| With the increasing development of country’s social economy,the scale of the power grid continues to expand,and the use of cables as a carrier for electrical energy transmission in power system is increasing year by year as well.The cables in the city are mostly laid underground in a harsh environment,and are prone to cause early faults due to local insulation defects,which evolve into permanent faults versus time.In the operation of power system,the early faults are of short duration and often undetectable,but monitoring system in the power grid can always record relevant data and observe current changes.Identifying early cable faults from current changes allows timely arrangement of maintenance personnel for damaged cables and elimination of fault hazards,which is of great significance for improving power supply reliability of the grid.Firstly,this paper analyzes the mechanism and characteristics of early faults.Then,the basic principles of Cassie arc model,Mayr arc model and Schavemaker arc model are introduced respectively to analyze the applicability of each arc model.The early fault simulation model of 10 k V distribution network based on Schavemaker arc model is established in PSCAD/EMTDC,and the early fault characteristics of ungrounded neutral system and resonant earthed system are discussed according to the simulation results,which verifies the effectiveness of the fault model established in this paper.Based on the established arc fault model,the influence of arc parameters on the early fault characteristics is studied.Based on the simulation signals obtained from the early fault model established in PSCAD/EMTDC,the time domain characteristics of early faults and various disturbance signals are analyzed.In order to extract the characteristics of cable early fault,the dispersion entropy(DE)feature is introduced.Aiming at the problem of limited information of single-scale DE and poor stability of multiscale dispersion entropy feature,the refined composite multiscale dispersion entropy(RCMDE)feature is proposed,and the results show that RCMDE features have higher stability and are more conducive to early fault identification.What’s more,Aiming at the problem of chaotic features after extraction due to the influence of noise on the monitoring signal in the actual field.For the actual field monitoring signal is affected by noise,which may lead to the confusion of the extracted features,it is proposed to remove the influence of noise signal by variational mode decomposition(VMD),which avoids the modal aliasing problem of empirical mode decomposition(EMD).According to the noise resistance test,it is shown that the denoising effect of VMD algorithm is better than EMD and wavelet denoising method.Finally,the cable early fault identification method based on deep confidence network(DBN)is proposed.The simulation signal is denoised by VMD,the time domain features and RCMDE features are extracted and normalized as the feature vector of the network input,and the network structure is optimized by using particle swarm optimization.The experimental results show that the proposed method can effectively identify cable early faults from other disturbances and has higher accuracy and reliability than the traditional classification methods. |
