Study On DC Arc Dynamic Model And Fault Detection Algorithm In PV System

DC arc is easily taken place because of loose contact points,aging insulation etc.While large photovoltaic(PV)system contains tens of thousands of contacts and lacks protection equipment from DC arc faults,which is easy to maintain and may cause electrical fire.As a result,DC arc brings huge safety risk to PV system and also operators.In addition,scientific research on DC arc faults in PV system just start recently,there are still many shortages in the following aspects:low fit of dynamic mathematical arc model for low-voltage and small-current dc system,the lack of systematic evaluation for feature extraction methods,arc fault location and fault type identification,etc.In this paper,the energy accumulation value Q in Mayr arc model is improved by linearizing the voltage and current in the dc arc transition process,to build up the dynamic mathematical model of low-voltage and small-current DC arc.Through the constant voltage and variable current DC arc experiments on the simulation platform,the calculation methods of parameters U_a、P_o andτare determined,and the agreement among the static arc resistance model,the original Mayr model and also the improved Mayr model besides the experimental waveform is analyzed to verify the effectiveness of the parameter calculation method and the dynamic mathematical model proposed in this paper.Based on the improved Mayr model,a simulation model of the DC arc fault in the PV system is built,the measured data of the Yunjiakou centralized PV system in Qinghai Province is applied as a template,and the output characteristics of the series and parallel DC arc faults were simulated and analyzed when some PV modules are shaded.However,in three cases,the output characteristics are overlapped,which can not effectively identify the operation state of the system,and it is necessary to obtain DC arc detection algorithm.Finally,six feature parameters of DC arc are extracted from the output current of PV system by means of frequency domain analysis,wavelet packet decomposition,power spectrum extraction and chaos analysis.They are the sum of spectrum amplitude,the ratio of standard deviation of amplitude between low frequency part and high frequency part,the correlation coefficient of spectrum amplitude,the sum of the variance of wavelet packet coefficient,the frequency at the fourth largest value of power spectrum amplitude and the maximum Lyapunov index.Based on the feature quantities,the logistic regression parameters were optimized through 8208 sets of normal operation data,3296 sets of interference data,and 8272 sets of DC arc data collected in the PV system.The mode identification of DC arc in PV system is carried out by logistic regression classification method.By analyzing the value of each feature to the pattern identification,the long-time-consuming Lyapunov index is removed and a high-accuracy DC arc detection algorithm for PV system is finally obtained and applied.Experimental data showed the feasibility of detection algorithm with an error rate of 0.24%and a miss rate of zero.