Research On Low-Voltage AC Series Arc Fault Detection Method Based On Modal Decomposition Technology

With the development of economy and the continuous progress of science and technology,electric energy has become an important resource for production and life in modern society.While meeting people’s production and living needs,the potential safety hazards of electricity use are increasingly prominent.Among them,electrical fires caused by electrical failures have become a prominent issue of electrical safety due to their characteristics of great harm and wide impact.As an important cause of ignition electrical faults,series arc faults have become one of the low-voltage line faults that are difficult to detect due to their weak fault characteristics and strong concealment.When a series arc fault occurs,there is impedance at the fault point,and the volume of the fault part is small,which can easily lead to the accumulation of heat,causing the temperature at the fault point to rise,igniting the presence of combustibles near the fault point,and ultimately causing a fire.In this paper,the research object is AC series arc fault detection at 220 V level,and the following research has been carried out:(1)The characteristics of arc fault current under different loads were studied.Aiming at the problem that the current data is greatly affected by the type of load when a series arc fault occurs,through analyzing the measured current data under different loads,the characteristics and common rules of the fault current data under different loads are summarized,laying a foundation for the subsequent proposed arc fault detection algorithm.(2)Aiming at the problem that the fault features in the original current signal are deeply hidden,and the fault features of different loads may be located in different frequency bands,this paper applies the modal decomposition algorithm to arc fault feature extraction,and proposes a method for selecting the intrinsic mode function(IMF)of the feature.After decomposing the original current signal into multiple IMFs using a modal decomposition algorithm,calculate the Spearman correlation coefficients of each IMF and the standard power frequency sinusoidal current signal,use the absolute value of the correlation coefficients to achieve rough selection of IMF components,and use energy values to achieve feature component selection.(3)An arc fault detection model based on convolutional neural network(CNN)is established.Using the feature IMF component as the driving data for the neural network effectively amplifies the input feature information,facilitating rapid training of the model and improving accuracy.The effects of network structure,optimizer,batch size,and iteration times on network performance were explored through experiments.Aiming at the problem that the size and number of convolutional kernels and the size of pooled kernels can be set with different parameters in each layer of the network,there are many parameter combinations that make it difficult to find the optimal parameters using the ergodic method,an improved fireworks algorithm is proposed to automatically find the optimal parameters of the network using the improved fireworks algorithm.The experimental results show that the parameter optimization method can effectively improve the detection accuracy of arc faults.