В ua kamaʻilio mākou e pili ana i ka wānana manawa. ʻO ka hoʻomau kūpono he ʻatikala e pili ana i ka ʻike ʻana i nā anomalies.
Ka hoʻohana
Hoʻohana ʻia ka ʻike anomaly ma nā wahi e like me:
1) Ka wanana o ka nahaha lako
No laila, i ka makahiki 2010, ua hoʻouka ʻia nā centrifuges Iranian e ka Stuxnet virus, ka mea i hoʻonohonoho i nā mea hana i ka hana maikaʻi ʻole a hoʻopau i kekahi o nā mea hana ma muli o ke kāʻei wikiwiki.
Inā ua hoʻohana ʻia nā algorithm ʻike anomaly ma nā mea hana, hiki ke pale ʻia ke kūlana hemahema.
ʻAʻole hoʻohana ʻia ka ʻimi ʻana i nā anomalies i ka hana o nā mea hana i ka ʻoihana nuklea wale nō, akā i ka metallurgy a me ka hana ʻana o nā turbines mokulele. A ma nā wahi ʻē aʻe kahi i ʻoi aku ka liʻiliʻi o ka hoʻohana ʻana i nā diagnostics predictive ma mua o nā poho hiki ʻole ma muli o kahi haki ʻole.
2) wānana hoʻopunipuni
Inā hoʻihoʻi ʻia ke kālā mai ke kāleka āu e hoʻohana ai ma Podolsk ma Albania, pono paha e nānā hou ʻia nā hana.
3) Ka ʻike ʻana i nā ʻano mea kūʻai like ʻole
Inā hōʻike kekahi mau mea kūʻai aku i ka hana ʻino, aia paha kahi pilikia āu i ʻike ʻole ai.
4) Ka ʻike ʻana i ka koi a me ka ukana ʻole
Inā hāʻule nā kūʻai aku ma kahi hale kūʻai FMCG ma lalo o ka manawa hilinaʻi o ka wānana, pono e ʻimi i ke kumu o ka mea e hana nei.
Nā ala e ʻike ai i nā anomalies
1) Kākoʻo i ka mīkini Vector me ka SVM papa hoʻokahi
He kūpono i ka wā e hahai ai ka ʻikepili i loko o ka hoʻonohonoho hoʻomaʻamaʻa ʻana i ka puʻunaue maʻamau, akā aia i ka hoʻonohonoho hoʻāʻo nā anomalies.
Hoʻokumu ka mīkini vector kākoʻo papa hoʻokahi i kahi ʻili nonlinear a puni ke kumu. Hiki ke hoʻonoho i ka palena ʻoki ʻia i manaʻo ʻia he anomalous.
Ma muli o ka ʻike o kā mākou hui DATA4, ʻo One-Class SVM ka algorithm i hoʻohana pinepine ʻia no ka hoʻoponopono ʻana i ka pilikia o ka ʻimi ʻana i nā anomalies.
2) Hoʻokaʻawale i ke ʻano o ka nahele
Me ke ʻano "random" o ke kūkulu ʻana i nā kumulāʻau, e komo nā emissions i nā lau i ka wā mua (ma kahi hohonu pāpaʻu o ka lāʻau), ʻo ia hoʻi. ʻoi aku ka maʻalahi o ka "hoʻokaʻawale." ʻO ka hoʻokaʻawale ʻana o nā waiwai anomalous e loaʻa i nā iterations mua o ka algorithm.
3) Elliptic envelope a me nā ʻano helu helu
Hoʻohana ʻia ke puʻunaue mau ʻia ka ʻikepili. ʻOi aku ka pili o ke ana ʻana i ka huelo o ka hui ʻana o nā puʻunaue, ʻoi aku ka anomali o ka waiwai.
Hiki ke hoʻokomo pū ʻia nā ʻano hana helu ʻē aʻe i kēia papa.
Kiʻi mai dyakonov.org
4) Nā ʻano metric
Loaʻa nā ʻano algorithm e like me k-kokoke hoalauna, k-kokoke loa, ABOD (angle-based outlier detection) a i ʻole LOF (local outlier factor).
He kūpono inā like ka mamao ma waena o nā waiwai i nā hiʻohiʻona a i ʻole maʻamau (i ʻole e ana i ka boa constrictor i nā parrots).
Manaʻo ka algorithm o nā hoalauna kokoke loa e loaʻa nā waiwai maʻamau i kekahi wahi o ka multidimensional space, a ʻoi aku ka mamao o nā anomalies ma mua o ka hyperplane hoʻokaʻawale.
5) Nā ʻano hui
ʻO ke ʻano o nā ʻano huipuʻu inā ʻoi aku ka nui o kahi waiwai ma mua o ka nui o ka mamao mai nā kikowaena cluster, hiki ke manaʻo ʻia he anomalous ka waiwai.
ʻO ka mea nui ka hoʻohana ʻana i kahi algorithm e hoʻopili pololei i ka ʻikepili, e hilinaʻi ana i ka hana kikoʻī.
6) ʻO ke ʻano ʻāpana kumu
He kūpono kahi i hōʻike ʻia ai nā kuhikuhi o ka hoʻololi nui loa o ka hoʻopuehu ʻia.
7) Algorithms ma muli o ka wānana manawa
ʻO ka manaʻo inā hāʻule kahi waiwai ma waho o ka wā hilinaʻi wānana, manaʻo ʻia ka waiwai he anomalous. No ka wānana ʻana i kahi kaʻina manawa, hoʻohana ʻia nā algorithm e like me triple smoothing, S(ARIMA), boosting, etc.
Ua kūkākūkā ʻia nā algorithm wānana manawa ma ka ʻatikala mua.
8) Ke aʻo ʻana i mālama ʻia (regression, classification)
Inā ʻae ka ʻikepili, hoʻohana mākou i nā algorithms mai ka regression linear a i nā pūnaewele recurrent. E ana kāua i ka ʻokoʻa ma waena o ka wānana a me ka waiwai maoli, a e huki i ka hopena i ka nui o ka haʻalele ʻana o ka ʻikepili mai ka maʻamau. He mea nui ka algorithm e lawa ai ka mana generalization a ʻaʻole i loaʻa i ka hoʻonohonoho hoʻonaʻauao nā waiwai anomali.
9) Nā hōʻike hoʻohālike
E hoʻokokoke kākou i ka pilikia o ka ʻimi ʻana i nā anomalies ma ke ʻano he pilikia o ka ʻimi ʻana i nā ʻōlelo paipai. E hoʻokaʻawale i kā mākou hiʻohiʻona matrix me ka hoʻohana ʻana i ka SVD a i ʻole nā mīkini factorization, a lawe i nā waiwai i loko o ka matrix hou i ʻokoʻa loa mai nā mea mua me he anomalous.
Kiʻi mai dyakonov.org
hopena
Ma kēia ʻatikala, ua nānā mākou i nā ala nui e ʻike ai i ka anomaly.
ʻO ka ʻike ʻana i nā anomalies hiki ke kapa ʻia ma nā ʻano he nui. ʻAʻohe algorithm kūpono a i ʻole kahi ala, ʻo ka hoʻohana ʻana e hoʻonā i nā pilikia āpau. Hoʻohana pinepine ʻia kahi ʻano hana e hoʻoponopono i kahi hihia kikoʻī. Hoʻohana ʻia ka ʻike anomaly me ka hoʻohana ʻana i nā mīkini vector kākoʻo hoʻokahi papa, hoʻokaʻawale i nā ululāʻau, metric a me nā kumu cluster, a me ka hoʻohana ʻana i nā mea nui a me ka wānana manawa.
Inā ʻike ʻoe i nā ala ʻē aʻe, e kākau e pili ana iā lākou ma nā ʻōlelo i ka ʻatikala.
Source: www.habr.com