This thesis collects all the work done by the PhD candidate at the University of Catania and Enel Green Power, whose collaboration has allowed the possibility of carrying out the doctorate research. Artificial intelligence is a field of computer science that deals with developing algorithms and systems capable of performing tasks that emulates human capabilities trying to achieve human being level of performance. Using complex algorithms and data, artificial intelligence enables machines to perform advanced tasks that, in the past, it would not have been possible to carry out automatically. Such technology is profoundly influencing our lives, entering fields such as medicine, industrial automation, transportation, and even artistic creativity. In this thesis the PhD Candidate will mainly deal with artificial intelligence applied to the world of photovoltaic panels with the aim of improving their maintainability. Initially, the automatic classifications of various types of defects present in photovoltaic panels identifiable with the test based on electroluminescence will be investigated; in this study a new dataset and a preliminary benchmark to make an automatic and accurate classification of defects in solar cells will be proposed. This dataset will include five classes of defects and the pre-trained ResNext50 network will reach 0.07 Hamming Distance. Subsequently, considering the importance of the operating temperature of the photovoltaic modules in the study of the degradation of the photovoltaic panels and that the operative temperature cannot always be recovered due to the absence of sensors, conventional physics models with machine learning models will be compared for the estimation of the temperature of monofacial and bifacial photovoltaic modules installed in two different locations of Italy. In that comparison, machine learning models will turn out to be slightly better than conventional physics models. Furthermore, a benchmark evaluation about compression techniques of stereoscopic images will be performed on large and standardized datasets including 60 stereopairs that differ by resolution and acquisition technique in which an Human Visual System assessment experiment will be implemented which will involve more than a hundred people in order to verify the perceived quality of the decoded images.
Questa tesi raccoglie tutto il lavoro svolto dal dottorando presso l'Università di Catania ed Enel Green Power, la cui collaborazione ha permesso la possibilità di svolgere il dottorato di ricerca. L'intelligenza artificiale è un campo dell'informatica che si occupa dello sviluppo di algoritmi e sistemi in grado di eseguire compiti che emulano le capacità umane cercando di raggiungere il livello di prestazioni dell'essere umano. Utilizzando algoritmi e dati complessi, l’intelligenza artificiale consente alle macchine di svolgere compiti avanzati che, in passato, non sarebbe stato possibile svolgere automaticamente. Tale tecnologia sta influenzando profondamente le nostre vite, entrando in campi come la medicina, l’automazione industriale, i trasporti e persino la creatività artistica. In questa tesi il dottorando si occuperà principalmente di intelligenza artificiale applicata al mondo dei pannelli fotovoltaici con l'obiettivo di migliorarne la manutenibilità. Inizialmente verranno indagate le classificazioni automatiche delle varie tipologie di difetti presenti nei pannelli fotovoltaici identificabili con il test basato sull'elettroluminescenza; in questo studio verranno proposti un nuovo set di dati e un benchmark preliminare per effettuare una classificazione automatica e accurata dei difetti nelle celle solari. Questo set di dati includerà cinque classi di difetti e la rete ResNext50 pre-addestrata raggiungerà 0,07 di Hamming Distance. Successivamente, considerando l'importanza della temperatura operativa dei moduli fotovoltaici nello studio del degrado dei pannelli fotovoltaici e che la temperatura operativa non sempre può essere recuperata a causa dell'assenza di sensori, verranno confrontati modelli fisici convenzionali con modelli di machine learning per la stima della temperatura di moduli fotovoltaici monofacciali e bifacciali installati in due diverse località d'Italia. In questo confronto, i modelli di apprendimento automatico si riveleranno leggermente migliori dei modelli fisici convenzionali. Inoltre, verrà effettuata un benchmark sulle tecniche di compressione delle immagini stereoscopiche su dataset ampi e standardizzati comprendenti 60 immagini stereoscopiche che differiscono per risoluzione e tecnica di acquisizione in cui verrà implementato un esperimento di valutazione del sistema visivo umano che coinvolgerà più di un centinaio di persone con il fine di verificare la qualità percepita delle immagini decodificate.
Intelligenza Artificiale Applicata all'Energia Solare
GRISANTI, MARCO
2024
Abstract
This thesis collects all the work done by the PhD candidate at the University of Catania and Enel Green Power, whose collaboration has allowed the possibility of carrying out the doctorate research. Artificial intelligence is a field of computer science that deals with developing algorithms and systems capable of performing tasks that emulates human capabilities trying to achieve human being level of performance. Using complex algorithms and data, artificial intelligence enables machines to perform advanced tasks that, in the past, it would not have been possible to carry out automatically. Such technology is profoundly influencing our lives, entering fields such as medicine, industrial automation, transportation, and even artistic creativity. In this thesis the PhD Candidate will mainly deal with artificial intelligence applied to the world of photovoltaic panels with the aim of improving their maintainability. Initially, the automatic classifications of various types of defects present in photovoltaic panels identifiable with the test based on electroluminescence will be investigated; in this study a new dataset and a preliminary benchmark to make an automatic and accurate classification of defects in solar cells will be proposed. This dataset will include five classes of defects and the pre-trained ResNext50 network will reach 0.07 Hamming Distance. Subsequently, considering the importance of the operating temperature of the photovoltaic modules in the study of the degradation of the photovoltaic panels and that the operative temperature cannot always be recovered due to the absence of sensors, conventional physics models with machine learning models will be compared for the estimation of the temperature of monofacial and bifacial photovoltaic modules installed in two different locations of Italy. In that comparison, machine learning models will turn out to be slightly better than conventional physics models. Furthermore, a benchmark evaluation about compression techniques of stereoscopic images will be performed on large and standardized datasets including 60 stereopairs that differ by resolution and acquisition technique in which an Human Visual System assessment experiment will be implemented which will involve more than a hundred people in order to verify the perceived quality of the decoded images.File | Dimensione | Formato | |
---|---|---|---|
Tesi di Dottorato di Ricerca.pdf
accesso aperto
Dimensione
6.11 MB
Formato
Adobe PDF
|
6.11 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/77452
URN:NBN:IT:UNICT-77452