Research On Fault Diagnosis Method Of Photovoltaic Inverter Based On Deep Learning

With the gradual depletion of traditional fossil energy and the intensification of global environmental pollution problems,the Energy Internet,which is the core of the third industrial revolution,gradually uses renewable energy to replace fossil energy and transforms the energy consumption structure into low-carbon energy.In the production of renewable and low-carbon energy,solar energy can be said to be the largest and inexhaustible energy source,and it has received extensive research and attention.The operation process of the photovoltaic power generation system:the solar cell module array converts the energy output from the solar energy and sends it to the DC power distribution cabinet through the DC combiner box,which is inverted by the grid-connected inverter to AC power,and finally integrated into the energy Internet.The open circuit of insulated gate bipolar transistor(IGBT)in the grid-connected inverter will not only affect the normal operation of photovoltaic inverter,but also reduce power generation efficiency of the entire system.Most of the photovoltaic power plants are built in high plateau areas with harsh environments,and the maintenance costs are relatively high.Therefore,it is a problem worth studying to quickly and effectively diagnose opencircuit faults of photovoltaic inverters.In this thesis,a photovoltaic power generation simulation system is built.From the perspective of reducing the number of sensors,the DC side current is selected as the detection quantity,and the DC side current signal is collected to construct a fault data set.In order to better fit the real operating environment of photovoltaic power generation system,the fault data set introduces noise with different signal-to-noise ratios.In the field of power system fault diagnosis,traditional fault diagnosis methods are difficult to apply to complex and intelligent power equipment.Based on the depth of data-driven learning algorithm adaptable and high precision,has a very broad application prospects in fault diagnosis.Therefore,in view of the open circuit fault of photovoltaic inverter,this paper proposes three deep learning algorithms:(1)A photovoltaic inverter fault diagnosis method based on EMD-CNN:Use empirical mode decomposition to obtain multi-scale features from fault data,and convert the features into a two-dimensional matrix as the input of the convolutional neural network.After that,features are extracted through CNN and classified;(2)CNN fusion LSTM photovoltaic inverter fault diagnosis method:Use CNN to automatically extract features from the data set and input them into the long-and short-term memory network,and then LSTM learns contextual features.Finally,complete the classification of the data set;(3)Multi-label PV inverter fault diagnosis method:Based on the feature extraction of CNN and LSTM,a multi-label-based loss function is set up to learn the correlation between faults to improve classification performance.Finally,according to the output value of the network,it can be determined whether the IGBT at the corresponding position has an open circuit fault.Experimental results show that the three algorithms have high reliability in the fault diagnosis of photovoltaic inverters.CNN model can extract the deep local features of the data and improve the feature distinguishing ability.The fault diagnosis method based on EMD-CNN can use EMD to eliminate redundant features of data and improve feature diversity.CNN-LSTM learns time correlation information from advanced features to improve the classification accuracy of photovoltaic inverter fault diagnosis.The fault diagnosis method based on multiple tags has the ability to locate the open circuit fault of the IGBT,which greatly reduces the fault diagnosis time.