the faulty bridge arms and the types of power tubes;in the fault location stage,a location method based on KNN abnormal value detection is proposed to locate the faulty submodule.The research results show that the proposed structural fault diagnosis method just required little dataset,and has the advantage of higher robustness and faster diagnosis speed. |
