Research On DC Arc Fault Detection In Photovoltaic System

With the rapid development of photovoltaic industry,the frequency of fire in photovoltaic system is increasing,which seriously endangers the normal operation of photovoltaic system.Some research results show that the direct cause of fire in photovoltaic system is DC arc fault.DC arc has no zero-crossing point,periodicity and other characteristics.It has the characteristics of high temperature,high heat and short duration.At the same time,the structure of photovoltaic system is complex,which makes DC arc fault detection of photovoltaic system very difficult.It has important application value for the research of DC arc fault detection in photovoltaic system.Firstly,based on the simulation of DC arc fault in photovoltaic system,an experimental platform for DC arc fault data acquisition in photovoltaic system is built,and the hardware and software of the acquisition device are designed and implemented.The experimental data of arc fault and normal working conditions are collected through the experimental platform.The current peak to peak values and current variance in in time domain,and high frequency components in frequency are extracted as the characteristics of arc fault.Then,the theory and practical application of mathematical morphology are introduced,and the feasibility of mathematical morphology in DC arc fault detection is discussed.Two different mathematical morphological operators and their combinations are proposed and applied to DC arc fault to study their effects.The first one is based on morphological open-close gradient transform to analyze arc fault;the second one is based on different mathematical morphological filters to analyze arc fault.The experimental results show that the two methods can extract arc fault features from different angles,and can be used in the application of DC arc fault detection,which provides another method and idea for solving DC arc fault.Finally,a scheme of DC arc fault recognition based on cyclic neural network is designed.The cyclic neural network is suitable for dealing with time series related tasks.In this experiment,the cyclic neural network is used to extract features,combined with the features extracted before,and trained in the neural network.Finally,the classification result of DC arc fault is achieved with 98.24%accuracy.Experiments show that the DC arc fault recognition scheme based on cyclic neural network has high accuracy,and has certain reference value and reference significance for future research.