Fault Identification And Localization Of A Time-Frequency Domain Joint Impedance Spectrum Of Cables Based On Deep Learning

With the development of urbanization,power cables have been widely used.Under the influence of the operating environment and the factory structure of cables,cables will have different degrees of moisture and aging,and it is difficult to detect the existence of defects because they are laid underground,and the defects will gradually develop into open circuit,short circuit,high resistance grounding,low resistance grounding and other fault under the action of electric field.Effectively and timely diagnosis of cable faults is one of the means to ensure the reliable operation of power systems.In view of the shortcomings of the existing methods,which require a large amount of manual identification of fault data for diagnosis and have a low diagnostic accuracy,this paper proposes a fault identification and localization of a time-frequency domain joint impedance spectrum of cables based on deep learning method.The research on cable fault identification and localization can avoid causing significant economic losses and diagnose the power system in time before permanent faults occur can improve its operational reliability.Firstly,this paper based on the equivalent model of power cable distribution parameters,model and analyze the cables under normal operation and fault conditions,and derive the relevant parameters of the cable lines to obtain the expressions of the headend input impedance amplitude and phase under various operating conditions.Inverse fast Fourier transform(IFFT)is used to transform the cable headend input impedance from frequency domain to time domain and obtain the corresponding real part expressions of the headend input impedance in time and frequency domain.Then build the model of single core cross linked polyethylene(XLPE)cable by PSCAD simulation software to obtain the modulus of the cable under normal operation and four types of faults including open circuit,short circuit,high resistance grounding and low resistance grounding.Use MATLAB to calculate the amplitude and phase values of the headend input impedance and the real part of the headend input impedance of the cable for normal operation and four types of faults conditions.Compare and analyze the change patterns of the cable headend input impedance spectrum under normal and different fault conditions which verify the cable operating conditions can be discriminated by the headend input impedance spectrum.The headend input impedance spectrum is subjected to IFFT to obtain the headend input time-frequency domain impedance spectrum of the cable,and extract the real part of the time-frequency domain impedance,and the waveform of it is verified to locate the fault by the time-frequency domain impedance spectrum.Finally,two deep learning methods,long short-term memory network(LSTM)and deep belief network(DBN),are used to construct fault identification and localization of cables based on deep learning model respectively.The headend input impedance amplitude and phase of normal operation and different faulty cables are selected as the original samples for the input cable fault type identification model;the real part of the faulty cable headend input time-frequency domain impedance is selected as the original samples for the input cable fault location model.The samples are divided into the ratio of 4:1 as training set and test set,which to analyze the effectiveness of two deep learning methods for cable fault diagnosis.The simulation results show that the proposed deep learning methods have good fault feature extraction capability,high fault type identification and location accuracy,which can be extended to the operation of smart grid in combination with the actual situation to save labor cost to a certain extent.