Fault Diagnosis Research Of Photovoltaic Grid-Connected Inverter Based On Wavelet Neural Network

With the attention and development of new energy all around the world,solar power has been widely used.This new energy is rich and non-polluting.In the solar power generation structure,grid-connected inverter has a very important role.If the inverter fails and does not get timely diagnosis,it will produce economic losses and bring a threat to people's life.So the fault diagnosis of the inverter is necessary.In this paper,a fault diagnosis method based on wavelet neural network is used to diagnose different fault types of the inverter.Firstly,the photovoltaic power generation technology was introduced,and the topology structure of PV grid-connected inverter was described.The Three-phase Three-Level Inverter of a type of the diode neutral point clamped was selected for the study,and its working principle and fault types were analyzed.The typical 11 fault types of A phase on the inverter circuit are studied.Then,the fault simulation model of three-level inverter was built in matlab/simulink simulation environment,and 11 different fault types were simulated in this model.From the simulation results,the fault information was obtained.That is bridge arm voltage,the upper bridge arm voltage and the lower bridge arm voltage.Finally,the diagnosis of different fault types of the inverter was completed.The whole process was divided into three stages:fault information collection,fault feature extraction and fault type recognition.Fault information collection was to collect the bridge arm voltage waveform of different fault types.The fault feature was extracted by wavelet transform,The collected voltage waveforms were decomposed by three-layer wavelet packet,and the voltage signals of different fault types were divided into eight bands of energy values.The energy values were normalized,and the processed data constituted the eigenvector.These feature vectors were used as the input sample data of the designed BP network,and then the desired output target was encoded.Finally the DC side voltage was set to 720V,700V and 680V,and the modulation ratio was chosen to be 0.2 to 0.9.24 sets of sample data of each fault type can be obtained,and a total of 264 fault characteristics were obtained.The datas of 2/3 were selected for training data and the datas of 1/3 were selected for testing data.The input sample datas were trained and tested by BP neural network.The simulation results show that the method has higher diagnostic accuracy,easy implementation and certain engineering application value.