Research On Fault Modeling And Diagnosis Methods For The Photovoltaic Three Level Inverter

As the popularization and application of the PV micro-grid systems across C hina, the three- level inverter get a lot of use because of its obvious advantage in high voltage and large power field, the problem that the main circuit of three-level inverter is prone to happen open-circuit faults also increasing prominent and need to be addressed. So it is of great significance to study the detection and diagnosis of common faults in three- level inverter. This paper study the wavelet packet transform and sparse coding algorithm in the application of fault feature extraction about the three- level inverter, through the thorough analysis the common algorithm of fault detection and diagnosis about inverter in recent years. And completed the fault diagnosis of three level inverter with the help of the sectional diagnostic thinking. That is to complete the typical fault diagnosis with expert system method, to complete the atypical fault diagnosis with the fault dictionary method based on BP neura l network.Firstly, This paper started with the main circuit structure and working mechanism three- level inverter, analyzed its common failure types in detail, and turn it into two categories failure of typical and atypical. Then set up simulation model about fault of three- level inverter in SIMULINK, and analysis all of the faults through simulation, It is concluded that the fault feature information contain three special rule. This paper completed the typical fault diagnosis with expert system method, according to the rule of the typical fault waveform amplitude, this method in the diagnosis of high speed and accuracy. Because the recognition degree of atypical fault waveform is not high, So the fault feature extraction method is necessary to use to extract the fault features. This paper study the wavelet packet transform and sparse coding algorithm in the application of fault feature extraction about the three- level inverter separately. The wavelet packet transform through three layers wavelet packet decomposition of the d, q two-phase voltage after decoupling, with the energy of the spectrum formed 16 dimensional fault feature vector; the sparse coding algorithm based on ICA got the 10 dimensional fault characteristic vectors through sparse the three phase voltage signal. We can diagnosis the atypical fault with the fault dictionary diagnosis method based on BP neural network respectively, after got the feature vector of atypical fault. Through the validation of simulation platform, the ideas of these two k inds of fault diagnosis are of high speed, high accuracy, good feasibility. The speed and accuracy of fault diagnosis under which based on sparse coding are obviously better than that of wavelet packet transform. Sparse coding for fault feature extraction of three- level inverter is an innovation of this article. Using expert system and fault dictionary method based on BP neural network of sectional diagnosis way of thinking is another highlight of this article. Expert system is greatly reduced the burden of the neural network training, improve the efficiency of the whole of the diagnosis system effectively.