Isizukulwana esilandelayo seemodeli eziqikelelwayo sifika ngoncedo lweGoogle DeepMind kunye noPhando lukaGoogle kunye ne-WeatherNext 2, uyilo lwe-AI eyilelwe ukuzisa uqikelelo oluluncedo noluchanekileyo. Kwimeko yezehlo zemozulu eziye zanda ngokugqithisileyo, ukukwazi ukulindela iimeko ezingenakwenzeka kodwa ezinempembelelo ephezulu Iba ngundoqo kucwangciso.
Phakathi kweenkqubela zayo eziphawulekayo, le nkqubo inokuvelisa uqikelelo 8 amaxesha ngokukhawuleza kunye nesisombululo ukuya kuthi ga kwiyure enyeIkwavelisa amakhulu eevolutions zeatmospheri ezinokwenzeka ukusuka kwindawo enye yokuqala. Ngokweemvavanyo zangaphakathi, idlula imodeli yangaphambili kwi I-99,9% yeenguqu kunye nexesha lexesha ukusuka kwi-0 ukuya kwiintsuku ze-15.
Yintoni i-WeatherNext 2 kwaye kutheni ibalulekile?
I-WeatherNext 2 yimodeli yoqikelelo olunikwe amandla e-AI edityaniswe kuyo undoqo wenkqubo kaGoogle yokuqikelelaYile nto enika amandla imisebenzi yemozulu kwiimveliso zabo. Oku kuguqulela ukuphuculwa okubonakalayo kubasebenzisi bokugqibela kunye namacandelo amaninzi axhomekeke emoyeni.
UGoogle sele eqalisile ukusebenzisa obu buchwepheshe ukuphucula uqikelelo kuPhendlo, iGemini, kunye nePixel Weather, ngezicwangciso Ndiphinde ndafikelela kwiiMaphu zikaGoogle I-API yayo yemozulu sele idityanisiwe kwiqonga lemephu. Ukwenziwa komntwana abe ngowakho ngokuthe ngcembe kwaye kugxile ekunikezeni ulwazi olungaguquguqukiyo noluneenkcukacha.
Indlela esebenza ngayo: indlela yokusebenza kweNethiwekhi yokuVelisa
Ukusebenza kweWeatherNext 2 kuxhomekeke kwindlela ebizwa INethiwekhi eSebenzayo yokuVelisa (FGN)Le ndlela ingenisa ingxolo ngokuthe ngqo kwi-architecture yemodeli, ukwenzela ukuba izibikezelo zihlale ziyinyani kwaye zihambelana nomzimba, ukuphepha ukungahambi kakuhle phakathi kwezinto eziguquguqukayo.
I-FGN iqeqesha inkqubo kwizinto ezibizwa ngokuba yi-meteorologists "i-marginals," oko kukuthi, izinto ezizimeleyo ezifana nobushushu obuthile, isantya somoya, okanye ukufuma. Ukusuka apho, imodeli ifunda ukuxela kwangaphambili "amalungu." iinkqubo ezintsonkothileyo nezinxibeleleneyo, njengeendawo ezinkulu zobushushu okanye imveliso elindelekileyo yefama yomoya.
Ukuqala kwimo enye yokuqala ye-atmospheri, i-WeatherNext 2 yenza amakhulu eendlela ezinokwenzekaEzi zifaniso zibuyiselwa kwimodeli ngokwayo ukuhlanganisa ukuhambelana phakathi kwezinto eziguquguqukayo kunye nemimandla, ukubamba zombini ukuguquka okunokwenzeka kunye nokugqithisa okubaluleke kakhulu kulawulo lomngcipheko.
Ngaphantsi kwe-hood, inkqubo ibala ezine 6-iyure shift ngosukuihlala iqala kweyona meko yamva nje yehlabathi. Le cadence ivumela uluhlu lweemeko ukuba luhlaziywe njengoko i-atmosphere itshintsha kwaye kuvela iimpawu ezintsha.
Isantya, isisombululo, kunye nokuchaneka
Kwimeko yokubala, ukutsiba kubalulekile: Uqikelelo ngalunye lubaleka ngaphantsi komzuzu kwi-TPU enye, ngelixa iindlela ezisekelwe kwiimodeli zomzimba eziqhelekileyo zinokufuna iiyure kwii-supercomputers zemisebenzi ethelekisekayo.
Ngaphandle kwesantya, ibonelela isisombululo sexeshana ukuya kutsho kwiyure enye kunye nophuculo oluphawulekayo kwizinto eziguquguqukayo ezifana nobushushu, umoya, ukufuma, imvula, kunye noxinzelelo. Ngokubanzi, idlula imodeli yangaphambili kwi-99,9% yezinto eziguquguqukayo kunye namaxesha amaninzi ukusuka ku-0 ukuya Iintsuku ze15ukuvula umnyango wezigqibo ezinolwazi.
Enye inzuzo kukukwazi ukubandakanya uluhlu olupheleleyo lwamathuba, kuquka ukuba nokwenzeka okuphantsi kodwa iziganeko ezinempembelelo ephezuluLe ndlela enokwenzeka ibalulekile kwimisebenzi yokucwangcisa kunye neemeko ezingxamisekileyo ngaphandle kokujongela phantsi iimeko ezigqithisileyo.
Apho uya kuyibona kunye nendlela yokufikelela kuyo
UGoogle udibanise iWeatherNext 2 kwi Khangela, iGemini kunye nePixel Weather appkwaye kulindeleke ukuba ifike kwiiMephu zikaGoogle kunye neqonga leMozulu yeAPI. Abasebenzisi baya kuqaphela uqikelelo oluneenkcukacha ngakumbi kunye noluhambelanayo yonke imihla.
Kumashishini, izazinzulu, kunye nabaphuhlisi, ukufikelela kufumaneka nge I-Google Cloud Vertex AI, iBigQuery kunye neNjini yoMhlabaLe ndlela yenza kube lula ukuphonononga iiseti zedatha, ukwakha usetyenziso olungqale kushishino, kunye nokuvavanya imingcipheko ngegranularity enkulu.
Imephu yendlela ibandakanya ukudibanisa imithombo emitsha yedatha kunye nokwandisa ukufikelela, ngenjongo yokukhawulezisa uphando kunye nokwenza izisombululo ezixhomekeke kuqikelelo oluphezulu.
Impembelelo enokubakho eSpain naseYurophu
Kwimixholo efana neSpain kunye neYurophu iphela, uqikelelo olukhawulezayo noluchaneke ngakumbi lunokuguqulelwa ulawulo oluphuculweyo lwamaza obushushu, izaqhwithi zomoya kunye nemvula enkuluAmacandelo afana namandla, ukushukumiseka, okanye ezolimo zikholisa ukuxhamla kwixesha elingaguqukiyo leentsuku eziyi-0-15.
Ngokuvelisa iimeko ezininzi, i-WeatherNext 2 ivumela uvandlakanyo lwe iimeko ezimbi kakhulu kunye nokusasazwa kweziphumoOku kuluncedo ngakumbi kukhuseleko loluntu, abaqhubi beziseko ezingundoqo, kunye nokucwangciswa kobutyebi bamanzi xa kujongwe iziganeko ezigqithisileyo.
Kwicandelo lamandla ahlaziyekayo, ukuphuculwa kwezinto eziguquguqukayo ezifana nomoya kunye ne-cloud cover kunceda ukuqikelela umoya kunye nokuveliswa kwamandla elanga ngeenkcukacha ezininzi zexesha, ukwandisa ukusebenza kwenkqubo yombane kunye nokunciphisa ukungaqiniseki ekuboneleleni.
Imodeli entsha idibanisa isantya, isisombululo, kunye nokungaguquguquki komzimba ukuhambisa Uqikelelo oluluncedo ngakumbi kubasebenzisi kunye nemibuthoUkudityaniswa kwayo kwiimveliso zikaGoogle kunye nokufikelela ngelifu kuvula ukusetyenziswa kwayo kwiimeko ezininzi, ukusuka ekuhlolweni kwemozulu yemihla ngemihla ukucwangcisa imisebenzi echaphazela imozulu.
