Research On Fault Diagnosis Of Inverter Based On Two-Dimensional Convolutional Neural Network

New energy power generation occupies an important position in the development of the world today.Photovoltaic power generation technology has also made great progress in recent years,and photovoltaic power generation technology is also developing towards large-scale grid connection.With the improvement of equipment automation,the failure of grid connected inverters will bring great harm to the entire power grid system.Therefore,in order to reduce the impact on people and social public property,it is particularly important to ensure the stability and safety of the power generation system through fault diagnosis of grid connected inverters.In order to apply the advantages of deep learning in recognizing two-dimensional images to fault diagnosis of three-phase grid connected inverters,this thesis proposes a inverter fault diagnosis model based on two-dimensional convolutional neural network(2D-CNN).Due to the different current signals of inverters in different operating states,the images formed by encoding their data in time series are also different,which makes image recognition technology available for time series classification and for identifying inverter fault current signals.Firstly,the main circuit topology of the three-phase grid connected inverter is described,and a voltage source three-phase two-level inverter is selected as the research object of this article.A simulation model of the three-phase grid connected inverter is built in MATLAB/Simulink.After analyzing the causes and types of faults in the inverter,all 22 operating conditions of the inverter are simulated using the Simulink simulation model,It includes 1 normal working condition and 21 open circuit fault conditions,and collects the Aphase current signal output by the inverter as one-dimensional time-domain signal data,which serves as the data basis for subsequent research.Secondly,the method of converting one-dimensional time domain signals into twodimensional images is studied.Converting a one-dimensional time-domain signal into a twodimensional image not only preserves the time dependence of the signal,but also allows deeper features of the signal to be mined.In this thesis,four methods,namely,the gramian angular summation field(GASF),the gramian angular difference field(GADF)in the gramian angular field(GAF),the markov transition field(MTF)and recurrence plots(RP),are used to convert one-dimensional time-domain signals into two-dimensional images,and then the converted two-dimensional images are used as a 2D-CNN dataset.Thirdly,the structure of 2D-CNN and the structures and principles of each layer in the network are analyzed and studied,and a 2D-CNN model suitable for fault diagnosis of threephase grid connected inverters in this thesis is established.The model structure is simpler than the classical model,and it also has high accuracy in identifying faults.Finally,the constructed 2D-CNN is introduced into the fault diagnosis of three-phase gridconnected inverters,and a 2D-CNN based inverter fault diagnosis method is proposed.Firstly,one-dimensional inverter fault signal data are converted into two-dimensional images using GASF,GADF,MTF and RP respectively.Different two-dimensional images are constructed into datasets,and different datasets are used as inputs to the 2D-CNN model.Then,the datasets are trained through the 2D-CNN model,Finally,fault detection,classification,and location are completed.The simulation results show that the recognition accuracy of this method is less affected by noise data,and has strong anti-interference ability within a certain range of noise.Compared with other traditional methods,this method has faster diagnosis speed,better accuracy and reliability.