Modelling and worldwide assessment of performance and aging of photovoltaic modules and systems

UNIVERSITY OF LJUBLJANA

FACULTY OF ELECTRICAL ENGINEERING

Julián Andrés Ascencio Vásquez

Modelling and worldwide assessment of performance and aging of photovoltaic modules and systems

Doctoral dissertation

Ljubljana, 2021

Julián Andrés Ascencio Vásquez

Modelling and worldwide assessment of performance and aging of photovoltaic modules and systems

Doctoral dissertation

Supervisor Prof. Dr. Marko Topič

Ljubljana, 2021

Julián Andrés Ascencio Vásquez

Modeliranje in globalno vrednotenje učinkovitosti delovanja ter staranja fotonapetostnih modulov in

sistemov

Doktorska disertacija

Mentor Prof. Dr. Marko Topič

Ljubljana, 2021

IZJAVA

Doktorska disertacija je plod samostojnega dela.

Pomoč mentorja in sodelavcev sem navedel v zahvali.

Julián Andrés Ascencio Vásquez

STATUTORY DECLARATION

I declare that I have authored this thesis independently, that I have not used other than the declared sources, resources and that I have explicitly marked all material which has been quoted either literally or by content from the used sources.

Julián Andrés Ascencio Vásquez

THESIS COMMITTEE

Prof. Dr. Janez Krč, Chairman

University of Ljubljana - Faculty of Electrical Engineering, Ljubljana/Slovenia

Assoc. Prof. Dr. Marko Jankovec, Member

University of Ljubljana - Faculty of Electrical Engineering, Ljubljana/Slovenia

Prof. Dr. Ralph Gottschalg, Member

Fraunhofer Center for Crystalline Silicon Photovoltaics, Halle/Germany

Prof. Dr. Marko Topič, Mentor

University of Ljubljana - Faculty of Electrical Engineering, Ljubljana/Slovenia

ACKNOWLEDGEMENTS

Special thanks to my mentor Prof. Dr. Marko Topič, who gave me the chance, motivation, and inspiration to complete a fruitful doctoral training. He has led me through the field of photovoltaic research and industry, pushed me to achieve ambitious challenges, supported me in the bad times, joined celebrations on the good times, and, for sure, wisely advised me on my next steps in the solar professional career. Thank you for everything.

I would also address special thanks to Dr. Kristjan Brecl, who has been my day-to-day mentor through unnumerous technical (and not too technical) discussions. Your support inside and outside the working place made my Slovenian stay simply remarkable! Thank you for showing confidence, a critical review of all scientific inputs and constructive suggestions, tips, and ideas.

Also, I would like to thank all my colleagues at LPVO, who have contributed to the pleasant, fun, and very high-quality work environment since day one. Especially thanks to Stefan Mitterhofer, with whom I spent three years of arduous work, and of course, extraordinary times after work in Ljubljana and many other cities in Europe. Thank you to Ms. Olga Zakrajsek, Mr. Joze Stepan, Marko Jankovec, Rok Kimovec, Andrej Campa, Matevz Bokalic, Jost Balent, Milan Kovacic, Janez Krc, Bostjan Glozar, Benjamin Lipovec, and the entire LPVO for sharing with me your knowledge, advice, friendship, and for showing and teaching me the best of the Slovenian culture.

I gratefully acknowledge my colleagues and mentors during the SOLAR-TRAIN project, who were not only colleagues but also excellent partners in numerous training schools, conferences, secondments, etc. Special thanks go to Ismail Kaaya, Sascha Lindig, Chiara Baretta, Luis Castillon, Guillermo Oviedo-Hernandez, Djamel Monsour, Karl-Anders Weiss, David Moser, Nikola Hrelja, Mike van Iseghem, Paolo V. Chiantore, Gernot Oreski, Ralph Gottschalg, Nikoleta Kirinaki, Ana Rosa Lagunas, Amantin Panos Mixeli, Aziz Nairi, Ashenafi W. G, and Tom Betts.

I also thank many people I crossed during my doctoral training at Fraunhofer ISE, EDF R&D, PCCL, EURAC, Loughborough University, CENER, and BayWa r.e. for the warm hospitality and support in my research.

After very exciting collaborations and inspiring discussions during the doctoral training, I also want to thank Juan Carlos Osorio, Marios Theristis, and Jakob Bevc for supporting and participating in my research.

Gracias mamá – Patricia, papá – Julián por su apoyo infinito en este largo viaje lejos de casa!

Very special thanks go to all my FAMILIA (Rita Rubio, Ercilia Vergara, Maria Garcés, Juan Vásquez, and Julián Ascencio M.) and AMIG@S for all the support and encouragement throughout this journey. Thank you to Gabija Bacinskaite, for your constant support and motivation to pursue my dreams. Thank you to Diego Barrera, Maria Patricia Ascencio, Victor Ascencio, Ignacio Molina, Bastian Lotina and many more important people who contributed to

holding me high, and everyone else who helped me in any way and supported me during my studies.

For financial support, thank you to the European Union for supporting and funding this PhD work through the H2020 programme SOLAR-TRAIN under grant agreement No 721452. I would like to thank the Public Research Agency of the Republic of Slovenia and the Public Fund of the Republic of Slovenia for the development of human resources and scholarships

ABSTRACT

Solar Photovoltaic (PV) modules and systems keep improving performance and reliability, and at the same time, reducing their costs. PV plants are being installed all over the world, and installed capacity is growing exponentially. Those indicators get better thanks to research and learning in the PV industry. Thus, the performance and aging of PV modules and PV systems installed at different periods, locations, and climates are the main focus of this doctoral dissertation.

PV technologies' energy yield, optimal operation, and service lifetime are highly weather- dependent (irradiance, temperature, among others). Manufacturers typically offer indoor testing results under controlled environmental conditions, but once devices are brought outdoors for operation, the behaviours could highly differ due to local uncontrollable weather and climate conditions. This doctoral thesis aims to go one step ahead in understanding, modelling and assessing PV performance and aging of PV modules and PV systems worldwide.

Combining the ground-based measurements with Geographical Information System (GIS) datasets, the mapping of performance and aging of PV modules and PV systems is aimed. The most relevant factors impacting the incident irradiance and PV performance behaviour are studied for this task, identifying accurate empirical models to generate global layers of the key performance indicators (KPIs). Quantitative indicators and risk maps are produced, visually presenting locations where energy yield is higher and where degradation modes are more prone to occur with higher probability.

Since PV modules are being installed at practically any location and climate globally, developing a scheme to standardize the evaluation of regions with similar climate conditions is considered in this thesis. The mapping of climate variables and stresses will be overlapped to create new climate zones regarding PV modules' performance and degradation. Those results simplify the understanding of the impact of climate on PV performance and degradation.

Moreover, the development of modelling algorithms and combination with GIS datasets are further exploited to simulate and compare PV systems' performance under different climate change scenarios taken for the latest literature.

The second chapter of this doctoral thesis focuses on the environmental variables that PV systems can experience at any global location. For optimal PV performance, incident irradiance, ambient temperature and wind speed play the most crucial role to achieve a high energy yield.

For the PV aging assessment, ultraviolet (UV) irradiance, thermal stresses and humidity are considered critical variables in service lifetime predictions (SLP). Here, data sources are identified and processed. Also, the modelling and validation of the mentioned weather variables are given.

The most relevant weather variables are then combined with the theory for PV modelling in terms of performance and aging. In the third chapter of this doctoral thesis, Empirical, Physical and Machine Learning models are given for the comprehensive modelling of PV technologies,

including thermal modelling, effective irradiance taking into account spectral losses, reflectance losses, soiling losses and snow losses, energy yield modelling for different technologies and degradation models for assessing the degradation rates of PV systems. This chapter achieves a broad overview of the essential parameters and theories to consider when analysing a PV system at any lifecycle stage.

Also, in the third chapter, the development of algorithms for advanced data analytics are presented to support the assessing real-time monitoring of weather and electrical data coming from weather stations and PV facilities. Those algorithms will help to check and filter raw datasets. Additionally, a clear definition of the data type will be proposed (unrealistic values, unstable measurements, etc.).

In the fourth chapter, the performance and aging of PV modules are studied. Firstly, the processing, exploration and exploitation of outdoor Current-Voltage (I-V) characteristics of three commercial polycrystalline silicon (pcSi) PV modules are presented. Each measured I-V curve is processed with advanced data techniques considering the different stages: filtering, fitting, parameter extraction, normalization, and long-term evaluation. In this chapter, a novel technique called “Empirical Weather Normalization” (EWN) is proposed and validated for all different I-V curve parameters.

In a parallel task and collaboration with Fraunhofer ISE, the uncertainties of physical degradation models for outdoor PV modules are assessed in different climate zones. For this purpose, different weather models for UV irradiance, PV Module Temperature, and Relative Humidity are benchmarked and used to quantify the PV module degradation rates and failure times in Ljubljana (Slovenia), Zugspitze (Germany), Gran Canaria (Spain) and Negev (Israel).

In the fifth chapter, the performance and aging of PV systems are evaluated. A novel numerical algorithm called “Typical Daily Profiles” (TDP) is proposed for this task. The algorithm is used to evaluate the PV performance of the entire fleet in Chile. Also, the potential and current state of the Chilean PV industry is given. Thanks to the TDP approach, critical technical aspects can be summarized for fixed and tracking PV systems at the country level.

Besides, a comprehensive evaluation of a 17kW PV system installed on the Faculty of Electrical Engineering's rooftop, University of Ljubljana, Slovenia, is carried out. Here, common data issues in monitoring systems are mentioned and considered for further PV performance modelling using heuristic approaches and Machine Learning approaches at different irradiance accuracy levels. The Light Gradient Boosting Machine (LightGBM) approach is highlighted as a high-speed and accurate modelling technique, proving that novel machine learning techniques could be integrated into PV systems' performance loss detection pipelines.

In a more global scope, which focus on categorizing geographical regions in terms of PV performance aspects, the sixth chapter covers the mapping of climate zones, performance, aging and weather risks for PV systems. An improvement of the Köppen-Geiger (KG) climate zone scheme is achieved by developing and publishing the Köppen-Geiger-Photovoltaic (KGPV) climate classification. The KGPV enhances the original KG with solar irradiation to the already temperature and precipitation weather variables. The models presented in the previous chapter

are the basis for developing maps for PV energy yield, Performance Ratio and PV module temperature, and PV Degradation Maps, including hydrolysis-degradation and photo-degradation thermo-mechanical degradation, and total degradation rates of PV modules. Then, the PVWeatherMaps are developed following a strict framework from weather data to risk maps.

The compilation of risk maps includes Snow Shading, Soiling Deposition and Thunderstorm risks. Finally, the correlation of PV indicators and climate zones, together with the application of PVWeatherMaps on an extensive and global PV portfolio, are presented.

The last technical chapter introduces climate change (i.e., climate evolution) aspects into the performance and aging of PV modules and PV systems. First, the illustration of ambient temperature and KGPV zones towards 2100 under the SSP585 climate scenario is given, followed by the evolution of PV performance indicators for a typical polycrystalline silicon (pcSi) PV system and the increase of PV degradation rates ambient temperature over the last decades.

Finally, a novel approach called “Climate Change Yield Assessment” (CCYA) is introduced, showing the framework of further yield assessments, including not only historical data (e.g., TMY) but also the latest climate change scenarios.

Most of the content in this thesis has been published or submitted in recognized international scientific journals under the following titles:

1) “Methodology of Köppen-Geiger-Photovoltaic climate classification and implications to wor ldwide mapping of PV system performance”, published in Solar Energy [1],

2) “Global climate data processing and mapping of degradation mechanisms and degradation rates of PV modules”, published in Energies [2],

3) “Advanced PV performance modelling based on different levels of irradiance data accuracy”, published in Energies [3],

4) “Assessment of uncertainties and variations in PV modules degradation rates and lifetime predictions using physical models”, published in Solar Energy [4],

5) “Outdoor PV module degradation assessment based on the Empirical Weather Normalization applied to I-V curves”, submitted to IEEE Journal of Photovoltaics [5], 6) “Typical Daily Profiles, a novel approach for photovoltaics performance assessment: case

study on large-scale systems in Chile”, published to Solar Energy [6],

7) “Global patterns for PV performance assessment and weather risk identification”, to be submitted to high impact factor scientific journal [7],

8) “Climate Change Yield Assessment (CCYA) for Solar Resource and Photovoltaic Systems”, to be submitted to high impact factor scientific journal [8].

RAZŠIRJENI POVZETEK Uvod

Solarni fotonapetostni moduli (PV), ki prihajajo na trg, iz leta v leto izboljšujejo zmogljivost in zanesljivost ter hkrati znižujejo svoje stroške. Sončne fotonapetostne elektrarne se pospešeno gradijo po vsem svetu, pri čemer kumulativna nameščena nazivna moč sončnih elektrarn raste eksponentno. Tako zmogljivost kot zanesljivost sta osrednja tema raziskav in izboljšav v fotonapetostni industriji ter sta v središču raziskav te doktorske disertacije.

Učinkovitost pretvrobe, optimalno delovanje in življenjska doba fotonapetostnih tehnologij so zelo odvisni od vremenskih pogojev (med drugim od sončnega obsevanja, temperature zraka ipd.). Proizvajalci običajno ponujajo rezultate preskusov tipov PV modulov v nadzorovanih laboratorijskih pogojih, toda ko so moduli postavljeni na prosto, se lahko njihovo delovanje in učinkovitost močno razlikuje zaradi lokalnih nenadzorovanih vremenskih in podnebnih razmer.

Namen te doktorske disertacije je narediti korak naprej pri razumevanju, modeliranju in ocenjevanju zmogljivosti in staranju PV modulov in PV sistemov po vsem svetu, tako na konktretnih tipih PV modulov kot posplošeno na globalnem nivoju.

Povzetek

Uvodoma se osredotočimo na meteorološke podatke in podatkovne zbirke, njihovo obdelavo, modeliranje in medsebojno validacijo. Namen združevanja zemeljskih meritev meteoroloških parametrov z nabori podatkov Geografskega informacijskega sistema (GIS) je v celoviti obravnavi obratovalnih pogojev in preko njih priti do teritorialne analize učinkovitosti in staranja PV modulov in PV sistemov. V tem sklopu preučujemo najpomembnejše dejavnike, ki vplivajo na sončno obsevanje in delovanje PV naprav, pri čemer določimo natančne empirične modele za generiranje globalnih slojev ključnih kazalnikov uspešnosti (KPI). V analizi pripravimo kvantitativne kazalnike in karte tveganj, ki vizualno prikažejo območja, kjer je donos energije večji in kjer je stopnja degradacije zmogljivosti večja.

Ker so PV moduli nameščeni skoraj povsod, na vseh možnih lokacijah po svetu ne glede na raznolikost podnebij, smo v okviru disertacije razvili novo metodologijo in klasifikacijsko shemo za standardizirano vrednotenje regij s podobnimi obratovalnimi in podnebnimi razmerami. S kartiranjem podnebnih spremenljivk in njihovega vpliva na PV naprave smo zasnovali nova podnebna območja glede učinkovitosti delovanja in degradacije PV modulov. Ti rezultati so poenotili in poenostavili razumevanje vpliva podnebja na delovanje in staranje fotonapetostnih naprav. Poleg tega smo razvite algoritme za modeliranje v kombinaciji z nabori podatkov GIS

izkoristili za simulacijo in primerjavo delovanja PV sistemov v prihodnosti v luči različnih scenarijev podnebnih sprememb, povzetih po najnovejši literaturi.

V drugem poglavju je poudarek na okoljskih spremenljivkah, ki jih lahko PV sistemi izkusijo na kateri koli globalni lokaciji. Za doseganje visokih izkoristkov energije so za doseganje optimalne PV zmogljivosti najpomembnejše vpadno obsevanje, temperatura okoliškega zraka in hitrost vetra. Za oceno staranja PV se ultravijolično (UV) obsevanje, toplotni stresi in vlaga štejejo za kritične spremenljivke v napovedih življenjske dobe (SLP). Tu so viri podatkov identificirani in obdelani. Podana sta tudi modeliranje in validacija omenjenih vremenskih spremenljivk.

V tretjem poglavju te vremenske spremenljivke nato združimo s teorijo za PV modeliranje v smislu zmogljivosti in staranja. Za celovito modeliranje PV tehnologij podamo empirične in fizikalne modele ter modele strojnega učenja, vključno s toplotnim modeliranjem, efektivnim obsevanjem, ki upošteva spektralne izgube, izgube na račun odboja svetlobe, umazanije ali snega.

Modeliranje vključuje modeliranje donosa energije v poljubnem časovnem obdobju za različne tehnologije in modele staranja za oceno stopnje degradacije PV sistemov. Razvoj splošnih algoritmov za oceno sprotnega spremljanja vremenskih in električnih podatkov, ki prihajajo iz vremenskih postaj in PV naprav, smo podprli z napredno analitiko podatkov z metodami strojnega učenja in umetne inteligence. Ti algoritmi bodo pomagali preveriti in filtrirati neobdelane nabore podatkov. Poleg tega bo predlagana jasna opredelitev vrste podatkov (kot so nerealne vrednosti, nestabilne meritve itd.).

V četrtem poglavju smo proučili zmogljivosti in staranje PV modulov. Najprej smo predstavili analizo treh komercialnih PV modulov iz polikristalnega silicija (pcSi), ki temelji na izmerjenih tokovno-napetostnih (I-V) karakteristikah pod kontroliranimi pogoji. Vsaka izmerjena krivulja I- V je obdelana z naprednimi podatkovnimi tehnikami, ki upoštevajo različne stopnje: filtriranje, prilagajanje, ekstrakcija parametrov, normalizacija in dolgoročno vrednotenje. V poglavju predlagamo in potrdimo novo tehniko, ki smo jo poimenovali "Empirična vremenska normalizacija" (EWN), za vse zmogljivostne parametre krivulje I-V.

V partnerski doktorski disertaciji Ismaila Kayaa v okviru projektnega sodelovanju s Fraunhofer ISE smo ocenjevali negotovosti modelov fizične degradacije PV modulov v različnih podnebnih pasovih. V ta namen smo primerjali različne vremenske modele za ultravijolično (UV) obsevanje, temperaturo PV modula in relativno vlažnost, ki se uporabljajo za kvantificiranje stopnje degradacije in/ali časa do odpovedi PV modula na različnih lokacijah različnih podnebij: v Ljubljani (Slovenija), Zugspitze (Nemčija), Gran Canaria (Španija) in Negevu (Izrael).

V petem poglavju se posvetimo ocenjevanju zmogljivosti in staranja PV sistemov. Za to nalogo predlagamo nov numerični algoritem, imenovan "Tipični dnevni profili" (TDP). Algoritem se

uporablja za ovrednotenje zmogljivosti celotne flote sončnih elektrarn v Čilu (skupne nazivne moči 2,6 GW). Zahvaljujoč pristopu TDP lahko kritične tehnične vidike vrednotenja povzamemo na ravni države za fiksne in sledilne PV sisteme, ki jih TDP algoritem avtomatsko razpoznava in ločuje. Poleg tega se izvede celovita ocena 17kW PV sistema, nameščenega na strehi Fakultete za elektrotehniko, Univerza v Ljubljani, Slovenija. Tu so omenjena in obravnavana pogosta vprašanja s podatki v sistemih za spremljanje za nadaljnje modeliranje PV zmogljivosti z uporabo hevrističnih pristopov in pristopov strojnega učenja pri različnih stopnjah natančnosti obsevanja.

Pristop Light Gradient Boosting Machine (LightGBM) je poudarjen kot hitra in zelo natančna tehnika modeliranja.

Šesto poglavje zajema kartiranje podnebnih pasov ter zmogljivosti, staranja in vremenskih tveganj za PV sisteme. Izboljšanje sheme podnebnega pasu Köppen-Geiger (KG) smo dosegli z razvojem klasifikacijske sheme Köppen-Geiger-Photovoltaic (KGPV). KGPV dopolnjuje oz.

nadgrajuje prvotno shemo KG s sončnim obsevanjem. Tudi modeli, predstavljeni v predhodnih poglavjih, so osnova za razvoj globalnih kart za energijski donos in zmogljivostno razmerje PV naprav, kot tudi globalnih kart za temperaturo PV modulov in globalnih kart degradacije PV, vključno s hidrolizno-razgradnjo, foto-razgradnjo, termo-mehansko razgradnjo in skupno stopnjo degradacije učinkovitosti PV modulov. Vse skupaj strnemo v PVWeatherMaps, ki so rezultat razvoja po rigoroznih korakih od vremenskih podatkov do zemljevidov oz. kart tveganj. Takšna priprava kart tveganj vključuje nevarnosti senčenja na račun snega v zimskih mesecih, nalaganja umazanije zaradi prisotnosti prahu in pomanjkanja padavin ter na račun ekstermnih mehanskih obrementive zaradi močnih neviht. Na koncu poglavja je predstavljena korelacija kazalnikov PV in podnebnih območij, skupaj z uporabo PVWeatherMaps na obsežnem in globalnem portfelju PV.

Zadnje vsebinsko poglavje (7.) prinaša analizo vpliva podnebnih sprememb (tj. Podnebni razvoj) na učinkovitost delovanja in staranje PV modulov in PV sistemov. Najprej podamo ilustracijo razvoja temperature okolice in območij KGPV do leta 2100 v podnebnem scenariju SSP585, čemur sledi razvoj kazalnikov delovanja PV za tipični PV sistem in povečanje stopnje degradacije PV zaradi povišanja temperature okolice v zadnjih desetletjih. Na koncu je predstavljen nov pristop, imenovan “Ocena donosa podnebnih sprememb“ (CCYA – Climate Change Yield Assessment), ki prikazuje okvir nadaljnjih ocen energijskega donosa, ki vključuje ne samo zgodovinske podatke (npr. TMY), temveč tudi najnovejše scenarije podnebnih sprememb.

• Rezultati doktorske disertacije so bili objavljeni v naslednjih izvirnih znanstvenih člankih, izdanih v mednarodnih revijah s faktorjem vpliva (JCR IF): J. Ascencio- Vásquez, K. Brecl and M. Topič, “Methodology of Köppen-Geiger-Photovoltaic climate classification and implications to worldwide mapping of PV system performance”, Solar Energy 2019, 191, 672–685 [impact factor (2020): 5.742].

• J. Ascencio-Vásquez, I. Kaaya, K. Brecl, K.A. Weiss and M. Topič, “Global climate data processing and mapping of degradation mechanisms and degradation rates of PV modules”, Energies 2019, 12, 4749 [impact factor (2020): 3.004].

• J. Ascencio-Vásquez, J. Bevc, K. Reba, K. Brecl, M. Jankovec and M. Topič,

“Advanced PV performance modelling based on different levels of irradiance data accuracy”, Energies 2020, 13 (9), 2166 [impact factor (2020): 3.004]

• I. Kaaya, J. Ascencio-Vásquez, K.A. Weiss and M. Topič, “Assessment of uncertainties and variations in PV modules degradation rates and lifetime predictions using physical models”, Solar Energy 2021, 218, 354-367 [impact factor (2020): 4.5.742]

• J. Ascencio-Vásquez, J.C. Osorio, K. Brecl, E. Muñoz-Cerón and M. Topič,

“Typical Daily Profiles, a novel approach for photovoltaics performance assessment: case study on large-scale systems in Chile”, Solar Energy, 225, 1 September 2021, Pages 357-374 [impact factor (2020): 5.742]

Dodaten izvirni znanstveni članek je bil poslan v objavo v mednarodni reviji s faktorjem vpliva:

• J. Ascencio-Vásquez, K. Brecl, M. Jankovec and M. Topič, “Outdoor PV module degradation assessment based on the Empirical Weather Normalization applied to I-V curves”, Submitted to IEEE Journal of Photovoltaics [impact factor (2020):

3.887]

in dva članka sta v zaključni fazi izdelave:

• J. Ascencio-Vásquez, K. Brecl and M. Topič, “Global patterns for PV performance assessment and weather risk identification”, to be submitted to high impact factor scientific journal.

• J. Ascencio-Vásquez, K. Brecl, M. Topič, et al.“Climate Change Yield Assessment (CCYA) for Solar Resource and Photovoltaic Systems”, to be submitted to high impact factor scientific journal.

Rezultati so bili predstavljeni na številnih mednarodnih konferencah (kot predavanje ali poster) in delavnicah ter objavljeni v konferenčnih zbornikih:

• J. Ascencio-Vásquez, K. Brecl and M. Topič, “Geographical approach for weather risk identification and PV performance assessment”, Proceedings of the 37^th European Photovoltaic Solar Energy Conference and Exhibition, September 7^th – 11^th, 2020, Lisbon, Portugal.

• S. Lindig, J. Leloux, J. Ascencio-Vásquez, M. Topič and D. Moser, “Climate- related dependence of Performance Losses of over 3,500 PV Systems”, Proceedings of the 37^th European Photovoltaic Solar Energy Conference and Exhibition, September 7^th – 11^th, 2020, Lisbon, Portugal.

• K. Reba, J. Bevc, J. Ascencio-Vásquez, M. Jankovec and M. Topič, “Photovoltaic Energy Production Forecasting using LightGBM”, Proceedings of the 53rd

International Conference on Microelectronics, Devices and Materials, September 25^th - 27^th, 2019, Bled, Slovenia.

• L. Castillon, J. Ascencio-Vásquez, A.P. Mehilli, G. Oreski, M. Topič and K.A.

Weiss, “Parallel natural weathering of laminated backsheets across Europe”, Proceedings of the 36^th European Photovoltaic Solar Energy Conference and Exhibition, September 9^th – 13^th, 2019, Marseille, France.

• J. Ascencio-Vásquez, K. Brecl and M. Topič, “Validation of PV system performance modelling in view of Köppen-Geiger-Photovoltaic climate classification”, Proceedings of the 35^th European Photovoltaic Solar Energy Conference and Exhibition, September 24^th – 28^th, 2018, Brussels, Belgium.

• J. Ascencio-Vásquez, K. Brecl and M. Topič, “Köppen-Geiger-Photovoltaic climate classification”, Proceedings of the IEEE 7^th World Conference on Photovoltaic Energy Conversion, June 10^th – 15^th, 2018, Hawaii, USA.

• K.-A. Weiss, I. Kaaya, C. Barretta, J. Ascencio-Vásquez, N. Hrelja, L. Pitta Bauermann, E. Klimm, D. Moser, M. Topič, M. Van Iseghem, G. Oreski, T. Betts, A. R. Lagunas, P. Chiantore, “Training the Next Generation of PV Reliability Experts (Photovoltaic Life Time Forecast and Evaluation) – The Marie Sklodowska-Curie Actions (MSCA) Project SOLAR-TRAIN”, Proceedings of the IEEE 7^th World Conference on Photovoltaic Energy Conversion, June 10^th – 15^th, 2018, Hawaii, USA.

• J. Ascencio-Vásquez and M. Topič, “An Overall Data Analysis Methodology for PV Energy Systems”, Proceedings of the 53rd International Conference on Microelectronics, Devices and Materials, October 4^th - 6^th, 2017, Ljubljana, Slovenia.

Izvirni prispevki k znanosti

Ocenjujemo, da predložena doktorska disertacija vsebuje naslednje izvirne prispevke k znanosti:

• Splošni algoritem za oceno sprotnega spremljanja vremenskih in električnih podatkov

Razvoj novega splošnega algoritma za preverjanje in filtriranje neobdelanih podatkovnih nizov s PV mest (vključno s podnebnimi in električnimi spremenljivkami). Poleg tega smo izoblikovali jasno opredelitev vrste podatkov glede njihove uporabnosti/neuporabnosti (kot so nerealne vrednosti, nestabilne meritve, itd.).

• Kartiranje zmogljivosti in stopnje degradacije PV modulov in PV sistemov

Z združitvijo podnebnih podatkov, parametrov PV tehnologij in modelov v literaturi smo izdelali kvantitativne kazalnike in zemljevide oz. karte tveganj, ki vizualno prikazujejo lokacije, kjer je energijski donos večji ali kjer je stopnja degradacije višja.

• Globalna shema klasifikacije podnebnih sprememb za PV tehnologije

Uspešen razvoj sheme za standardizano vrednotenje regij s podobnimi podnebnimi razmerami. Kartiranje podnebnih spremenljivk in obremenitev se bo prekrivalo, da se bodo ustvarila nova podnebna območja glede učinkovitosti in degradacije PV modulov.

• Povprečni kazalniki učinkovitosti in stopnje degradacije za podnebna območja Rezultate zgornjih izvirnih prispevkov smo povezali in s povprečenjem poenostavili razumevanje vpliva podnebja na učinkovitost delovanje in degradacijo PV naprav.

• Projekcije PV zmogljivosti v različnih scenarijih podnebnih sprememb

S simulacijo različnih scenarijih podnebnih sprememb, upoštevanih v najnovejši literature, primerjamo delovanje PV sistemov do leta 2100 na globalnem nivoju.

Zaključek

Doktorska disertacija <<MODELIRANJE IN GLOBALNO VREDNOTENJE UČINKOVITOSTI DELOVANJA TER STARANJA FOTONAPETOSTNIH MODULOV IN SISTEMOV>> poglobljeno obravnava večino tem o oceni učinkovitosti in staranja PV modulov in PV sistemov. Analize smo izvajali ne le za nekaj lokacij, temveč smo v okviru disertacije razvili metodologijo ocenjevanja in vrednotenja tudi na vseh različnih podnebnih območjih po vsem svetu. Širok obseg te naloge omogoča celovito razumevanje delovanja PV sistemov v različnih časovnih okvirih in obdobjih ter kjer koli po svetu.

Najprej v tem doktorskem delu je podan podroben pregled vremenskih virov podatkov, ki so nam bile na razpolago. Predstavljeni so tudi fizični modeli za oceno lokalnih vremenskih dejavnikov, ki združujejo posebne vremenske postaje in globalne zemljevide. Podatkovne zbirke o podnebnih razmerah (ERA-Interim, ERA5, CAMS, med drugim) smo vključili v nadaljnje aplikacije analize PV. Globalno obsevanje je bilo potrjeno z uporabo enega najobsežnejših podatkovnih baz piranometrov po vsem svetu, imenovanega BSRN. To delo kaže, da podatkovni niz ERA5 Reanalysis in podobni viri (npr. Ponovna analiza in satelitski podatki) ponujajo pomembno alternativo pridobivanju velikega števila bistvenih spremenljivk za PV, kot so sončno obsevanje, temperatura zraka in hitrost vetra za modeliranje zmogljivosti PV, in UV obsevanje, nočno- dnevna temperaturna spremembe in relativna vlažnost za mehanizme razgradnje PV (hidroliza, termomehanska in fotorazgradnja) in ocene skupne stopenje degradacije.

Kar zadeva tehnike modeliranja PV, ostajajo široko uporabljene metode temperature PV modula, kot sta modela Ross in Faiman, zelo učinkovite rešitve za oceno temperature PV modulov.

Predstavili smo uporabo globalnih podatkov za kartiranje temperature PV modula za različne PV tehnologije, ki kažejo, kakšna bi lahko bila najvišja možna obratovalna temperatura PV modula (približno 125 stopinj Celzija). Te informacije lahko pomagajo tudi pri določanju natančnejših in bolj verodostojnih laboratorijskih preskusov za različne sestavne dele sončnih elektrarn.

Za simulacije energijskega donosa smo empirične modele primerjali s pristopi strojnega učenja.

Čeprav so empirični modeli lahko nekako povezani s fizičnimi lastnostmi PV sistemov, lahko napredna matematika umetnih nevronskih mrež, podpornih vektorskih strojev in pristopov za povečanje gradienta izboljšajo natančnost ocen. V tem delu smo uporabili zelo zmogljiv algoritem LightGBM in dokazali njegovo uporabnost v PV aplikacijah. Poleg tega je fizično modeliranje stopenj degradacije vroča tema v PV industriji; v disertaciji je poudarek na ocenjevanju negotovosti v različnih scenarijih (različna natančnost modela). Podana so tudi priporočila za še natančnejšo oceno stopnje degradacije.

Izvedli smo obsežno analizo na velikem naboru I-V krivulj treh PV modulov, ki delujejo v Ljubljani v Sloveniji, da bi izluščil razvoj notranjih parametrov zmogljivosti. V disertaciji predlagamo novo metodologijo, imenovano "Empirična normalizacija vremena" (EWN), ki pomaga objektivno primerjati različne krivulje I-V, izmerjene v različnih zunanjih pogojih in različnih časovnih oknih. Podamo stopnje degradacije preučevanih PV modulov skupaj s stopnjo evolucije vseh zmogljivostnih parametrov krivulje I-V, ekstrahiranih z modelom enojne diode.

Za oceno učinkovitosti PV sistemov sta predstavljeni dve študiji: velika skupina sončnih elektrarn v skupni moči 2,6 GW, ki deluje v Čilu, in visoko učinkovit 17 kW strešni PV sistem, ki deluje v Sloveniji. Po predlogu na novo razvite metodologije, imenovane “Tipični dnevni profili“ (TDP), je v Čilu ocenjenih več kot 160 PV sistemov. Za fiksne in sledilne PV sisteme so podane zanimive korelacije ključnih kazalnikov učinkovitosti v več podnebnih pasovih. Potencial PV v Čilu je izpostavljen ne le zaradi rekordnega letnega obsevanja, temveč tudi zelo visokega specifičnega energijskega donosa v primerjavi z evropskimi regijami.

Na primeru 17 kW PV sistema, ki deluje na strehi Fakultete za elektrotehniko Univerze v Ljubljani, smo empirične modele in modele strojnega učenja primerjali za različne stopnje natančnosti obsevanja. LightGBM je v ocenjenih scenarijih predstavljen kot zelo učinkovita rešitev za natančno in hitro metodo.

Študije o zmogljivosti in staranju PV modulov in PV sistemov smo ekstrapolirali na globalni nivo po vsem svetu. Najprej smo predlagali novo klasifikacijo Köppen-Geiger-Photovoltaic (KGPV), ki razširja prvotno shemo Köppen-Geigerja (KG) in dodaja sončno obsevanje kot novo komponento. Dvanajst podnebnih območij vključuje shemo KGPV, kjer prva črka opredeljuje

„glavno podnebje“ in druga govori o „stopnji obsevanja“.

Mehanizme degradacije in skupno stopnje degradacije smo tudi ovrednotili na globalnem nivoju oz. jo kartirali po vsem svetu. Z uporabo podatkov ERA5 smo ocenili stopnjo degradacije v vseh regijah in jih pobezali s podnebnimi območji KGPV.

V nadaljevanju smo predstavili nadgradnjo dosedanjih globalnih zemljevidov oz. kart kot PVWeatherMaps, vključno z oznakami vremenske nevarnosti, ki temeljijo na umazaniji, senčenju zaradi snega in ekstremnih obremenitvah zaradi neviht. Za določitev teh vplivov po vsem svetu so ključni nabori podatkov o podnebnih satelitih in ponovni analizi združeni z empiričnimi modeli. Označevanje tveganja je določeno glede na pričakovane vrednosti za posamezno komponento. Severna Indija, severovzhodna Kitajska, osrednji del ZDA, južno od Španije v severnem delu Afrike in Avstralija imajo veliko tveganje za izgube na račun senčenja zaradi zaprašenosti PV modulov.

Ob koncu disertacije osvetlimo učinkovitost delovanja in staranje fotonapetostnih tehnologij še v luči razvoja globalnih podnebnih sprememb v preteklih in prihodnjih desetletjih. Zadnja leta kažejo zvišanje temperature okolice, kar povzroča zviševanje temperature PV modula, zmanjšano razmerje zmogljivosti in višjo stopnje degradacije PV naprav. Za reševanje tega vprašanja uvedemo in predstavimo novo metodologijo, imenovano "Ocena donosa podnebnih sprememb“

(CCYA). Proti letu 2100 SSP585 prikazuje najslabši scenarij podnebnih sprememb in mi ga vključimo v analizo ter prikažemo njegov vpliv na delovanje PV v različnih podnebnih pasovih.

Medtem ko temperatura okolice še naprej narašča, je globalno obsevanje bolj lokalno odvisno in dolgoročni energijski donos PV energije bo treba analizirati za vsak primer PV sistema posebej.

Table of Contents

1 Introduction ... 1

1.1 Global PV context ... 1

1.2 Advanced monitoring, modelling and assessment of weather and PV

technologies ... 3 1.3 Proposed original scientific contributions ... 5 1.4 Outline of the dissertation ... 8

2 Weather Data Sources, Processing, Modelling and

Validation ... 10

2.1 Weather data sources ... 11

2.1.1 Ground measurements ... 11 2.1.2 Satellite-based models ... 12 2.1.3 Climate Reanalysis-based models ... 13 2.1.4 Climate projections ... 14

2.2 Data aggregation and filtering ... 14

2.2.1 Single time-series ... 14 2.2.2 Global gridded datasets ... 20 2.2.3 Combination of single location and global gridded data ... 22

2.3 Climate Modelling ... 24

2.3.1 Ambient Temperature ... 24 2.3.2 Relative Humidity ... 24 2.3.3 Wind Speed ... 25 2.3.4 Sun Irradiances ... 25

2.4 Typical Meteorological Year (TMY) ... 31 2.5 Typical Daily Profiles (TDP) ... 31 2.6 Weather datasets used in this thesis... 35 2.7 Validation of Data Sources and Models ... 39

2.7.1 Relative Humidity models and ground measurements ... 39 2.7.2 UV irradiance models and ground measurements ... 41 2.7.3 ERA-Interim, BSRN and Modelling ... 42 2.7.4 ERA 5, Explorador Solar and Modelling ... 43

2.8 Summary... 46

3 Theory for PV Modelling ... 47

3.1 Effective Solar Irradiance Modelling ... 48

3.1.1 Modelling of Soiling Losses ... 49 3.1.2 Modelling of Snow losses ... 49 3.1.3 Modelling of Spectral Losses ... 51

3.1.4 Modelling of Angular Losses ... 51

3.2 Thermal Modelling ... 51

3.2.1 Models to estimate the PV module temperature ... 52 3.2.2 Accuracy of PV module temperature models ... 53 3.2.3 Comparison of PV module temperature estimations worldwide ... 55 3.2.4 Realistic maximum PV module temperature ... 56

3.3 PV Performance Modelling and Estimations ... 57

3.3.1 Modelling of PV systems’ energy yield ... 58 3.3.2 Empirical Modelling on operational data ... 60 3.3.3 Machine Learning Approaches on operational data ... 61

3.4 PV Degradation Modelling and Estimations ... 63

3.4.1 Degradation Modes and Failures ... 63 3.4.2 Reliability Modelling ... 64 3.4.3 Physical-based PV Module Degradation Models ... 65 3.4.4 Statistical methods for performance loss estimations ... 66 3.4.5 Uncertainties in Outdoor Performance Loss Estimations ... 68

3.5 Data Sources at PV Module and System Level ... 71 3.6 Summary ... 73

4 Performance and aging of PV modules ... 74

4.1 Outdoor PV module degradation assessment ... 74

4.1.1 I-V tracer and monitoring system ... 76 4.1.2 Filtering and Fitting of I-V curves ... 77 4.1.3 The Empirical Weather Normalization (EWN) approach ... 77 4.1.4 Description of the case study ... 80 4.1.5 Effect of Irradiance and Temperature on I-V parameters ... 80 4.1.6 Application of the Polynomial Filtering and EWN algorithms ... 81 4.1.7 Summary and limitations of the EWN algorithm ... 84

4.2 Uncertainties in modelled PV degradation rates ... 86

4.2.1 Workflow for uncertainties assessment of physical-based degradation rate

estimations ... 87 4.2.2 Uncertainties in modelled degradation rates due to reliability models ... 87 4.2.3 Uncertainties in modelled degradation rates due to physical-based PV

degradation models... 89 4.2.4 Uncertainties in modelled degradation rates due to PV module temperature models 91

4.2.5 Uncertainties in modelled degradation rates due to UV irradiance models ... 92 4.2.6 Uncertainties in modelled degradation rates due to relative humidity models .... 93 4.2.7 Discussions ... 95

4.3 Summary ... 96

5 Performance and aging of PV systems ... 98

5.1 The PV industry in Chile ... 98

5.1.1 Regional solar resource and worldwide comparison ... 99 5.1.2 The current state of the Chilean PV industry ... 100 5.1.3 Technical aspects of PV systems in Chile ... 101 5.1.4 Data availability for this study ... 103

5.2 PV performance assessment on utility-scale systems ... 103

5.2.1 Application of the TDP approach on operational data ... 104 5.2.2 Identification of mounting structure type ... 105 5.2.3 Year-to-Year TDP of PV systems’ output power ... 108 5.2.4 Sensitivity analysis of Y2Y-TDPs ... 109 5.2.5 Key performance indicators based on Typical Daily Profiles ... 110 5.2.6 Modelling for fixed and tracking PV systems ... 111 5.2.7 KPIs on neighbouring PV systems and weather stations ... 111 5.2.8 Analysis of PV systems in Chile using the TDP approach ... 113

5.3 Benchmark of empirical and ML approaches for Yield Modelling ... 116 5.4 Summary... 118

6 Mapping of PV climate zones, performance, and aging ...119

6.1 The interest in PV related climate zones ... 120 6.2 Köppen-Geiger-Photovoltaic Scheme ... 121

6.2.1 Temperature-Precipitation zones: first classification level ... 122 6.2.2 Irradiation zones: second classification level ... 124 6.2.3 Reduction of KGPV classification labels ... 125 6.2.4 Worldwide Köppen-Geiger-Photovoltaic map ... 127

6.3 Mapping of PV Performance indicators ... 128 6.4 Mapping of PV Degradation indicators ... 129 6.5 Weather Risks ... 132

6.5.1 Thunderstorms... 134 6.5.2 Snow Shading Risk ... 135 6.5.3 Soiling Deposition ... 136

6.6 Correlation among PV indicators and climate zones ... 137

6.6.1 PV performance in different climate zones ... 137 6.6.2 PV degradation in different climate zones ... 138 6.6.3 Soiling rates in different countries ... 140

6.7 Summary... 142

7 Climate evolution and its impact on PV ...143

7.1 Predictions based on climate change scenarios ... 143

7.1.1 SSP-RF climate change scenarios in the CMIP6 ... 145

7.2 Evolution of ambient temperature ... 146

7.3 KGPV zones towards 2100 ... 147

7.4 Evolution of the PV Performance and Aging ... 148

7.5 Yield assessment under different climate change scenarios ... 150

7.5.1 Solar resource assessment using climate change scenarios ... 150 7.5.2 Climate Change Yield Assessment (CCYA) for PV systems ... 156 7.5.3 CCYA applied to a PV system in Bolzano, Italy ... 157

7.6 Summary ... 164

8 Conclusions and Outlook ... 166

8.1 Original Scientific Contributions ... 166 8.2 List of publications ... 167 8.3 General Conclusions ... 169 8.4 Outlook for future research ... 172

Bibliography ... 174

List of Figures

Figure 1.1: Total solar PV installed capacity shares 2012-2019 (Source: [10]). ... 1 Figure 1.2: Projections of the installed capacity for solar PV and wind systems in the frame of the Net Zero pathway of the IEA (source: [19]). ... 2 Figure 1.3: Climate and electrical modelling scheme for PV systems (source: [1]). ... 5 Figure 1.4: Topics in which this doctoral work aims to go beyond the state-of-the-art... 7 Figure 2.1: Categories of weather data sources based on the temporal scale. ... 11 Figure 2.2: Representation of the main physical weather phenomena and primary input data sources considered in the ERA5 climate reanalysis model to obtain the global temperature changes over time (Source: [64]). ... 13 Figure 2.3: Levels of data accuracy for the measured and modelled global plane-of-array irradiance (source : [3]). ... 16 Figure 2.4: Distribution of five irradiance clusters together with the 5th, 50th and 95th percentile values. The clusters are plotted using a kernel-density estimate technique with Gaussian kernel.

The irradiance class width was set to 50 W/m². ... 17 Figure 2.5: Example of a global grid with a spatial resolution of 0.5°×0.5°. ... 21 Figure 2.6: Slovenian grid with different spatial resolutions: a) 0.1°x0.1°, b) 0.25°x0.25°, c) 0.5°x0.5° and d) 1°x1°. ... 21 Figure 2.7. Overview of applications (text in grey), data sources (indicated in red ovals), and the cross-relationships between spatial and temporal scale, and data accuracy in solar data ... 22 Figure 2.8: Representation of interpolation techniques to extract single locations from global gridded data. ... 23 Figure.2.9: a) Illustration of the irradiance flow from the Sun to the Earth surface. b) Extra- terrestrial irradiance and Surface Clear-Sky Irradiance as a function of wavelengths. c) Range of the Spectral Irradiance indicating the upper limits for ultraviolet irradiance A and B. ... 26 Figure 2.10: Sun and PV module motion for one-axis tracking mounting. ... 29 Figure 2.11: Sun and PV module motion for two-axis tracking mounting. ... 29 Figure 2.12: a) Spectral irradiance from 280 nm to 440 nm under overcast sky conditions. b) Spectral irradiance from 280 nm to 440 nm under clear sky conditions. ... 30 Figure 2.13: Practical example of the AVG-TDP approach applied to nine years of measured PV energy yield. ... 33 Figure 2.14: Comparison of TMY and TDP algorithms using data extracted from the PVGIS tool [35]. a) Hourly irradiance and monthly irradiation between 2007 and 2016 in Ljubljana, Slovenia.

b) Hourly irradiance and selected data for the construction of the TMY. c) Selected hourly

irradiance for TMY and monthly irradiation from the TMY and Average-TDP with operators P50 and Mean. d) Average-TDPs with operators P50 and Mean. ... 34 Figure 2.15: Locations of weather stations, including irradiance measurements. ... 37 Figure 2.16: Geographical distribution of the weather stations included in the WRMC-BSRN. 38 Figure 2.17: Illustration of the ETOPO1 Global relief model for the land surface. ... 38 Figure 2.18: Global population density projected to 2020. Data retrieved from SEDAC Population Density v4.11. ... 39 Figure 2.19: Monthly boxplots of measured and modelled relative humidity in the four locations:

a) Negev, Israel; b) Gran Canaria, Spain; c) Zugspitze, Germany; d) Ljubljana, Slovenia. The plots correspond to the year 2014 for all the locations. Each box's upper and lower limit corresponds to the 25% and 75% percentile of the measured and modelled values. ... 40 Figure 2.20: Monthly box plots of measured and modelled UV irradiation in Negev, Israel a) and Gran Canaria, Spain b) The plots correspond to 2014 for both locations. ... 42 Figure 2.21: Comparison of hourly synthetic data from ERA-Interim and hourly measured data from 22 stations of the BSRN. a) Ambient Temperature, b) Global Horizontal Irradiance. Legends indicate the 𝑅2, RMSE and MBE for each station. ... 43 Figure 2.22: a) CDF plots for the weather station SPA. b) CDF plots for the weather station PANG. ... 45 Figure 2.23: a) MBE and its trend versus altitude. b) RMSE and its trend versus altitude. ... 45 Figure 3.1: Generalized information flow to estimate the PV performance and degradation from climate datasets. Climate datasets such as satellite-based and reanalysis models are used to extract the local climate. The PV module conditions are estimated and transformed to stress factors that will trigger degradation mechanisms through modelling. ... 48 Figure 3.2: Annual averages of daily soiling rate in several cities of Chile correlated to the annual averages of AOD. (Adapted from [130]). ... 49 Figure 3.3: a) Photography of the rooftop at the Faculty of Electrical Engineering, University of Ljubljana, Slovenia on the 8th February 2020. b) Schematic for portrait orientation and the measured I-V curves partially covered and in case of a clean module. c) Schematic for landscape orientation and the measured IV curves partially covered and in case of a clean module. ... 50 Figure 3.4: Annual Energy Losses due to snow correlated to the annual sum of Snow Depth for tilt-equals-latitude fixed PV modules in locations over the United States. (Adapted from [134]).

... 51 Figure 3.5: Monthly box plots of measured and modelled PV module temperature in the four locations: a) Negev, Israel; b) Gran Canaria, Spain; c) Zugspitze, Germany; d) Ljubljana, Slovenia. The plots correspond to the year 2014 for Negev, Gran Canaria and Zugspitze and 2018 for Ljubljana. ... 54

Figure 3.6: a) Maximal module temperature using the Ross model. b) Maximal module temperature using the Faiman model. c) The absolute difference between the estimated 𝑇𝑚𝑜𝑑 using Faiman and Ross models. Annual average data from 2016 to 2018 from ERA5. ... 55 Figure 3.7: PV Monitoring Data and Energy Flow (source: [160]). ... 57 Figure 3.8: Flowchart for defining initial variables and parameters (modified from: [1]). ... 58 Figure 3.9. Flowchart for PV System Performance modelling. ... 60 Figure 3.10: Flowchart for the data processing, including filtering algorithm stage, training, testing, and validation of empirical models and machine learning approaches. ... 63 Figure 3.11: Strategy to calculate the Degradation Rates of PV modules taking into account the possible uncertainties. ... 71 Figure 3.12: Visualization of the 17-kW PV system installed on the Faculty of Electrical Engineering's rooftop, University of Ljubljana (Source: [3]). ... 72 Figure 4.1: Example of the I-V curve fitting for different irradiance levels. ... 77 Figure 4.2: Flowchart of steps to extract the performance trends of I-V curve parameters. ... 77 Figure 4.3: Single Diode Model of a solar cell (Source:[216]). ... 78 Figure 4.4: Representation of extraction of I-V curve parameters from a fitted I-V curve. ... 78 Figure 4.5: Components of the Empirical Weather Normalization (EWN) approach. ... 80 Figure 4.6: Visual representation of I-V parameters (𝑃𝑀𝑃𝑃, 𝐼𝑆𝐶, 𝑉𝑂𝐶, 𝐹𝐹, 𝑅𝑆, and 𝑅𝑆𝐻) over irradiance and time for PVM1. ... 81 Figure 4.7: Visual representation of the relationships among I-V parameters (𝑃𝑀𝑃𝑃, 𝐼𝑆𝐶, 𝑉𝑂𝐶, 𝐹𝐹, 𝑅𝑆, and 𝑅𝑆𝐻), irradiance, and PV Module Temperature of the PV Modules PVM1. ... 81 Figure 4.8: a) Application of the automatic filtering algorithm published in [3]; b) Raw and Filtered data; c) the first loop of the EWN algorithm based on irradiance ; d) the second loop of the EWN algorithm based on temperature. ... 82 Figure 4.9. Workflow for uncertainties assessment of physical-based lifetime estimations ... 87 Figure 4.10: a) Calibration and extrapolation of degradation rates in the operation of a monocrystalline PV module in Ljubljana, Slovenia. The dashed line at 0.8 indicates the -20%

performance loss of the initial power output. b) Change of failure time with degradation rates using different PV modules reliability models. ... 89 Figure 4.11. Normalized PV module measured power (blue) and model (red). ... 90 Figure 4.12: Variation of degradation rates and failure time using degradation rate models of Bala and Kaaya. The values indicate the estimated degradation rates and failure time. ... 91 Figure 4.13: Variations in degradation rates and failure times using measured and modelled PV module temperatures. The percentages are the relative differences compared to degradation rates evaluated using measured PV module temperatures. ... 92

Figure 4.14: Variations in degradation rates and failure time using measured and modelled 𝑈𝑉 doses. The percentages are the relative differences compared to degradation rates evaluated using measured 𝑈𝑉. ... 93 Figure 4.15: Variations in degradation rates and failure time using measured and modelled relative humidity. The percentages are the relative differences compared to degradation rates evaluated using measured relative humidity. ... 94 Figure 4.16: Change of degradation rate with the different climatic variables: average module temperature (blue), maximum temperature (red), 𝑈𝑉 dose (green) and relative humidity (black).

Degradation rates simulated by varying one variable and keeping other variables constant. For the x-axis, all variables are simulated from 0-100. ... 95 Figure 5.1: Worldwide comparison of the solar potential per Chilean region in terms of annual global horizontal irradiation (𝐻𝑎𝑛𝑛) in [kWh/m²/a]. Name regions defined in ISO 3166-2:CL.

... 99 Figure 5.2: Temporal evolution and geographical distribution of PV systems in Chile. a) Temporal evolution of a number of projects and installed PV capacity. b) Geographical distribution of PV systems in the Northern and Central areas. c) Geographical distribution of PV systems in the Central area. ... 101 Figure 5.3: TDP calculations in CHILE_139, 103.02MW - Tracking located in Colina, Chile. a) Angle viewpoint where high mountains are observed close to the PV system (source: Google Earth). b) Top-viewpoint of the PV system (source: Google Earth). c) Distributions of power output during summer and winter at solar noon. d) Average Typical Daily Profile using percentiles 50th and 90th. (e) Year-to-Year Typical Daily Profile using percentiles 50th and 90th. ... 105 Figure 5.4: Typical Daily Profiles using the 90th percentile for a) Fixed structures, b) Tracking structures, and c) Bifacial PV technologies. ... 106 Figure 5.5: Plotting of kurtosis vs skewness extracted from P90 AVG-TDPs of each PV system:

a) training and b) prediction, with the extracted exponential lines for Fixed and Tracking systems from training. ... 107 Figure 5.6: Comparison of P90 AVG-TDPs and P90 Y2Y-TDPs on PV systems. a) CHILE_139:

identification of initial operational phases. b) CHILE_85: curtailment during the years 2015 and 2016. c) CHILE_144: failure in 2015, followed by large performance losses during 2016 and 2017. ... 109 Figure 5.7: Results based on P90 Y2Y-TDP calculations for 𝐺𝐻𝐼 in the weather station IDEO. a) Number of hours processed, b) Average daily irradiance in W/m², and c) Daily global horizontal irradiation in kWh/m². ... 110 Figure 5.8: Neighbouring PV systems and weather stations. a) Weather station PALM and PV systems CHILE_16 and CHILE_17. b) Weather station IDEO and PV systems CHILE_11 and CHILE_43. c) Weather station CCA and the PV system CHILE_160 (Source: Google Earth Pro).

... 112

Figure 5.9: a) Number of projects and installed PV capacity. b) Annual Unit Capacity Factor of PV systems under at least one year of operation. c) Annual energy yield of PV systems under at least one year of operation. d) Correlation of annual energy yield and net performance serious, indicating the mounting configuration and size of PV systems. e) Geographical distribution of the net performance ratio for systems under at least one year of operation. ... 115 Figure 5.10: Linear regression of monthly Performance Ratio calculated using the Holt-Winters (HW) seasonal exponential smoothing for the 17-kW PV system in May 2014–Dec 2019. The average PR for the training set and test set is also presented. ... 116 Figure 6.1: a) The threshold criteria define main climate zones based on the Köppen-Geiger- Photovoltaic scheme [1]. b) KGPV main climate zones in Europe. c) KGPV main climate zones worldwide... 123 Figure 6.2: Evolution of the ambient temperature per climate zone calculating a 10-years rolling mean. The trend lines represent the linear regression with a 95% confidence interval extrapolated to 2030 2020 for the three data periods: 1950-2016 (black dashed line), 1950-1989 (solid black line) and 1990-2016 (coloured solid line). Grey areas represent the period selected to monthly average the climate variables (from 1990 to 2016). ... 124 Figure 6.3: a) Thresholds criteria to define irradiation zones based on the Köppen-Geiger- Photovoltaic scheme. b) KGPV irradiation zones in Europe. c) KGPV irradiation zones worldwide... 125 Figure 6.4: Köppen-Geiger-Photovoltaic climate classification in Western Europe with a spatial resolution a) 0.1°x0.1°, b) 0.25°x0.25°, and c) 0.5°x0.5°. ... 127 Figure 6.5: Köppen-Geiger-Photovoltaic climate classification with the 12 most relevant climate zones (Antarctica excluded). The first letter indicates the Temperature-Precipitation (TP)-zones:

A-Tropical, B-Desert, C-Steppe, D-Temperate, E-Cold and F- Polar. The second letter indicates the Irradiation (I)-zones: K-Very High, H-High, M-Medium and L-Low irradiation. ... 127 Figure 6.6: Worldwide mapping of Final Energy Yield for a PV system with typical c-Si PV modules. Parameters used: 𝑘𝑅𝑜𝑠𝑠 = 0.03 °C m2/W; 𝛾 = −0.45%/°C; 𝜂𝐵𝑜𝑆 = 85%... 129 Figure 6.7: Global mapping of degradation mechanisms. a) Hydrolysis-degradation, b) photo- degradation, c) thermomechanical degradation, and d) Total degradation rates for a specific monocrystalline silicon PV module using the Kaaya model. Climate data from ERA5 for the average between 2016 and 2018. ... 130 Figure 6.8: European categorization of total degradation rate based on temperature, humidity and UV irradiance for a specific monocrystalline silicon PV module. ... 131 Figure 6.9: Categorization of total degradation rates for a specific monocrystalline silicon PV module using the Kaaya model. Climate data from ERA5 average between 2016 and 2018. .. 131 Figure 6.10: a) Visual concept of the GIS data considering time and space. b) Information flow from global weather data and physical models towards key performance indicators and weather risk labels. ... 133

Figure 6.11: Geographical data generated for Thunderstorm risks in Europe. ... 134 Figure 6.12: Geographical data generated for Snow Shading risks in Europe. ... 135 Figure 6.13: Geographical data generated for Soiling risks in Europe. ... 136 Figure 6.14: Cities representing each KGPV zone in the Northern hemisphere. Additionally, Atacama ... 137 Figure 6.15: Spatial distribution of the degradation mechanisms in view of the Köppen-Geiger- Photovoltaic (KGPV) climate zones. ... 139 Figure 6.16: Spatial distribution of the total degradation rates in view of the KGPV climate zones.

The average total degradation rate per climate zone is indicated below each label in %/a. ... 140 Figure 6.17: Soiling risk map for different countries. ... 141 Figure 6.18: Comparison of daily soiling rates from literature and the modelled soiling rates over an entire country. ... 141 Figure 7.1: a) Illustration of Shared Socioeconomic Pathways (SSP) scenarios based on challenges for mitigation and adaptation. b) Radiative forcing levels in W/m² combined with the SSP scenarios to build the SSP-RF matrix. ... 146 Figure 7.2: 10-years rolling mean evolution of the Land-Surface ratio for a) TP-zones and b) I- zones. ... 148 Figure 7.3: Evolution of relevant PV performance indicators for c-Si PV modules under the SSP585 climate change scenario. Annual differences of 10 years rolling mean: a) Ambient Temperature, b) Global Horizontal Irradiation, c) Maximal Module Operating Temperature and d) Performance Ratio... 149 Figure 7.4: Temporal evolution of the global ambient temperature and the total degradation rates.

... 149 Figure 7.5: A proposed methodology for solar resource assessment using climate change scenarios. ... 151 Figure 7.6: Application of the solar resource assessment on the station 03987 retrieved from DWD [285], corrected to the climate change scenarios SSP119, SSP126, SSP245, SSP370, SSP434, SSP460, and SSP585.a) Correlation and correction of estimated irradiance based on measured irradiance. b) time-series of measured and corrected irradiance data. c) Correlation and correction of estimated irradiation based on measured irradiation. d) time-series of measured and corrected irradiation data. ... 152 Figure 7.7: Solar resource assessment under climate change scenarios for available DWD stations.

The gain/loss values are calculated by the differences between the 10-year rolling mean values and the measured average value. ... 153

Figure 7.8: Solar resource assessment under climate change scenarios for European WRMC- BSRN stations. The gain/loss values are calculated by the differences between the 10-year rolling mean values and the measured average value. ... 154 Figure 7.9: Mapping of WRMC-BSRN stations in Europe with the gain/loss of the solar irradiance by 2030, 2050 and 2100. The gain/loss values showed in the maps have been calculated as the average of all SSPs climate change scenarios. ... 154 Figure 7.10: A proposed methodology for the Climate Change Yield Assessment (CCYA) for PV systems. ... 156 Figure 7.11: Differences between monthly average of measured irradiances and modelled under different climate change scenarios. ... 157 Figure 7.12: Differences between monthly average of measured temperatures and modelled temperatures under different climate change scenarios. ... 158 Figure 7.13: Relation of modelled monthly mean global horizontal irradiance and measured daily DC yield. The star symbols represent the centre of each cloud per month of the year. ... 158 Figure 7.14: Evolution of the monthly DC energy yield of a PV system operating in Bolzano, Italy. ... 159 Figure 7.15. Specific DC energy yield in kWh/kWp considering the irradiance variability under the different climate change scenarios. The 10-years rolling mean is added to visualize the trends easily. ... 159 Figure 7.16: DC energy yield gain/loss compared to the measured P10, P50, and P90 under irradiance variability. a) Irradiance variability versus measured P10; b) Irradiance variability versus measured P50; c) Irradiance variability versus measured P90... 160 Figure 7.17: Temperature loss factor calculated for the different climate change scenarios indicating the temperature variability over the years of operation. The 10-years rolling mean is added to visualize the trends easily. ... 161 Figure 7.18: DC energy yield gain/loss compared to the measured P10, P50, and P90 under temperature variability with a power-temperature coefficient equal to -0.45%/°C and Ross coefficient equal to 0.03°C m²/W. a) Temperature variability versus measured P10; b) temperature variability versus measured P50; c) temperature variability versus measured P90.

... 161 Figure 7.19. Power loss factor calculated for different linear degradation patters with rates 0%/year, 0.25%/year, 0.50%/year and 1.0%/year. Those patterns indicate the power variability evaluated. ... 162 Figure 7.20: DC energy yield gain/loss compared to the measured P50 under different scenarios of power variability. a) Power variability under 0.0%/year versus measured P50; b) Power variability under 0.25%/year versus measured P50; c) Power variability under 0.5%/year versus measured P50; d) Power variability under 1.0%/year versus measured P50. ... 163

Figure 7.21: Comparison of climate change scenarios considering irradiance variability, temperature variability and power variability. ... 164 Figure 8.1: Main stages and results developed in this doctoral thesis. ... 172

List of Tables

Table 2.1: Definition of common PV performance monitoring issues. ... 15 Table 2.2: Description of three different levels of data accuracy (case: Plane-of-Array irradiance).

... 15 Table 2.3: Description of filtering procedures for different time resolution data. ... 16 Table 2.4: Algorithm for data filtering (Clustering). ... 18 Table 2.5: Filtering algorithm step-by-step on the datasets for each level of data accuracy. ... 19 Table 2.6: Annual average measured and modelled relative humidity with measured and modelled data. Also, the relative differences between measured and modelled values are given. ... 40 Table 2.7: UV irradiance calibration and modelling validation in Negev and Gran Canaria. .... 41 Table 2.8: Measured and modelled annual UV dose [kWh/m²/a]. 𝑅𝑒𝑙𝑑𝑖𝑓𝑓 is the relative difference between the modelled and measured UV dose. ... 41 Table 2.9: Comparison of modelled and measured solar irradiance. Stations of the Explorador Solar and modelled data using ERA5 and PVLib. ... 44 Table 3.1: Definition of loss factors for incident global irradiance. ... 48 Table 3.2: Typical values of Power-Temperature Coefficients (𝛾) per technology. ... 52 Table 3.3: Extracted coefficients 𝑘𝑅𝑜𝑠𝑠 of Ross model and 𝑢0 , 𝑢1 of Faiman model. Calibration was done using measured module data of 2013 for Negev, Gran Canaria and Zugspitze and 2015 data for Ljubljana. 𝑛𝑅𝑀𝑆𝐸𝑐𝑎𝑙 and 𝑅𝑀𝑆𝐸𝑐𝑎𝑙 correspond to the normalized-root-mean-square- error (nRMSE) for calibration. ... 53 Table 3.4. Measured and modelled temperatures in the four locations using Faiman and Ross model. 𝑇𝑎𝑣𝑔, 𝑇𝑚𝑎𝑥, and 𝑇𝑚𝑖𝑛 are the average, maximum and minimum PV module temperature in [°C]. The maximum and minimum temperatures are evaluated as the 95^th and 5^th percentiles of the annual temperature distribution. 𝑅𝑒𝑙𝑑𝑖𝑓𝑓 is the relative difference between the modelled and measured temperatures. ... 54 Table 3.5: Increase of Temperature due to the Irradiance at different mounting conditions. Typical values of 𝑘𝑅𝑜𝑠𝑠 for different mounting conditions are taken from [154]. ... 56 Table 3.6: PV module temperature at different ambient temperature values and mounting conditions. (𝐺 = 1300 W/m²). Typical values of 𝑘𝑅𝑜𝑠𝑠 for different mounting conditions are taken from [154]. ... 56 Table 3.7: Definition of the main parameters of PV performance modelling. ... 58 Table 3.8: Key performance indicators (KPIs) for PV performance assessment. ... 59

Table 3.9: Simplified relation among the failure modes and the shapes of IV curves based on information taken from [176]. ... 63 Table 3.10: Reliability models for power degradation with derived failure time. β and Г are parameters associated with the materials, µ is the shape parameter, while 𝑘 [𝑦𝑒𝑎𝑟 − 1] is the degradation rate and (𝑡𝑓) is the failure time. ... 65 Table 3.11: Uncertainty factors for degradation rates and performance loss estimations. ... 68 Table 4.1: Characteristics of PV module evaluated under outdoor exposure... 80 Table 4.2: Overall results of the filtering algorithm and Empirical Weather normalization on the three modules evaluated. This table gives the annual rate of the I-V curve parameters. ... 83 Table 4.3: Fitting parameters and uncertainty indicators for each component at the filtering stage with percentile curves 50th, 5th, and 95th and the EWN algorithm applied on each I-V curve parameter based on irradiance (i.e., EWNG) and ambient temperature (i.e., EWNG, T). ... 83 Table 4.4: Model parameters during calibration and estimated degradation rates of a monocrystalline PV module operating in Ljubljana, Slovenia. The nRMSE and nMBE give the calibration uncertainties, while the lifetime is calculated using the derived failure functions, and it corresponds to a -20% of losses of the initial power output. µ corresponds to the shape parameter, while Г is the model parameter. ... 88 Table 4.5: Extracted parameter values using the Bala model [179]. ... 90 Table 4.6: Extracted parameter values using the Kaaya model [178]. ... 90 Table 4.7: Measured climatic variables of the four locations in annual averages used to assess the two degradation rate models. 𝑇𝑎𝑣𝑔, 𝑇𝑚𝑎𝑥 , and 𝑇𝑚𝑖𝑛 are the average, maximum and minimum PV module temperature in [°C]; 𝑅𝐻 is the relative humidity in [%]; 𝑈𝑉 dose is the integral ultraviolet irradiation in [kWh/a/m²]. ... 90 Table 5.1: Average annual irradiation, minimum and maximum for plotting, name and code per region. ... 100 Table 5.2: Evolution of the large PV systems installed capacity in Chile [229] ... 100 Table 5.3: Main failures of eight large scale PV systems in Chile (Source: [54]). ... 102 Table 5.4: Net Performance Ratio of four PV systems under different approaches... 112 Table 5.5: Summary of temporal events through long-term P90 TDPs in Chilean large-scale PV systems. ... 113 Table 5.6: Uncertainty indicators of ML approaches for different data accuracy levels, including the normalized root-mean-square-error (nRMSE) of the testing set and the processing time (P.time). Features considered: GPoA, Tamb, and sun position angles. ... 117 Table 5.7: nRMSE of each machine learning approach using different input features for the high accuracy dataset. ... 117

Table 6.1: Land Surface Ratio per climate zone indicating the land-surface portion for each climate zone over the total sum. ... 126 Table 6.2: Population Density per climate zone representing the average population density for each climate zone over the Globe. ... 126 Table 6.3: Surface-Population Density indicator per climate zone calculated by multiplying Table 6.1 and Table 6.2. ... 126 Table 6.4: Assumptions applied to KGPV labelling procedure merging to most relevant zones next to them indicated by the arrows. Different colours label TP-zones. ... 126 Table 6.5: Definition of thunderstorm risks based on 𝑇𝑇𝐼. ... 134 Table 6.6: Definition of snow shading risk based on snow depth. ... 135 Table 6.7: Definition of soiling risk based on soiling rates as a function of AOD at 550nm. .. 136 Table 6.8: Simulated performance indicators for selected cities in each KGPV climate zone.

(𝑘𝑅𝑜𝑠𝑠 = 0.03 °C m2/W, γ = -0.45%/°C, 𝜂𝐵𝑜𝑆=85%) ... 138 Table 7.1: Details of the measurement station where solar resource assessment under climate change scenarios are applied. ... 151 Table 7.2: Detailed results for the evaluated stations. The gain/loss and expected yield correspond to the climate change scenario SSP585. ... 155