Study On Detection Technology Of Low-voltage AC Series Arc Faults

As the requirements for electrical safety are further improved,research on AC arc fault detection methods has attracted the attention of scholars in more fields.Due to the complexity of the actual electrical circuit,the normal operating waveforms of various nonlinear load circuits are similar to the arc fault waveforms.At present,most algorithms can only effectively diagnose some simple loads,and the accuracy of arc fault detection for nonlinear loads is low and the rate of misjudgment Higher.Therefore,in order to further propose an arc fault detection algorithm with higher accuracy and lower misjudgment rate,this paper conducts research on low-voltage AC series arc fault detection technology.Firstly,design and build an arc fault experimental platform,including the arc generator unit,the cable test sample carbonization unit,and the experimental platform control circuit,in accordance with the national standard "General Requirements for Arc Fault Protection Appliance(AFDD)" GB/T 31143-2014,to achieve communication The simulation of series arc and the control switching of each function test loop.Collect and store normal operation waveforms and arc fault waveform data of different current levels and different load types(resistive loads,vacuum cleaners,electronic switching power supplies,air compressors,electronic light regulators,fluorescent lamp loads,halogen lamp loads)through the arc experiment platform,Establish arc fault waveform database.Then,the typical time-domain current waveforms of resistive and shielded loads under normal conditions and arc conditions are analyzed,and the corresponding dimensionless time-domain characteristics are proposed.At the same time,FFT transformation is performed on the single power frequency periodic current signals of different loads to determine the corresponding frequency domain characteristic index.Further analysis and comparison of the distribution of various features in the normal operating state and the arc fault state to verify that the extracted features can be used as input to the arc fault diagnosis model.Finally,based on the quantified feature importance of the random forest,the time and frequency domain feature importance ranking processing is performed according to the feature importance score,and further the feature indexes with higher scores are selected as the basis for judgment.Establish a fault diagnosis model based on a probabilistic neural network,explore the influence of the number of input features and the value of the smoothing factor of the neural network on the training effect of the model,and then determine the optimal value of the input features and smoothing factor of the model.At the same time,further analysis of the causes of certain misjudgments and missed judgments in the arc fault diagnosis model,and then the continuous arc fault detection method is proposed to optimize the arc fault diagnosis process,further reduce the arc fault misdiagnosis rate,and improve the accuracy of arc fault detection.The arc fault detection method proposed in this paper can effectively detect low-voltage AC series arc faults,and can be used as the core detection algorithm of AFDD.