Research On Fault Detection And Location Method Of Photovoltaic Inverter Based On Deep Learning

Photovoltaic inverter is one of the most important system balance devices in photovoltaic power generation system.Insulate-Gate Bipolar Transistor(IGBT)is the core component of photovoltaic inverter.When it fails,it will affect the photovoltaic power generation system.The stable operation of the IGBT may cause immeasurable economic losses if the fault is not dealt with in time,so it is of great significance to the fault detection and accurate positioning of the IGBT.This paper will use the deep learning method to carry out related research on the aging fault and open circuit fault of IGBT.The main contents are as follows:(1)A label-free aging fault prediction method based on temporal convolutional network is proposed.Firstly,the aging failure mechanism of IGBT is analyzed,and the external electrical parameter with obvious aging characteristics-collector-emitter turn-off peak voltage value is selected as the aging failure characteristic parameter;secondly,the TCN network model is constructed,using the research of NASA The central accelerated aging experimental data set is used for experimental verification;finally,both performance evaluation indicators verify the effectiveness and accuracy of the proposed method for predicting aging fault characteristic parameters.(2)A single-label open-circuit fault diagnosis method based on attention recurrent neural network is proposed.First,a two-stage threephase photovoltaic grid-connected power generation simulation system is constructed,and the DC side current data of 22 different open-circuit faults that may occur in photovoltaic inverters are collected;secondly,in order to learn the fault context information and focus on high Correlation features,an attention recurrent neural network model is designed to classify faults;finally,the model is trained and tested using the collected data,and the results show that the model can effectively classify faults under different signal-to-noise ratios Compared with the algorithm,it has better robustness.(3)A method for locating open-circuit faults of photovoltaic inverters based on multi-tags is proposed.First,in order to use the correlation between faults to improve the localization performance,a multi-label data processing method is introduced;secondly,the multi-label improvement of Attention-LSTM is carried out,which can realize direct fault classification and localization;For training and testing,the results show that the method has better classification and localization performance compared with other traditional algorithms.The three new methods proposed in this paper can use flexible receptive fields to process fault features in parallel to accurately predict IGBT aging;learn fault context information to improve fault classification accuracy and robustness;use multi-label methods to directly and accurately It can locate the faulty IGBT effectively,which effectively reduces the fault diagnosis time and has good application value.