Research On Fault Diagnosis Technology Of Photovoltaic Inverter Based On Convolutional Neural Network

Solar energy is one of the most promising new energy sources since the 21st century,whose wide distribution,large content,and pollution-free characteristics have been recognized worldwide.Photovoltaic power generation has become a way to efficiently use solar energy,transforming solar energy into electrical energy and delivering it to thousands of households,which facilitates the lives of residents and promotes humanity towards the era of clean energy.Photovoltaic inverter is the power electronic device in the photovoltaic power generation system that undertakes the conversion of electrical energy,which can convert the direct current generated by the photoelectric effect into alternating current and merge it into the grid or directly for the load.Under rated operating conditions,the failure of the photovoltaic inverter not only directly affects the efficiency of the photovoltaic power generation system,but affects the stable operation of the energy internet.Therefore,it is necessary to diagnose and eliminate it in time.In the past,most of the core devices of photovoltaic inverters,insulated gate bipolar transistors(IGBTs),were diagnosed with open-circuit fault diagnosis using three-phase voltage or three-phase current on the AC side,which required 3 sensors to monitor three-phase voltage or current signals.In this paper,the DC side current signal of photovoltaic inverter is used as the detection quantity to realize the open circuit fault diagnosis of the inverter,which can effectively reduce the number of sensors.Deep learning provides important technical support for the energy Internet's vitality features of smart and self-learning.Convolution Neural Networks(CNN)is one of the typical applications in deep learning.It can directly start from the low-level original signals and obtain deep feature representations through layer-by-layer learning to avoid the complexity that was caused by traditional artificial feature extraction and selection and uncertainty in order to improve the intelligence of the diagnosis process.Compared with other traditional fault diagnosis methods,the fault diagnosis method,which is based on convolutional neural network not only can get rid of the dependence on many signal processing techniques and diagnostic experience and complete the adaptive extraction of fault features.But also,the joint end-to-end learning of the detector and the convolutional neural network can effectively solve the problem of insufficient diagnostic ability that was caused by the mismatch of features and detection targets.This method has strong versatility and adaptability,and is very suitable for dealing with the problem of fault diagnosis of energy Internet photovoltaic inverters.The Simulink simulation platform was used to build a photovoltaic power generation system model,which considered the change in light intensity.The maximum power point tracking method was used to ensure that the system worked stably near the maximum power point.It can simulate the situation of open circuit failure of the photovoltaic inverter,and collect the DC side current signal data by using the photovoltaic system model in the steady state.After the data is preprocessed,the two methods of CNN and EMD-CNN proposed in this paper are used for fault diagnosis.The fault diagnosis method based on the CNN model directly extracts the features of the pre-processed one-dimensional current signal data,and then uses a fully connected detector to complete the fault diagnosis.The fault diagnosis method based on EMD-CNN firstly uses the empirical mode decomposition(EMD)method to obtain multiple inherent modal function(IMF)components of the fault data,convert the original one-dimensional signal into a two-dimensional signal,and then use two-dimensional CNN model to complete the feature extraction,and then connect the fully connected layer to diagnose the fault state.After experimental verification,both methods have achieved better fault diagnosis performance.In addition,fault diagnosis tests under different noise levels are also designed.The test results show that the fault diagnosis that is based on EMD-CNN model has higher accuracy and stronger robustness in noisy environments.