Π sathetha ngoqikelelo lothotho lwexesha. Ukuqhubeka okunengqiqo kuya kuba linqaku ekuchongeni izinto ezingaqhelekanga.
Isicelo
Ukufunyaniswa okungaqhelekanga kusetyenziswa kwiindawo ezifana nezi:
1) Ukuqikelelwa kokonakala kwezixhobo
Ke, ngo-2010, ii-centrifuges zase-Iranian zahlaselwa yintsholongwane ye-Stuxnet, eyabeka izixhobo ekusebenzeni ngokungekho ngqiqweni kwaye yakhubaza ezinye zezixhobo ngenxa yokunxiba okukhawulezileyo.
Ukuba i-algorithms yokufumanisa i-anomaly isetyenziswe kwisixhobo, imeko yokungaphumeleli yayinokuthi igwenywe.
Ukukhangela izinto ezingaqhelekanga ekusebenzeni kwezixhobo azisetyenziswa kuphela kwishishini lenyukliya, kodwa nakwi-metallurgy kunye nokusebenza kwee-injini zomoya. Kwaye kwezinye iindawo apho ukusetyenziswa kokuxilongwa kwangaphambili kungabizi kakhulu kunelahleko enokwenzeka ngenxa yokuphazamiseka okungalindelekanga.
2) Ukubikezelwa kobuqhetseba
Ukuba imali itsalwa kwikhadi olisebenzisa ePodolsk eAlbania, ukuthengiselana kunokufuneka kuqwalaselwe ngakumbi.
3) Ukuchongwa kweepateni ezingaqhelekanga zabathengi
Ukuba abanye abathengi babonisa ukuziphatha okungaqhelekanga, kusenokubakho ingxaki ongayaziyo.
4) Ukuchongwa kwemfuno engaqhelekanga kunye nomthwalo
Ukuba ukuthengisa kwivenkile ye-FMCG kuye kwehla ngaphantsi kwexesha lokuzithemba kwi-forecast, kuyafaneleka ukufumana isizathu sento eyenzekayo.
Iindlela zokuchonga izinto ezingaqhelekanga
I-1) Inkxaso yoMshini weVector kunye neClass One-Class ye-SVM
Ifanelekile xa idatha kwisethi yoqeqesho ilandela ukuhanjiswa okuqhelekileyo, kodwa isethi yovavanyo iqulethe i-anomalies.
Umatshini we-vector yenkxaso yeklasi enye yakha indawo engabonakaliyo malunga nemvelaphi. Kuyenzeka ukuseta umda wokuvala apho idatha ithathwa njengengaqhelekanga.
Ngokusekwe kumava eqela lethu le-DATA4, i-One-Class SVM yeyona algorithm isetyenziswa kakhulu ukusombulula ingxaki yokufumana izinto ezingaqhelekanga.
2) Indlela yehlathi yodwa
Ngendlela "engaqhelekanga" yokwakha imithi, ukukhutshwa kuya kungena emagqabini kumanqanaba okuqala (kwindawo enzulu yomthi), i.e. ukukhutshwa kwezinto ezikhutshwayo kulula βukukwahlula.β Ukwahlukaniswa kwamaxabiso angaqhelekanga kwenzeka kwi-iterations yokuqala ye-algorithm.
3) Imvulophu ye-elliptic kunye neendlela zamanani
Isetyenziswa xa idatha isasazwa ngokuqhelekileyo. Ukusondela komlinganiselo kumsila womxube wolwabiwo, kokukhona ixabiso elingaqhelekanga.
Ezinye iindlela zobalo nazo zingabandakanywa kule klasi.
Umfanekiso ovela kwi-dyakonov.org
4) Iindlela zeMetric
Iindlela ziquka algorithms ezifana k-abakufutshane abamelwane, k-ummelwane okufutshane, ABOD (angle-based outlier ubhaqo) okanye LOF (local outlier factor).
Kuyafaneleka ukuba umgama phakathi kwamaxabiso kwiimpawu uyalingana okanye uqhelekile (ukuze ungalinganisi i-boa constrictor kwi-parrots).
I-algorithm yabamelwane abasondeleyo k ithatha ukuba amaxabiso aqhelekileyo abekwe kwindawo ethile ye-multidimensional space, kwaye umgama ukuya kwi-anomalies uya kuba mkhulu kune-hyperplane eyahlulayo.
5) Iindlela zeCluster
Undoqo weendlela ze-cluster kukuba ukuba ixabiso lingaphezulu kwexabiso elithile kude namaziko e-cluster, ixabiso linokuthi lithathwe njengelingaqhelekanga.
Into ephambili kukusebenzisa i-algorithm edibanisa ngokuchanekileyo idatha, exhomekeke kumsebenzi othile.
6) Indlela yecandelo eliphambili
Ifanelekile apho izalathiso zolona tshintsho lukhulu kusasazo lugxininisiwe.
I-7) Ii-algorithms ezisekelwe kwi-forecast series time
Uluvo lwelokuba ukuba ixabiso liwela ngaphandle kwexesha lokuzithemba, ixabiso lithathwa njengelingaqhelekanga. Ukuqikelela uthotho lwexesha, i-algorithms efana ne-triple smoothing, i-S(ARIMA), i-boosting, njl.
Uluhlu lwexesha lokubikezela i-algorithms lwaxoxwa kwinqaku elidlulileyo.
8) Ukufunda okugadiweyo (ukuhlehla, ukuhlelwa)
Ukuba idatha iyasivumela, sisebenzisa i-algorithms ukusuka kwi-linear regression ukuya kuthungelwano oluqhubekayo. Masilinganise umahluko phakathi koqikelelo kunye nexabiso lokwenyani, kwaye sigqibe ekubeni idatha itenxa kangakanani na kwinto eqhelekileyo. Kubalulekile ukuba i-algorithm inamandla okwaneleyo okwenziwa ngokubanzi kwaye isethi yoqeqesho ayiqulathanga ixabiso elingaqhelekanga.
9) Iimvavanyo zemodeli
Masisondele kwingxaki yokukhangela izinto ezingaqhelekanga njengengxaki yokukhangela iingcebiso. Masibolise i-matrix yethu yeempawu sisebenzisa i-SVD okanye oomatshini bokwenza izinto, kwaye sithathe amaxabiso kwimatrix entsha ahluke kakhulu kweyokuqala njengento engaqhelekanga.
Umfanekiso ovela kwi-dyakonov.org
isiphelo
Kweli nqaku, sihlolisise iindlela eziphambili zokufumanisa i-anomaly.
Ukufumana izinto ezingaqhelekanga ngeendlela ezininzi kunokubizwa ngokuba bubugcisa. Ayikho i-algorithm efanelekileyo okanye indlela, ukusetyenziswa kwayo kusombulula zonke iingxaki. Amaxesha amaninzi iseti yeendlela ezisetyenziswayo ukusombulula imeko ethile. Ukufunyanwa kwe-Anomaly kuqhutyelwa kusetyenziswa umatshini we-vector yenkxaso yeklasi enye, ukwahlula amahlathi, iindlela ze-metric kunye ne-cluster, kunye nokusetyenziswa kwamacandelo enqununu kunye nokuqikelelwa kwexesha.
Ukuba uyazazi ezinye iindlela, bhala ngazo kwizimvo kwinqaku.
umthombo: www.habr.com