
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Ababhali: 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Shengyu Tao Haizhou Liu Chongbo Sun Haocheng Ji Guanjun Ji Zhiyuan Han Runhua Gao Jun Ma Ruifei Ma Yuou Chen Shiyi Fu Yu Wang Yaojie Sun Yu Rong Xuan Zhang Guangmin Zhou Hongbin Sun Isishwankathelo Iibhetri ezingalungisiweyo ezishenxisiweyo ezineendlela ezahlukeneyo zekathode zithintela ukwamkelwa kokurisayiklishwa ngokuthe ngqo ngenxa yendalo yazo ethile yekathode. Ukwanda kwee-bhetri ezishenxisiweyo kufuna ukuhlelwa ngokuchanekileyo ukuze kusetyenziswe ngokufanelekileyo ukurisayiklishwa ngokuthe ngqo, kodwa imingeni ivela kwiimbali zokusebenza ezahlukeneyo, abavelisi abahlukeneyo, kunye neengxaki zobumfihlo bedatha ababambisene nabo bokurisayiklisha (abanini bedatha). Apha sibonisa, ukusuka kwiseti yedatha eyahlukileyo ye-130 yee-bhetri ze-lithium-ion ezihlanganisa izinto ezisi-5 zekathode kunye nabavelisi abasi-7, indlela yokufunda umatshini eyahluleleneyo ingakwazi ukucalula ezi bhetri zishiyekileyo ngaphandle kokuxhomekeka kwidatha yokusebenza yangaphambili, ikhusela ubumfihlo bedatha bababambisene nabo bokurisayiklisha. Ngokusebenzisa izinto ezikhutshiweyo ukusuka kumjikelo wokutshaja-ukukhupha wokuphela kobomi, imodeli yethu ibonisa i-1% kunye ne-3% iimpazamo zokuhlela ikathode phantsi kweemeko zokurisayiklisha ze-bhetri ezifanayo kunye nezingafaniyo ngokulandelelana, ezibangelwa yindlela yethu entsha yokuvota nge-Wasserstein-distance. Ngokwezobuchwepheshe, indlela ecetywayo igxininisa ixabiso lokuhlelwa ngokuchanekileyo kwee-bhetri ukuze kube nobuchwepheshe bokurisayiklisha obunempumelelo nozinzileyo. Esi sifundo sibhengeza imodeli entsha yokusebenzisa idatha ye-privacy-sensitive ukusuka kwimithombo eyahlukeneyo, kuququzelela ukwenza izigqibo ngokubambisana nokuhlonipha ubumfihlo kwiinkqubo ezisasazekileyo. Isingeniso Iibhetri ze-Lithium-ion (LIBs), ezisebenza njengezixhobo zokugcina amandla, ziye zasetyenziswa ngokubanzi kwiindawo ezahlukeneyo, ukusuka kwimveliso yoshishino ukuya ebomini bemihla ngemihla njengendlela eyammkeliweyo yobuchwepheshe. Iingxelo zibonisa ukuba isikali semveliso yehlabathi ye-LIBs sidlule i-1.3 TWh ngo-2030  xa imfuno ephakamayo yee-bhetri idlula kude ukufumaneka kwezixhobo eziphambili zentsimbi ezifana ne-lithium kunye ne-cobalt , . Nangona kunjalo, ubomi obuqhelekileyo bwangoku beeemveliso ze-LIB bumalunga neminyaka emi-5–8, nto leyo ekhokelela ekwanda okukhulu kwee-bhetri ezishenxisiweyo kumazwe amaninzi. Ukuba azilawulwa ngokufanelekileyo, iibhetri ezishenxisiweyo ziya kubangela inkunkuma yezixhobo engazinzanga kunye nomonakalo wokusingqongileyo. Ngokwale meko, ukuphuhliswa kobuchwepheshe bokurisayiklisha ibhetri kufuna ukubaluleka okukhulu xa sijongana nezaa nkqubela yee-LIB ezilindelekileyo . 1 2 3 4 Uphuhliso lwamva nje kuphando malunga nokurisayiklishwa kwee-bhetri lugxile kwiindlela ze-pyrometallurgical, ze-hydrometallurgical, kunye ne-direct recycling . Ngokungafaniyo neendlela ze-pyrometallurgical kunye neze-hydrometallurgical, ukurisayiklishwa ngokuthe ngqo kuyahluka njengendlela eyahlukileyo. Le nkqubo ayibangeli monakalo wesibini kwisakhiwo sezinto, ivumela ukulungiswa okusebenzayo kunye nokubuyiselwa kokusebenza. Ngaphezu koko, ukurisayiklishwa ngokuthe ngqo kubonisa inzuzo ephezulu, njengoko kuthatha ukusetyenziswa kwamandla okuphantsi, ukuncitshiswa kokukhutshwa kwe-greenhouse gas, kunye neendawo ezikhaphukhaphu zokusingqongileyo , . Kwiimveliso zangempela, nangona kunjalo, ababuyiseli bamathumbu bahlala bekhawuleza ii-LIBs ezinezingxenye ezingaziwayo okanye iipakke zebhetri eziqukethe umxube weentlobo ezahlukeneyo zekathode. Ngokucinga ukuba ukurisayiklishwa ngokuthe ngqo kunokuxhomekeka kakhulu kwikathode, ubunzima obunjalo benza isicelo sokurisayiklisha ngokuthe ngqo singasebenzi ukuze kufumaneke ixabiso kwiibhetri ezishenxisiweyo . Kubalulekile ukugxininisa ukuba nkqu neebhetri ezibalulekileyo ezivela kwiintlobo ezixubeneyo zekathode zinokukhupha usebenzisa iindlela zokurisayiklisha eziqhelekileyo, ukusebenzisana phakathi kweentlobo ezahlukeneyo zekathode ngexesha lenkqubo yokurisayiklisha kungathatha umgangatho wemveliso. . Ngoko ke, ukuqonda ulwazi olubalulekileyo lwekathode kwicala lokurisayiklisha kuchaphazela kakhulu ukhetho lwendlela yokurisayiklisha ngokuthe ngqo kwaye ekugqibeleni kuphucula umgangatho wemveliso, inzuzo, kunye nokuzinza. 5 6 7 8 9 Ukufunyanwa kweenkcukacha zohlobo lwekathode lwebhetri elishiyekileyo kulungiswa kwinkqubo yokulungisa kwangaphambili, kodwa ayisebenzi ngokuphathelele imali xa ushishino lokurisayiklisha lukhula . Ukufumana ngokufanelekileyo ulwazi lwekathode yohlobo lwebhetri elishiyekileyo, uluntu lwenzululwazi kunye noshishino luye lwaqala inkqubo yokulandelela ubomi bebhetri  kunye neengcamango ezivelayo ezifana nephepha lebhetri  kunye ne-genome yedatha yebhetri . Nangona iibhetri ezininzi ziye zasetyenziswa phambi kwezi ntlanganiso, kukho isivumelwano esikhulayo sokuba ulwazi lwebhetri kufanele lufumaneke kuyo yonke inkqubo yobomi ukulungiselela ukwenza izigqibo zokujikeleza kwesibini . Oku kunokwenzeka ngokukodwa kwicandelo lokurisayiklisha kwebhetri, isikhululo sokugqibela sobomi besibini bebhetri, njengoko indlela yokurisayiklisha inokuxhomekeka kakhulu kwikathode. Nangona kunjalo, iinkqubo zokulandelela ubomi bebhetri okanye iiphepha lebhetri zenziwa zisebenze kusetyenziswa izixhobo zombane ezifana neebhakhowudi kunye neenkomunikeysen ezikufutshane, ezingabonelela ngotyalo-mali olunzima kwaye zingahambelani ngokubanzi nabayili beebhetri ezahlukeneyo. Ngaphezu koko, izixhobo zombane zisazama ukulawulwa ngokuqinisekileyo kubomi bazo bonke, nto leyo ekhokelela kwiizixhobo ezidliweyo kunye nokungafumaneki kwisigaba sokurisayiklisha njengoko inkqubo yemveliso yemodeni ye-LIB ingadibanisi imveliso nokurisayiklisha . Ngoko ke, izinto ezintsha ezininzi zifunwa kakhulu ukuba kufunyanwe ukuhlelwa okusebenzayo kwekathode yohlobo lwebhetri kusetyenziswa ulwazi olufumaneka lula kwintsimi kuphela , , ngokuphambene nedatha yembali ebhalisiweyo okanye ngendlela yenkxaso yabantu, kuququzelela ukwamkelwa kokurisayiklisha ngokuthe ngqo ukuphucula umgangatho kunye nenzuzo yeemveliso ezigayisiweyo. 1 10 11 12 13 14 15 16 Kule minyaka mihlanu idlulileyo, ukufunda umatshini kuye kwavela njengezixhobo ezinokwenzeka ukujongana nemibuzo evulekileyo kuzo zonke iinkalo zebhetri. Kwezinye izihloko eziphathelene neebhetri, ukufunda umatshini kuthe kuvumele ukuba sifumanise ngokuzenzekelayo iinkqubo ezintsonkothileyo zebhetri , , , ukuchaza ubomi obushiyekileyo obusebenzayo , , , , , ukuvavanya imeko yempilo , , , ukwenza isikali sokujikeleza , , ukuchaza ukusasazwa kwesiphelo , nkqu nokukhokhela uyilo lwebhetri , , kwaye uchaze ukusebenza okungunaphakade ngokukhawuleza emva kokwenziwa . Kwimeko yokurisayiklisha ibhetri, imisebenzi embalwa iye yajonga ukufunda umatshini malunga nezinto zekathode , , oku kubangelwa yidatha enqabileyo yebhetri, ngakumbi kwabo bajikeleze kude kube sekupheleni kobomi. Ubuninzi bezifundo ezipapashwe zibonisa izisayizi ezilinganiselweyo  kwaye ziphantsi kakhulu kwii-bhetri ezahlukeneyo zekathode . Ubunqabileyo bubangelwa yindleko enzima, ixesha elide lokuvavanya , kwaye okubaluleke kakhulu, ubumfihlo bedatha ngenxa yeengxaki zorhwebo okanye izilangazelelo. Ngokungafaniyo, umcimbi wobumfihlo uqinisekisa idilemma apho idatha ebhetri ekhoyo, nangona inkulu ngomthamo kunye nobume obahlukeneyo ukusuka kumacala amaninzi afana nabavelisi bebhetri, izicelo ezisebenzayo, imibutho yezifundo, kunye namaqonga emithathu, ayinakwabelwana. Idilemma enjalo ibiza ukuphanda ngokuhlela ikathode ukuze kwenziwe isikali sokurisayiklisha kwebhetri ngendlela yokubambisana ngelixa ikhusela ubumfihlo. 17 18 19 20 21 22 23 24 19 25 26 27 28 29 30 31 32 33 34 35 36 37 I-Federated machine learning, njengemodeli esasazekileyo nekhusela ubumfihlo, inamandla okusombulula zombini intsebenziswano yamacala amaninzi (elingana, umthamo wedatha yebhetri kunye nobume) kunye neengxaki zobumfihlo ngokusebenzisa ukufunda koomatshini ngokubambisana , , . Kulo lonke ulandelelwano lokufunda, abanini bedatha abasazekileyo benza ufundiso lwasekhaya ngamandla abo ezobuchwepheshe, bakrokra iiparameters/imiphumo yemodeli efundisiweyo, kwaye bazithumele kumququzeleli ophakathi ukuze zihlanganiswe. Iinyani ezibonisa ukuba iiseti zedatha eziluhlaza aziphumi kubanini bazo bedatha kunye neeparameter/imiphumo edluliselweyo ikhrokriwe ngokufanelekileyo ukukhusela ubumfihlo bedatha. I-Federated machine learning iye yajongwa ngokubanzi kumacala amaninzi okusebenzisa, kuquka impilo kawonke-wonke , , ukuxilonga izifo , , , urhwebo nge-intanethi , i-Internet of Things , ikhompyuter yeselula , kunye nesikiti esihlakaniphile , , . Le ndlela ingenza utshintsho kwimodeli yophando oluqhutywa yidatha kumacala amakhulu amandla ngokuvumela intsebenziswano ekhusela ubumfihlo, ngakumbi kwabo bane-access elinganiselweyo kwidatha. Malunga necandelo lokurisayiklisha ibhetri, i-federated machine learning ithathela ingqalelo amathuba amakhulu okusebenzisa inani elikhulu ledatha yebhetri ekhoyo kodwa engenakwabelwana ngenxa yeengxaki zobumfihlo. Ngale modeli yokubambisana nokhusela ubumfihlo, ukuhlela ibhetri elishiyekileyo kunokwenziwa ngokuchanekileyo okuphezulu, ukusebenza, ukwanda, kunye nokugqibelela, kwenziwe lula umgangatho kunye nenzuzo yeemveliso ezigayisiweyo. Ngokwazi kwethu, izifundo ze-federated machine learning ezigxile ekurisayiklishweni kwebhetri azikaze zixelelwe. 38 39 40 41 42 43 44 45 46 47 48 49 50 51 Kolu phando, senze ukuhlelwa kwekathode yezinto ezishiyekileyo ibhetri, sisebenzise idatha ekhoyo yebhetri ukusuka kubabambisene abaninzi, njengabavelisi bebhetri, abaqhubi bezicelo ezisebenzayo, imibutho yophando lwezifundo, kunye namaqonga emithathu, ngendlela yokufunda umatshini yokubambisana nokhusela ubumfihlo njengoko kuchaziwe kuMfanekiso 1. Imodeli yethu ye-federated machine learning ifundisiwe kusetyenziswa umjikelo omnye wedatha yovavanyo lwentsimi ngokusebenzisa inkqubo yokuvelisa izinto eyiyo, ngaphandle kolwazi olungaphambili lwembali yemeko yokusebenza. Siyayithelekisa amandla okuqikelela emodeli yethu ye-federated machine learning kunye neyemodeli zasekhaya ezifundiswa ngokuzimeleyo ngokusekelwe kwidatha yasekhaya phantsi kweemeko zokurisayiklisha ze-bhetri ezifanayo nezingafaniyo. Umcimbi wobungafaniyo usonjululwe yindlela yethu ecetywayo yokuvota nge-Wasserstein-distance. Ukuvavanywa kwezobuchwepheshe kokurisayiklishwa kwebhetri elishiyekileyo kusetyenziswa isakhiwo sethu se-federated machine learning, sigxininisa ukubaluleka kunye nokufuneka kokuhlelwa ngokuchanekileyo kwee-bhetri ezishiyekileyo. Siyayichaza ngokupheleleyo intsingiselo yemodeli, iingcinga zokurisayiklisha ibhetri, kunye nembono ebanzi yenkqubo ezayo yokurisayiklisha ibhetri edibaniswe ne-federated machine learning.    Iimithombo ezininzi zedatha, ezifana nabavelisi bebhetri (Image courtesy Addionics), abaqhubi bezicelo ezisebenzayo (ipakeji yebhetri kwindawo esezantsi yeTesla. Image courtesy of Tesla), imibutho yophando lwezifundo, kunye namaqonga emithathu, ingaba ngabanikeli bedatha. Ibhola yedatha ayiguquleli phakathi kwabanikeli okanye ithunyelwe kumthengisi webhetri. Endaweni yoko, abanikeli bedatha baqeqeshela iimodeli zasekhaya kwaye babelane ngeeparameter zemodeli nomthengisi webhetri ukwakha imodeli yehlabathi. Indlela ecetywayo yokuvota nge-Wasserstein-distance idibanisa iimodyeli zasekhaya kwi-model yehlabathi, ezinzileyo kukunqongophala kwedatha kunye ne-noise. Abathengisi bebhetri bangasebenzisa imodeli edityaniswe ngokudibeneyo ukuze kuhlelwe ibhetri, idibene nedatha yovavanyo lwentsimi ekulula ukuyifumana.   Isakhiwo sethu se-federated machine learning sinika inkuthazo kubabambisene ukuba babelane ngedatha ngelixa bekhusela ubumfihlo bedatha ngokungafaniyo nemodeli yemveli yesiqithi sedatha. a b Iziphumo Uqokelelo lwedatha kunye nokwenziwa ngokomgangatho Iinkqubo zebhetri ezahlukeneyo kwiintlobo ezahlukeneyo zebhetri zidla ngokuba ne-dimensional ephezulu kwaye kunzima ukuyichaza ngenxa yeemeko zokusebenza ezahlukeneyo, ukwahluka kwemveliso, kunye nokusetyenziswa kwembali . Ukufumana isisombululo kwidilemma le, siye saqokelela kwaye senza ngokomgangatho i-130 iibhetri ezishiyekileyo ezineentlobo ezi-5 zekathode ezivela kubavelisi abasi-7 ukwakha idatha engaphandle kokusasazwa, ngokulinganayo idatha engafaniyo. Ngenxa yezinto ezahlukeneyo zokusetyenziswa kwembali, amandla eebhetri eziqokelelweyo angaphantsi kwe-90% yamandla akalelweyo. Izinto zekathode zebhetri yi-lithium cobalt oxide (LCO), i-nickel manganese cobalt (NMC), i-lithium ferrophosphate (LFP), i-nickel-cobalt-aluminum oxide (NCA), kunye neentlobo ezixubeneyo ze-NMC-LCO, ezizeziqhubekekiswe zaba ngamaqela asi-9 ngokusekelwe kubavelisi (Itheyibhile Encedisayo 1). Siyayibandakanya ngabom iibhetri ezineembali ezahlukeneyo zokusetyenziswa, ukusuka kuvavanyo lwelebhu ukuya kwimidlalo yokuqhuba izithuthi zombane, ukuze sifundise imodeli egqibeleleyo kumthengisi webhetri engazibopheleli kwimbali yokusetyenziswa kunye neentlobo zebhetri. 52 Ukuze senziwe ngokomgangatho, yonke idatha efunwa ngumthengisi ngumjikelo wangoku (ovavanywa entsimini) kunye novavanyo olunye lokutshaja kunye nokukhupha, okungekho lula ukuyiphumeza kwiimeko eziphambili. Idatha efunyiweyo yokuqala iyacocwa ngokuzalisa iziphaluka ezilahlekileyo, ukubuyisela izinto ezingaphandle, kunye nokwenza uhlolo lwe-median. Iingxaki ezibangelwa ngabantu kunye nezingafaniyo zekathode zigcinwa ngabom, nangona kunjalo, ukwenza imodeli izinzile kwiingxaki ezingapheleleyo. Idatha ke ibe yindawo efanayo yokuzalisa umgca (uMfanekiso Encedisayo 1) kwaye iphuculwe ukuze kuncitshiswe ubuninzi, kunye neseti edibeneyo yee-parameters zenkqubo yomgangatho (Inqaku Eli ncedisayo 1). Izinto ezikhutshiweyo kwinkqubo yenkqubo yomgangatho ziyabonakala, into ebaluleke kakhulu kwiimarike zorhwebo. Ngokwazi kwethu, yeyona ndlela yokuqala apho idatha yeebhetri ezahlukeneyo ezivela kumacala amaninzi kunye nembali yokusetyenziswa isetyenziswa ukuncedisa uyilo lwesicwangciso sokurisayiklisha ibhetri. Umganekiso 2a, 2b bonisa inkqubo yokuvelisa izinto. Sigxile kumgca wokutshaja kunye nokukhupha weebhetri ezishiyekileyo kumjikelo wokugqibela, oko kukuthi, umjikelo omnye wokutshaja kunye nomjikelo omnye wokukhupha (iifigures ezingezizo zangoko-2–5). Kumjikelo wokutshaja, izinto ezili-15 zikhutshiwe kumgca we-voltage-capacity kunye nomgca we-dQ/dV, apho u-V kunye no-Q bebhekisa kumaxabiso voltage kunye ne-capacity, ngokulandelelana. Isitshixo esifanayo sezinto zikhutshiwe kumjikelo wokukhupha. Ngenxa yoko, izinto ezingama-30 zikhutshiwe ngokuphakama, njengoko kuboniswe ukusuka ku-F1 ukuya ku-F30. Jonga Itheyibhile Encedisayo 2 kunye neNqaku Eli ncedisayo 2 ukuze ufumane inkcazo eneenkcukacha zezinto. Umfanekiso 2c ubonisa amaxabiso ezinto ezibalulekileyo kunye nezabelo zezinto ezikhethiweyo ezivela kwiqela ngalinye. Uninzi lwama-coefficients amathambo ezinto ezibalulekileyo kwiindidi ezahlukeneyo zidibanisa kwingingqi ye-−1 ukuya ku-0 (ngombala oluhlaza okukhanyayo) kwaye azinakwahlulwa, zibonisa ubunzima bokucalula uhlobo lwebhetri ngokusebenzisa umjikelo omnye wedatha yebhetri. Ubunzima kulindelekile kuba iimeko zokusebenza ezahlukeneyo zinokuchaphazela iinkqubo zokutshaja-ukukhupha kwee-bhetri ukuze izinto ezikhutshiweyo zibe neqhina kakhulu nangona zineentlobo zebhetri ezahlukeneyo (uMfanekiso Encedisayo 6). Endaweni yokutolika ngokuthe ngqo izinto ezikhutshiweyo kusetyenziswa ulwazi lweg

Part of HackerNoon's growing list of open-source research papers, promoting free access to academic material.

Le audio iveliswe ngolwimi lokuqala lwebali!

LeNketho lobumfihlo ye-AI ingaguqula inkunkuma yebhetri ibe yinzuzo enkulu

About Author

IZIMVO

ZIJONGE IIMPAWU

ELI NQAKU LINIKEZELWE KU

Related Stories

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps