Electrical Fault Diagnosis Of Grid Connected Photovoltaic Inverter

Nowadays, with the rapid development of renewable energy sources, photovoltaic power generation has been widely concerned all over the world, and now, the solar energy grid generation has gradually become a kind of mainstream application mode. Because photovoltaic power plants are often built in Gobi, islands and other remote areas where have extremely poor physic-geographical environment, it is particularly important to ensure the grid-connected photovoltaic inverter, which is the core device of electrical power generating system, to be dependable and stable during operating. It will cause the entire system break down or even stop running once faults occurring in the inverter without timely and efficient repairing, and the consequences will be quite terrible. Therefore, it is very necessary to study the fault detection and diagnosis methods of inverter.Actually, photovoltaic grid-connected inverter is a complex assembly of many kinds of power electronic devices, thus, the existing traditional fault diagnosis methods for power electronic circuits do hardly apply. Therefore, aimimg at the nonlinear and non-stationary characteristics of the fault signals, this dissertation combines the two methods to extract the fault characteristic value of the inverter on the basis of learning empirical mode decomposition method and fractal geometry theory, and BP neural network is used to classify the patterns, and fairly good results are achieved. The effectiveness of the method is verified by comparing with the wavelet packet decomposition and reconstruction method.Firstly, in this dissertation, the structure and working principle of neutral point clamped three level photovoltaic grid-connected inverter are introduced in the first place, then, the primary fault modes are analyzed, and the open circuit faults of the main circuit are classified and encoded. After that, the simulation model is built and the simulation for different faults is carried out.Secondly, the fault signals, including the output line-to-line voltages of the inverter and the grid-connected phase currents, are collected. Afterwards, the signals are decomposed and reconstructed by using the wavelet packet analysis method and the feature values are extracted. And then, the fault pattern classification is realized by using the three layer BP network.Thirdly, the electrical signals are decomposed adaptively by making use of the characteristics of empirical mode decomposition method and a series of intrinsic mode functions would be obtained. Then, the box dimension and the generalized fractal dimension of these intrinsic mode functions are calculated by using the fractal geometry theory, and after that, the eigenvectors are built as the inputs of BP neural network.Finally, the three methods are analyzed and compared. The result indicates that the method of empirical mode decomposition combined with fractal theory is effective to extract fault feature values.