Research On Fault Diagnosis Method Of Cascaded H-bridge Seven-level Inverter Based On Multi-feature Fusion CNN

Grid-connected photovoltaic inverters are one of the key parts of photovoltaic grid-connected power generation systems,and their low failure rate and low maintenance cost play an important role in the development of the photovoltaic industry.As an important topology composed of cascaded H-bridges,cascaded multilevel inverters have the characteristics of high voltage,large capacity and strong scalability,and have become the focus of research in the field of inverters.However,the cascaded H-bridge multilevel inverter is composed of multiple cascaded H-bridge cascades,and the number of internal insulated gate bipolar transistors is significantly increased compared with the traditional low-level inverter.If the polar transistor fails,the output three-phase voltage is unbalanced,which will have a serious impact on the system.Research on fault diagnosis and fault location methods applicable to cascaded H-bridge multilevel inverters has great significance in improving the safety and stability of photovoltaic grid-connected power generation systems and ensuring power generation efficiency.Existing inverter fault diagnosis methods mainly focus on the method of single feature combined with traditional classifier.Given that different faults of cascaded H-bridge multilevel inverters are very similar and it is difficult for the traditional shallow classifiers to cope with high-dimensional feature inputs,which restrict the accuracy of fault diagnosis.The commission is improved from two levels of feature extraction and fault diagnosis.(1)Feature extraction(1)Time-domain feature extraction: collect the original three-phase current signal,use an improved multi-scale principal component analysis method to filter the components of different scale which characterize the fault characteristics and reconstruct them to obtain the high-dimensional time-domain signal features.Compared with traditional time-domain signals,the time-domain signals obtained by reconstruction have a higher signal-to-noise ratio and less redundant interference information,which is conducive to improving the accuracy of later fault diagnosis.(2)Time-frequency domain feature extraction: Hilbert yellow transform is performed on each scale component obtained by screening,and the marginal spectrum is extracted as the high-dimensional time-frequency domain signal feature that characterizes the fault characteristics.(2)Fault diagnosis(1)Given that traditional classifiers are difficult to deal with high-dimensional feature inputs thus the accuracy of fault diagnosis is limited,a fault diagnosis method for cascaded H-bridge multilevel inverters based on CNN is proposed.By comparing with the commonly used classification methods,the results verify the superiority of the proposed method.(2)Given that different faults of the cascaded H-bridge multilevel inverters are similar which hinder classifiers’ ability to distinguish different faults with a single feature.Combined with the idea of information fusion,the two features extracted above are used as the input of the dual-channel CNN model for training to construct a multiple Feature Fusion CNN Fault Diagnosis Model.The comparative analysis results show that the proposed time-domain and time-frequency domain feature extraction methods are more conducive to characterizing fault characteristics and improving the accuracy of fault diagnosis by comparing the proposed time-domain and time-frequency domain feature extraction methods with traditional time-domain and time-frequency domain feature extraction methods.The proposed multi-feature fusion CNN fault diagnosis has a fault diagnosis accuracy rate of 95%,which has a higher recognition rate and stronger adaptability than a fault diagnosis strategy that combines a single feature with a shallow classifier.