Research On Arc Protection Early-warning And Forecasting System Of LV/MV Switchgear

With the widly application of aggregated switchgears in LV/MV Power Network,arcing faults in LV/MV busbars happen frequently due to various factors such asequipment and running environment. Arcing faults have characteristic as followsrapid expanding, unpredictability, great destructiveness, and could cause an explosionand leads switchgears to burn. So far, there isn’t have busbar protection specificallyfor LV/MV switchgear, therefore, development of the rapid and effective arcprotection is extremely urgent. At present, there are refined of arc protection system athome and abroad, most of them use one or two of arc signal and short circuit currentsignal, which appear after the occurrence of arc, as the basis for protection. Thismethod is passive, it always cut off the fault quickly, but have caused some lossalready at the same time. It is very important to research early arcing fault detectingmethods.To solve the selection of feature in strong noise, firstly, wavelet analysis, waveletpacket analysis, feature extraction techniques and pattern recognition technology areresearched in this paper, take how to extract the feature and pattern recognition, whichbased on wavelet packet transform and fuzzy c-means clustering algorithm, as theresearch key.Then, a specific experiment system is set up, which is to study the change ofarcing sound, light, current of medium-low voltade switchgear under differentconditions of discharge test. Analyzed from two aspects of time domain andfrequency domain, find the internal relations of these three signals, and their variationin the discharge process. Experimentation is provided that it is feasible to make thearc sound as a early warning singal，and the arc light and current signal as theprotection signal of arcing fault.After that, with the time-frequency domain localized nature of wavelet packetanalysis, which can also divided signal into multi-scale in the whole frequency range,fault arc signal is decomposed to three components, the feature which cancharacterize the nature of the arc sound is obtained. A pattern recognition system isestablished based on fuzzy c-means clustering algorithm to training and testing thefeature.Finally, designed the arc protection early-warning and Forecasting system basedon DSP, Devised the hardware and software, and verified the feasibility of the system.