Garken Dazuzzukan Bishiyoyi na Tsarin Kima na Goyon Bayan Vector Regression don Hasashen Wutar Lantarki ta Rana

Teburin Abubuwan Ciki

1. Gabatarwa & Bayyani

Wannan takarda, "Garken Dazuzzukan Bishiyoyi na Tsarin Kima na Goyon Bayan Vector Regression don Hasashen Wutar Lantarki ta Rana," ta magance wata kalubale mai mahimmanci a cikin tsarin wutar lantarki na zamani: rashin tabbas da rashin ci gaba na samar da wutar lantarki ta hasken rana (PV). Yayin da shigar makamashi mai sabuntawa cikin grid ya ƙaru, ingantaccen hasashe ya zama mafi mahimmanci don kiyaye kwanciyar hankali, inganta tanadin aiki, da ba da damar ayyukan kasuwa masu inganci. Marubutan sun ba da shawarar sabon tsarin haɗakarwa mai matakai biyu wanda ke amfani da ƙarfin dabarun koyon inji guda biyu da aka kafa: Kima na Goyon Bayan Vector Regression (SVR) don samar da hasashe na farko da kuma Dazuzzukan Bishiyoyi (RF) a matsayin babban mai koyo don haɗa waɗannan hasashe da inganta su.

Babban ƙirƙira ya ta'allaka ne a amfani da RF ba don sarrafa bayanan yanayi na danye ba, amma don yin sarrafa bayanai ko haɗakar hasashe. Garken RF yana shigar da hasashe daga tsarin SVR da yawa (ta amfani da hasashe na yanzu da na baya) tare da bayanan yanayi masu dacewa don samar da ingantaccen hasashen wutar lantarki ta rana mai zuwa. Wannan hanyar ta wuce sauƙaƙan matsakaici ko haɗa bayanan yanayi, tana nufin ɗaukar hadaddun mu'amala marasa layi tsakanin hanyoyin hasashe daban-daban.

2. Hanyar Aiki & Tsarin Fasaha

2.1 Tsarukan Koyon Injin na Asali

Kima na Goyon Bayan Vector Regression (SVR): Ana amfani da SVR a matsayin mai hasashe na asali. Yana aiki ta hanyar nemo aiki $f(x) = w^T \phi(x) + b$ wanda ya bambanta da ainihin manufa $y_i$ da aƙalla ƙimar $\epsilon$ (bututu marar hankali-epsilon), yayin da yake kasancewa mai laushi kamar yadda zai yiwu. An tsara wannan a matsayin matsala mai ma'ana, yana mai da shi mai ƙarfi ga yin wuce gona da iri, musamman tare da bayanai masu girma kamar haɗaɗɗun fasali na yanayi da tarihin wutar lantarki.

Dazuzzukan Bishiyoyi (RF): Ana amfani da RF a matsayin mai haɗawa. Yana aiki ta hanyar gina ɗimbin bishiyoyin yanke shawara yayin horo da kuma fitar da matsakaicin hasashe (don regression) na kowane bishiya. Ƙarfin sa na asali na sarrafa alaƙar da ba ta layi ba, darajar muhimmancin fasali, da ba da ƙarfi ga hayaniya ya sa ya zama mai kyau don gane waɗanne hasashen SVR (kuma a ƙarƙashin waɗanne sharuɗɗa) suka fi aminci.

2.2 Tsarin Haɗakarwa na Garken

Tsarin da aka tsara garken ne:

1. Mataki na 1 (Masu Hasashe na Asali): Ana horar da tsarin SVR da yawa, mai yuwuwa ta amfani da ma'auni daban-daban, saitin fasali na shigarwa (misali, jinkirin wutar lantarki, zafin jiki, hasken rana), ko tagogin horo. Kowane yana samar da hasashen rana mai zuwa.
2. Mataki na 2 (Babban Mai Koyo): Ana horar da tsarin Dazuzzukan Bishiyoyi. Abubuwan shigarsa (fasali) sune hasashe daga duk tsarin SVR na Mataki-1 don lokacin da aka yi niyya, tare da ainihin bayanan yanayi (fitowar NWP) na wannan lokacin. Fitowarsa (manufa) ita ce ainihin wutar lantarki ta rana da aka lura. RF yana koyon yin ma'auni da haɗa hasashen SVR da kyau bisa ga yanayin yanayi na yanzu.

Wannan hanyar ta fi na gargajiya na matsakaicin ƙira, saboda RF na iya koyon ma'auni masu dogaro da mahallin, yana aiwatar da zaɓin hasashe mai hankali da gyara.

3. Tsarin Gwaji & Sakamako

3.1 Bayanan Gwaji & Ma'aunin Kimantawa

Binciken mai yiwuwa yana amfani da bayanan tarihi na shekara guda daga tsarin wutar lantarki ta rana (PV), gami da fitar da wutar lantarki da madaidaicin masu canjin yanayi (hasken rana, zafin jiki, rufe gajimare). Bayanan Hasashen Yanayi na Lissafi (NWP) suna aiki a matsayin babban shigarwa don hasashen rana mai zuwa. Ana kimanta aikin ta amfani da ma'auni na kuskure na yau da kullun kamar Tushen Matsakaicin Kuskuren Square (RMSE), Matsakaicin Kuskure na Cikakke (MAE), da kuma mai yiwuwa Matsakaicin Kuskuren Kashi na Cikakke (MAPE), tare da kwatanta tsarin haɗakarwa da tsarin SVR ɗaya ɗaya da sauran dabarun haɗawa na ma'auni (misali, sauƙaƙan matsakaici, regression na layi mai ma'auni).

3.2 Nazarin Aiki & Kwatanta

Takardar ta ba da rahoton cewa garken RF-SVR ya fi duka tsarin SVR na ginshiƙansa da sauran hanyoyin haɗawa a tsawon lokacin kimantawa na shekara. Wannan yana nuna cewa dabarun haɗawa mara layi na RF ta yi nasarar ɗaukar mu'amalar da masu haɗawa na layi suka rasa. Sakamakon ya tabbatar da hasashen cewa haɗakar hasashe ta hanyar babban mai koyo mai ƙarfi na iya fitar da ƙarin siginar hasashe daga tarin hasashe daban-daban amma masu alaƙa.

Bayyani na Ginshiƙi (Ra'ayi): Taswirar sanduna za ta nuna ƙimar RMSE/MAE don: a) Tsarin dagewa, b) Mafi kyawun tsarin SVR guda ɗaya, c) Matsakaicin tsarin SVR, d) Haɗawar regression na layi, e) Garken RF-SVR da aka tsara. Sandar RF-SVR za ta kasance mafi gajarta, tana nuna mafi girman daidaito. Taswira mai ƙari na layi na iya nuna hasashe da ainihin wutar lantarki don mako mai wakilci, yana nuna inda garken ya gyara kurakuran da kowane tsari ya yi.

4. Nazari Mai Zurfi & Ra'ayi na Masana'antu

Fahimta ta Asali: Aikin Abuella da Chowdhury aiki ne mai aiki, mai mayar da hankali kan injiniya, ba ci gaba na ka'ida ba. Sun gane cewa a cikin duniyar hasashen hasken rana mai rikitarwa, babu "mafi kyau" tsari guda ɗaya. Maimakon neman unicorn, sun tura "kwamitin ƙwararru" (SVR da yawa) da "shugaban kwamiti mai hankali" (Dazuzzukan Bishiyoyi) don haɗa mafi kyawun amsa mai yuwuwa. Wannan ba game da ƙirƙirar sabon AI ba ne, amma game da sarrafa kayan aiki na yanzu, waɗanda aka gwada da yaki—alamar balaga a cikin amfani da ML don tsarin makamashi.

Kwararar Hankali & Ƙarfi: Hankali yana da inganci kuma yana kwatanta mafi kyawun ayyuka a gasar ML (kamar GEFCom2014 da aka ambata). Ƙarfinsa yana cikin sauƙi da sake yin samfurinsa. SVR da RF suna samuwa ko'ina, ana fahimtar su da kyau, kuma suna da sauƙin daidaitawa idan aka kwatanta da madadin koyo mai zurfi. Tsarin matakai biyu kuma yana ba da fahimta: muhimmancin fasalin RF na iya bayyana wane tsarin SVR (ko ma'aunin yanayi) ya fi tasiri a ƙarƙashin takamaiman sharuɗɗa, yana ba da fahimtar aiki mai mahimmanci fiye da lambar hasashe baƙar fata.

Kurakurai & Iyakoki: Bari mu yi magana a sarari: wannan hanya ce ta 2017. Tsarin yana da jeri da tsayayye. Tsarin SVR an daidaita su kafin a horar da RF, sun rasa damar ingantawa har zuwa ƙarshe wanda garken koyo mai zurfi na zamani (misali, amfani da hanyoyin sadarwar jijiya a matsayin masu koyo na asali da manyan masu koyo) zai iya bayarwa. Hakanan yana buƙatar ƙirar fasali mai mahimmanci kuma yana iya fuskantar wahala tare da bayanai masu yawan mitar sosai ko ɗaukar dogaro na sararin samaniya-lokaci mai rikitarwa a cikin rukunin PV da aka rarraba—kalubale inda Hanyoyin Sadarwar Jijiya na Zane (GNN) ke nuna alƙawari a yanzu, kamar yadda aka gani a cikin wallafe-wallafen baya-bayan nan daga cibiyoyi kamar National Renewable Energy Laboratory (NREL).

Fahimta Mai Aiki: Ga ƙungiyoyin hasashen aiki, wannan takarda ta kasance tsarin aiki don nasara mai sauri. Kafin nutsewa cikin koyo mai zurfi mai rikitarwa, aiwatar da wannan garken RF-on-SVR. Aikin ne mai ƙarancin haɗari, babban yuwuwar dawowa. Ainihin fahimta ita ce a ɗauki matakin "haɗakar hasashe" a matsayin muhimmin sashi na tsarin. Saka hannun jari don ƙirƙirar saitin hasashe na asali daban-daban (ta amfani da algorithms daban-daban, hanyoyin bayanai, da tsarin da ke da ilimin kimiyyar lissafi) sannan a yi amfani da mai haɗawa mara layi mai ƙarfi kamar RF ko Gradient Boosting. Wannan hanyar modular tana ba da tabbacin tsarin ku na gaba; zaku iya musanya sabbin ƙirar asali (kamar LSTM ko Transformer) yayin da suka tabbatar da ƙimarsu, yayin da kuke riƙe da ingantaccen tsarin haɗawa.

5. Cikakkun Bayanai na Fasaha & Tsarin Lissafi

Tsarin SVR: Idan aka ba da bayanan horo ${(x_1, y_1), ..., (x_n, y_n)}$, SVR yana warwarewa: $$\min_{w, b, \xi, \xi^*} \frac{1}{2} ||w||^2 + C \sum_{i=1}^n (\xi_i + \xi_i^*)$$ bisa ga: $$y_i - (w^T \phi(x_i) + b) \le \epsilon + \xi_i,$$ $$(w^T \phi(x_i) + b) - y_i \le \epsilon + \xi_i^*,$$ $$\xi_i, \xi_i^* \ge 0.$$ Anan, $\phi(x)$ yana zuwa sarari mafi girma, $C$ shine ma'aunin tsari, kuma $\xi_i, \xi_i^*$ sune masu canjin sallama.

Hasashen Dazuzzukan Bishiyoyi: Don regression, hasashen RF $\hat{y}_{RF}$ don vector shigarwa $\mathbf{z}$ (wanda ya ƙunshi hasashen SVR da bayanan yanayi) shine matsakaicin hasashe daga bishiyoyi $B$ guda ɗaya: $$\hat{y}_{RF}(\mathbf{z}) = \frac{1}{B} \sum_{b=1}^{B} T_b(\mathbf{z})$$ inda $T_b$ shine bishiyar yanke shawara ta $b$.

6. Tsarin Nazari: Nazarin Lamari na Ra'ayi

Yanayi: Mai sarrafa grid na yanki yana buƙatar haɗa hasashe daga tsarin wutar lantarki ta rana (PV) 50 da aka rarraba.

Aiwatar da Tsarin:

1. Mataki na Asali (Tsarin SVR): Horar da tsarin SVR guda uku don kowane wuri (ko tsarin duniya):
   • SVR_Phys: Yana amfani da bayanan NWP (hasken rana, zafin jiki) a matsayin fasali na farko.
   • SVR_TS: Yana mai da hankali kan fasalin lokaci-lokaci (jinkirin wutar lantarki, ranar mako, sa'a na rana).
   • SVR_Hybrid: Yana amfani da saitin fasali da aka haɗa.
2. Mataki na Meta (Dazuzzukan Bishiyoyi): Don sa'a da aka yi niyya gobe, shigarwa zuwa RF shine vector: $\mathbf{z} = [\hat{P}_{SVR\_Phys}, \hat{P}_{SVR\_TS}, \hat{P}_{SVR\_Hybrid}, GHI_{NWP}, Temp_{NWP}, CloudCover_{NWP}]$. RF, wanda aka horar da shi akan bayanan tarihi, yana fitar da ƙarshen haɗakar hasashe $\hat{P}_{Final}$.
3. Fitowa: Hasashe mafi daidaito da ƙarfi. Nazarin muhimmancin fasalin RF na iya bayyana cewa a ranakun da gajimare ke rufe sama, tsarin lokaci-lokaci (SVR_TS) yana samun ƙaramin ma'auni, yayin da tsarin da ke da ilimin kimiyyar lissafi (SVR_Phys) da bayanan rufe gajimare suka zama mafi mahimmanci.

Wannan tsarin yana ba da hanya mai tsari, ta atomatik don amfani da bambancin ƙira.

7. Aikace-aikace na Gaba & Hanyoyin Bincike

Ka'idodin wannan aikin sun wuce hasashen wutar lantarki ta rana:

• Hasashen Wutar Lantarki ta Iska: Aiwatar kai tsaye ta amfani da garken samfurin hasashen saurin iska daban-daban.
• Hasashen Kaya: Haɗa hasashe daga samfurin kaya na tattalin arziki, lokaci-lokaci, da koyon inji.
• Hasashe Mai Yuwuwa: Haɓaka mai haɗawa na RF don fitar da tazara na hasashe (misali, ta amfani da dazuzzukan regression na quantile) maimakon kawai hasashe na batu, wanda ke da mahimmanci don ayyukan grid masu sane da haɗari.
• Haɗawa tare da Koyo Mai Zurfi: Maye gurbin SVR da LSTM ko Maɓallan Haɗakar Lokaci a matsayin masu koyo na asali, da kuma amfani da Hanyar Sadarwar Jijiya a matsayin babban mai koyo, wanda aka horar da shi har zuwa ƙarshe. Bincike a wannan shugaban yana aiki, kamar yadda aka gani a cikin takardu daga manyan tarurruka kamar NeurIPS da ICLR.
• Lissafin Geza don Rarraba PV: Tura sigogi masu sauƙi na wannan tsarin garken don hasashe na ainihin lokaci a matakin inverter ko mai tara.

Gaba yana cikin garken masu motsi, masu daidaitawa waɗanda za su iya ci gaba da koyo da sabunta ma'aunin haɗawa a cikin kusa da ainihin lokaci yayin da sabbin bayanai da ayyukan ƙira ke gudana.

8. Nassoshi

1. Abuella, M., & Chowdhury, B. (2017). Random Forest Ensemble of Support Vector Regression Models for Solar Power Forecasting. A cikin Proceedings of Innovative Smart Grid Technologies, North America Conference.
2. Hong, T., Pinson, P., & Fan, S. (2016). Global Energy Forecasting Competition 2014. International Journal of Forecasting, 32(2), 896-913.
3. National Renewable Energy Laboratory (NREL). (2023). Hasashen Wutar Lantarki ta Rana. An samo daga https://www.nrel.gov/grid/solar-forecasting.html
4. Breiman, L. (2001). Dazuzzukan Bishiyoyi. Koyon Injin, 45(1), 5-32.
5. Smola, A. J., & Schölkopf, B. (2004). Darasi akan kima na goyon bayan vector regression. Kididdiga da Lissafi, 14(3), 199-222.
6. Zhu, J.-Y., Park, T., Isola, P., & Efros, A. A. (2017). Fassarar Hoton-da-Hoto mara haɗin gwiwa ta amfani da Hanyoyin Sadarwar Adawa na Ci gaba. Proceedings of the IEEE International Conference on Computer Vision (ICCV). (An ambata a matsayin misalin ingantaccen tsarin koyo mara layi).
7. Binciken baya-bayan nan akan Hanyoyin Sadarwar Jijiya na Zane don hasashen sararin samaniya-lokaci a cikin tsarin wutar lantarki (misali, daga taron GM na IEEE PES).