Series Arc Fault Detection Based On Deep Learning

Fault arc detection plays an important role in securing the safety of electrical equipment,and effectively prevents building electrical fires.In recent years,electrical fires caused by arc fault leads to nonnegligible damages on people's safety and properties.The relatively mature lowvoltage protection electrical products on the market are mainly to prevent the occurrence of overcurrent and leakage current,but these products cannot detect series arc fault on single-phase distribution lines.Therefore,the research on arc fault detection technology is of great significance to the reduction of electrical fires.In the AC power supply system,the series arc fault detection is the most difficult,and the difficulty of its detection algorithm is mainly two points: Firstly,it is necessary to ensure the accuracy of arc fault detection.The current signal of series arc fault is complex,which is closely related to the connected load.How to accurately determine the line status is still a problem at present.Secondly,the real-time detection needs to be guaranteed.That is,when an arc fault occurs,the Arc Fault Detect Device(AFDD)can act immediately or give an alarm.If the occurrence of a arc fault cannot be detected within the required time,then the algorithm also fails and cannot play a role in preventing electrical fires.In order to meet the above requirements,this paper studies the current traditional arc fault detection methods.In summing up the shortcomings of the traditional methods,a series arc fault detection method based on deep learning convolutional neural networks is designed.Because the series arc fault detection is based on a large amount of current data,and the current waveform of the series arc fault is different from the current waveform under normal conditions,the fault detection problem can be converted into a classification problem in deep learning for processing.This paper builds an experimental circuit based on the AFDD national standard,and designs a current data acquisition device to collect the current waveforms of some linear loads and non-linear loads as a database for convolutional neural network training and testing,including incandescent lamps,fluorescent lamps,vacuum cleaners and air compressors under different conditions.In the traditional arc fault detection methods,the characteristics of current in the time and frequency domains are studied,and an appropriate algorithm is designed to detect the arc fault based on these characteristics.In terms of arc fault detection based on convolutional neural network,a neural network suitable for detecting arc faults is designed,and the layer structure of the network,the evaluation indicators,and the selection of the optimization algorithm are determined.After the network design,the collected data is divided into training set and test set.The training set is used to train the neural network on the server to determine the parameters of the model.The test set is used to detect the effect of the model.Finally,the model has an excellent performance on the test set.The accuracy of series arc fault detection reaches 100%.In the end,the model is saved and processed and then imported into the embedded device for testing.The detection accuracy rate also reached 99.75%.This detection method not only improves the accuracy of the detection,but also ensures the timeliness of the detection.