Feature Extraction And Recognition Of Low-Voltage AC Fault Arc

With the vigorous development of science and technology and the remarkable improvement of consumers’ quality of life,various kinds of electrical equipment emerge in endlessly,and the occurrence rate of electrical fire accidents is higher and higher.Due to the complexity of the actual circuit,the normal current waveform of the nonlinear load loop is similar to that of the normal current waveform of other loads,the detection accuracy of multi-load loop is low.In order to detect the fault arcs in the circuit more accurately and quickly,the technology of detecting low-voltage AC series fault arcs is studied in this thesis.Firstly,an experimental platform for fault information acquisition is built to collect enough loop current for detecting fault arcs.Besides collecting the normal current and fault current when each electric apparatus works alone,many typical loads are combined according to the actual situation to collect the normal current and fault current.Furthermore,the time-frequency domain characteristics of the normal current and fault current are analyzed.In this thesis,fast Fourier transform(FFT)is applied to the single power frequency periodic current waveform of different loads,according to the correlation coefficient between the spectrum of other loads and the spectrum of pure resistive loads,the load is divided into switching power supply load and non-switching power supply load.The feature is extracted from the low frequency feature of current,Principal component analysis(PCA)is used to reduce the dimension of feature matrix.After dimensionality reduction,the amount of information retained by PCA is more than99%.Finally,the decision tree is selected as the fault detection model,and the optimal parameters of the model are determined by combining the grid search method with the Kfold cross test method.Combined with bagging and boosting,the decision tree model is optimized to improve the accuracy of fault detection model.Considering the difference of data observation and training principle of different algorithms,in order to make full use of the advantages of each model,a low voltage AC series fault arc detection model based on Stacking ensemble learning embedded by multiple machine learning algorithms is constructed.Compared with the traditional binary classification model,the Stacking ensemble learning method based on multi-model fusion has better classification effect.