Research On Fault Diagnosis Approach Of PV Inverter

The current environmental situation is worsening, photovoltaic power is a clean energy has been widespread concern. The PV inverter, as an important part of the photovoltaic power generation system, operating at high voltage, high power environments, so the fault can not be ignored. Now, for troubleshooting problems PV inverter is not yet comprehensive, researches in this area is relatively scarce. The topic for the current common PV inverter system, to study its occurrence and to determine failure modes, a new method uses extreme learning machine failure mode discrimination.In this paper, research on the fault diagnosis of photovoltaic inverter, take the most widely used of diode neutral point clamped (NPC) photovoltaic inverter currently for the study. First, we further study analysis of various type of power device potential failure of three-level NPC photovoltaic inverter; Secondly, further analysis the Related theory of Extreme Learning Machine (ELM) networks and its typical applications in practice, and also analyzes how to use the wavelet transform to extract feature vectors; Finally, under the MATLAB environment to build a photovoltaic inverter of NPC experimental platform, and compared the classification results of the limit learning machine algorithm and support vector machine (SVM) algorithm in the fault diagnosis of NPC PV inverter, as a result, the two pattern recognition methods can both achieve the purpose of classified the type of fault correctly, but the diagnostic accuracy of Extreme Learning Machine method is higher and faster. Fully demonstrated the proper operability and advantages of the fault diagnosis of Extreme Learning Machine as the paper researched.The main topic for fault diagnosis the problem three-level NPC PV inverter studied, presents a new method for learning and limits based on the combination of wavelet transform and support vector machine (SVM) algorithm is added contrast to prove usefulness of the method proposed by this project. Eventually obtained by simulation based on Extreme Learning Machine (ELM) network diagnostics and reliability advantages.