В mun yi magana game da hasashen jerin lokaci. Ci gaba mai ma'ana zai zama labarin gano abubuwan da ba su da kyau.
Aikace-aikacen
Ana amfani da gano anomaly a wurare kamar:
1) Hasashen lalacewar kayan aiki
Don haka, a cikin 2010, kwayar cutar Stuxnet ta Iran centrifuges ta kai hari, wanda ya saita kayan aikin zuwa aiki mara kyau kuma ya lalata wasu kayan aikin saboda saurin lalacewa.
Idan da an yi amfani da algorithms na gano abubuwan da ba a sani ba akan kayan aiki, da an guje wa yanayin gazawar.
Ana amfani da binciken abubuwan da ba a sani ba a cikin aikin kayan aiki ba kawai a cikin masana'antar nukiliya ba, har ma a cikin ƙarfe da aikin injin turbin jiragen sama. Kuma a wasu wuraren da yin amfani da tsinkaya ya fi arha fiye da asarar da za a iya yi saboda rashin tabbas.
2) Hasashen zamba
Idan an cire kuɗi daga katin da kuke amfani da shi a Podolsk a Albaniya, ana iya buƙatar ƙarin bincika ma'amalar.
3) Gano tsarin mabukaci mara kyau
Idan wasu kwastomomi sun nuna halayen da ba su dace ba, za a iya samun matsalar da ba ku sani ba.
4) Gano buƙatu da kaya mara kyau
Idan tallace-tallace a cikin kantin sayar da FMCG ya ragu a ƙasa da tazarar amincewa na hasashen, yana da daraja gano dalilin abin da ke faruwa.
Hanyoyi don gano abubuwan da ba su da kyau
1) Goyan bayan Injin Vector tare da SVM Class One Class One
Ya dace lokacin da bayanan da ke cikin tsarin horarwa ke biye da rarraba ta al'ada, amma saitin gwajin ya ƙunshi abubuwan da ba su dace ba.
Na'ura mai ba da tallafi mai aji ɗaya tana gina wani wuri marar layi a kusa da asalin. Yana yiwuwa a saita iyakar yankewa wanda aka ɗauki bayanan da ba su da kyau.
Dangane da ƙwarewar ƙungiyarmu ta DATA4, SVM-Class shine mafi yawan amfani da algorithm don magance matsalar gano abubuwan da ba su da kyau.
2) Ware hanyar daji
Tare da hanyar "bazuwar" na gina bishiyoyi, hayaki zai shiga cikin ganyayyaki a farkon matakan (a zurfin zurfin bishiyar), watau. fitar da hayaki ya fi sauƙi don “keɓe.” Warewa dabi'u marasa ma'ana yana faruwa a farkon abubuwan da ke faruwa na algorithm.
3) Ambulan Elliptic da hanyoyin ƙididdiga
Ana amfani dashi lokacin da ake rarraba bayanai akai-akai. Mafi kusancin ma'aunin shine zuwa wutsiya na cakuda rarrabawa, ƙimar ƙimar mara kyau.
Hakanan ana iya haɗa wasu hanyoyin ƙididdiga cikin wannan ajin.
Hoto daga dyakonov.org
4) Hanyoyin awo
Hanyoyi sun haɗa da algorithms kamar maƙwabta k-kusa, maƙwabcin k-kusa, ABOD (ganowar waje na tushen kusurwa) ko LOF (fasali na waje).
Dace idan nisa tsakanin dabi'u a cikin halaye ne daidai ko al'ada (don kada a auna boa constrictor a parrots).
Algorithm na k-kusa mafi kusa yana ɗauka cewa dabi'u na yau da kullun suna cikin wani yanki na sararin samaniya, kuma nisa zuwa abubuwan da ba su da kyau za su kasance mafi girma fiye da rabewar jirgin sama.
5) Hanyoyin tagulla
Ma'anar hanyoyin tari shine idan darajar ta fi wani adadi nesa da cibiyoyin tari, ana iya la'akari da ƙimar mara kyau.
Babban abu shine yin amfani da algorithm wanda ke tattara bayanai daidai, wanda ya dogara da takamaiman aiki.
6) Hanyar da ta fi dacewa
Dace inda aka haskaka kwatance mafi girman canji a tarwatsawa.
7) Algorithms dangane da hasashen jerin lokaci
Manufar ita ce idan ƙima ta faɗi a waje da tazarar amincewar tsinkaya, ana ɗaukar ƙimar mara kyau. Don hasashen jerin lokaci, ana amfani da algorithms irin su smoothing sau uku, S(ARIMA), haɓakawa, da sauransu.
Algorithms na hasashen jerin lokaci an tattauna su a cikin labarin da ya gabata.
8) Koyon kulawa (sakewa, rarrabawa)
Idan bayanan sun ba da izini, muna amfani da algorithms kama daga koma baya na layi zuwa cibiyoyin sadarwa na yau da kullun. Bari mu auna bambanci tsakanin tsinkaya da ainihin ƙima, kuma mu zana ƙarshe zuwa nawa bayanai suka karkata daga al'ada. Yana da mahimmanci cewa algorithm ɗin yana da isassun ƙarfin haɓakawa kuma tsarin horon bai ƙunshi ƙima mara kyau ba.
9) Gwajin samfuri
Mu tunkari matsalar neman abubuwan da ba su dace ba a matsayin matsalar neman shawarwari. Bari mu lalata matrix ɗin mu ta amfani da SVD ko injunan haɓakawa, kuma mu ɗauki ƙimar a cikin sabon matrix waɗanda suka bambanta da na asali a matsayin abin ban mamaki.
Hoto daga dyakonov.org
ƙarshe
A cikin wannan labarin, mun sake nazarin manyan hanyoyin gano abubuwan da ba su da kyau.
Gano abubuwan da ba su da kyau ta hanyoyi da yawa ana iya kiran su da fasaha. Babu ingantaccen algorithm ko hanya, amfani da shi yana warware duk matsalolin. Mafi sau da yawa ana amfani da saitin hanyoyin don warware takamaiman lamari. Ana aiwatar da gano anomaly ta amfani da injunan tallafi na aji ɗaya, ware dazuzzukan, hanyoyin awo da tari, da kuma amfani da manyan abubuwan da aka haɗa da hasashen jerin lokaci.
Idan kun san wasu hanyoyin, rubuta game da su a cikin sharhin labarin.
source: www.habr.com