Π takataura nezvekufanotaura kwenguva. Kuenderera mberi kune musoro kungava chinyorwa chekuziva zvisizvo.
Kushanda
Anomaly kuongororwa kunoshandiswa munzvimbo dzakadai se:
1) Kufanotaura kwekuparara kwemidziyo
Nekudaro, muna 2010, Iranian centrifuges yakarwiswa nehutachiona hweStuxnet, iyo yakamisa michina kuti isashande zvakanaka uye yakaremadza mimwe michina nekuda kwekukurumidza kupfeka.
Dai anomaly yekuona algorithms yakashandiswa pamidziyo, mamiriro ekutadza angadai akadziviswa.
Kutsvaga kweanomalies mukushanda kwemidziyo haishandiswe kwete muindasitiri yenyukireya chete, asiwo mune simbi uye kushanda kwema turbines endege. Uye mune dzimwe nzvimbo uko kushandiswa kwekufungidzira kwekufungidzira kwakachipa kupfuura kurasikirwa kunobvira nekuda kwekuparara kusingafungidzirwe.
2) Kufanotaura kwehutsotsi
Kana mari ikabviswa kubva pakadhi raunoshandisa muPodolsk kuAlbania, kutengeserana kungada kuongororwa zvakare.
3) Kuzivikanwa kwemaitiro asina kujairika evatengi
Kana vamwe vatengi vakaratidza maitiro asina kunaka, panogona kunge paine dambudziko rausiri kuziva.
4) Kuzivikanwa kwekuda kusingaite uye mutoro
Kana kutengesa muchitoro cheFMCG kwadonha pazasi penguva yekuvimba yekufanotaura, zvakakodzera kuwana chikonzero chezviri kuitika.
Nzira dzekuziva anomalies
1) Tsigira Vector Machine ine Imwe Kirasi Imwe-Kirasi SVM
Inokodzera kana iyo data iri museti yekudzidziswa ichitevera yakajairika kugovera, asi bvunzo seti ine anomalies.
Iyo imwe-kirasi yekutsigira vector muchina inovaka isina mutsara nzvimbo yakatenderedza kwakabva. Izvo zvinokwanisika kuseta cutoff muganho weiyo data inoonekwa seyakaoma.
Zvichienderana neruzivo rwechikwata chedu cheDATA4, Imwe-Kirasi SVM ndiyo inonyanya kushandiswa algorithm yekugadzirisa dambudziko rekutsvaga zvinokanganisa.
2) Isolate musango nzira
Ne "random" nzira yekugadzira miti, kubudiswa kuchapinda mumashizha pamavambo ekutanga (pakadzika kudzika kwemuti), i.e. zvinobuda zviri nyore "kuzviparadzanisa." Kusarudzika kwemaitiro akashata kunoitika mune yekutanga iterations yealgorithm.
3) Elliptic envelopu uye nzira dzehuwandu
Inoshandiswa kana data ichiwanzo kugoverwa. Iko kuswedera kweyero kumuswe wemusanganiswa wekugovera, iyo inonyanya kushamisa kukosha.
Dzimwe nzira dzenhamba dzinogona kuverengerwa mukirasi ino.
Mufananidzo kubva dyakonov.org
4) Metric nzira
Nzira dzinosanganisira algorithms senge k-pedyo nevavakidzani, k-pedyo muvakidzani, ABOD (angle-based outlier discover) kana LOF (local outlier factor).
Inokodzera kana chinhambwe chiri pakati pehunhu muhunhu hwakaenzana kana hwakajairwa (kuitira kuti usayera boa constrictor mumaparrots).
Iyo k-yepedyo vavakidzani algorithm inofunga kuti yakajairwa hunhu huri mune imwe nzvimbo yenzvimbo ine multidimensional nzvimbo, uye chinhambwe cheanomalies chichave chakakura kupfuura kune inoparadzanisa hyperplane.
5) Nzira dzeCluster
Izvo zvakakosha zvemasumbu maitiro ndezvekuti kana kukosha kwakawanda kupfuura imwe chiyero kure kubva kunzvimbo dzemasumbu, kukosha kunogona kunzi kunoshamisa.
Chinhu chikuru ndechokushandisa algorithm inonyatso unganidza data, zvinoenderana nebasa chairo.
6) Principal chikamu nzira
Inokodzera uko nzira dzeshanduko huru mukupararira dzinoratidzwa.
7) Algorithms yakavakirwa pane yakatarwa nguva yekufungidzira
Pfungwa ndeyokuti kana kukosha kukawira kunze kwekufanotaura kwekuvimba, kukosha kunoonekwa sekusanzwisisika. Kufanotaura nhevedzano yenguva, algorithms akadai seyakapetwa katatu, S(ARIMA), boosting, nezvimwe zvinoshandiswa.
Nguva yakatevedzana yekufungidzira algorithms yakakurukurwa muchinyorwa chakapfuura.
8) Kutariswa kudzidza (kudzoreredza, kupatsanura)
Kana iyo data ichibvumira, isu tinoshandisa algorithms kubva kumutsara regression kusvika kune inodzokororwa network. Ngatiyere mutsauko uripo pakati pekufanotaura uye kukosha chaiko, uye totora mhedziso kusvika papi iyo data inotsauka kubva pane yakajairwa. Izvo zvakakosha kuti iyo algorithm ine yakakwana generalization kugona uye kuti seti yekudzidziswa haina maitiro asina kunaka.
9) Miedzo yemuenzaniso
Ngatisvike padambudziko rekutsvaga anomalies sedambudziko rekutsvaga mazano. Ngatiparadzei yedu matrix tichishandisa SVD kana factorization michina, uye titore makoshero ari mumatrix matsva akasiyana zvakanyanya neayo ekutanga seasinganzwisisike.
Mufananidzo kubva dyakonov.org
mhedziso
Muchikamu chino, takaongorora nzira huru dzekuona anomaly.
Kuwana anomalies kunogona nenzira dzakawanda kunzi unyanzvi. Iko hakuna yakanaka algorithm kana nzira, kushandiswa kwayo kunogadzirisa matambudziko ese. Kazhinji seti yenzira inoshandiswa kugadzirisa imwe nyaya. Kuonekwa kweanomaly kunoitwa uchishandisa imwe-kirasi tsigiro vector michina, kupatsanura masango, metric uye masumbu nzira, pamwe nekushandisa zvakakosha zvikamu uye fungidziro yenguva.
Kana iwe uchiziva dzimwe nzira, nyora pamusoro pavo mumashoko kune chinyorwa.
Source: www.habr.com