Research On Fault Diagnosis Method Of Power Inverter Based On Neural Network

In recent years,with the rapid development of alternative energy,photovoltaic power generation technology has attracted the attention of many experts and scholars.Because the inverter is a core device in the photovoltaic power generation system,the safe operation of the inverter is an important indemnification for high-quality power output.In the photovoltaic grid-connected system,the failure of the inverter will affect the quality of output current of the grid-connected side directly.And it can directly affect whether the grid can operate safely or not.Therefore,the research of the fault diagnosis of inverter is of great significance.The faults of photovoltaic grid-connected three-phase inverters and cascaded H-bridge seven-level inverters are studied in this dissertation.Firstly,the works of photovoltaic grid-connected inverter and cascaded H-bridge seven-level inverter are analyzed briefly,and the corresponding mathematical models are established on the basis of their circuit principles.According to the number of power tube faults,the fault types of photovoltaic grid-connected three-phase inverters are classified.Fault codes are encoded for the two inverter models based on the type of fault.The simulation models are set up for the two inverters respectively,and the fault types of the inverter are analyzed according to the simulation models.And the output waveforms of voltage and current by different fault types are simulated on the basis of the simulation models.Then,the current mainstream fault feature extraction methods are summarized,including wavelet transform,principal component analysis and wavelet packet transform.The basic principles of BP neural network used in this article are summarized.Based on the above algorithm,the WTP-PCA-BP algorithm is proposed to analyze single-switch and double-switches open-circuit faults of photovoltaic grid-connected three-phase inverters.In order to verify the superiority of the proposed algorithm,the faults of the model are analyzed by FFT-BP and PCA-BP respectively.To compare the classification effects of the three algorithms obviously,t-SNE algorithm is used to visualize fault feature vector.With the comparisons of three algorithms,the proposed algorithm has not only good fault diagnosis accuracy but also good robustness to noise.Finally,the main ideas of deep learning are summarized,and three model structures of deep neural network are briefly introduced.The network structure and training process of the stack auto-encoding network are elaborated.The SAE-SOFTMAX algorithm is used to diagnose the open circuit fault of the cascaded H-bridge seven-level inverter.The simulation experiments verify that the algorithm has superior fault diagnosis characteristics.In addition,the FFT-BP algorithm and WPT-PCA-BP algorithm are applied to the fault diagnosis problem respectively,which have been compared with the results of SAE-SOFTMAX algorithm.In this dissertation,the t-SNE algorithm is used to visualize the fault features extracted by the three fault feature extraction methods.The results show that the algorithm of this article has many superior features than the other two algorithms.