Delta på prøveforelesning - 22. mai kl. 10:15 (rom 402, Institutt for teknologisystemer)
"How wind power forecasts are used in applications in different time frames, from seconds to days ahead".
Kreeringssammendrag
For å bekjempe klimaendringer er det essensielt å forbedre påliteligheten til fornybare energikilder som vindkraft. Arbeidet setter søkelys på bruk av dyp læring for å utvikle nøyaktige prognoser for vindkraft, inkludert modellering av interaksjoner mellom vindturbiner og værprognoser i nær fremtid. Forskingen viser til relevante arkitekturer og metoder for mer nøyaktige prognosemodeller, samt fleksible modeller for irregulære datasett og metoder for sannsynlighetsprognoser.
Hovedfunn
In the global effort to combat climate change, wind energy is of paramount importance in transitioning away from fossil fuels. But wind power comes with a challenge: it's not always blowing when we need it. That's where advanced prediction systems come in. Imagine a wind farm as a complex puzzle of turbines, each affecting the others. In this work, we study how deep learning enables us to account for these turbine interactions, resulting in more precise wind turbine power predictions. By looking at data from multiple locations over time, we show how graph neural networks and complex Transformer architectures can help improve our forecasts even further. We've also come up with new ways to handle tricky data gaps, making our predictions more reliable.
Furthermore, a key takeaway from this research is not only the utilisation of advanced techniques to improve wind energy forecasting but also the importance of incorporating probabilistic prediction models. These models provide valuable insights into the uncertainty associated with wind power, aiding in better planning for successful system integration. Ultimately, these advanced techniques empower us to harness wind energy more effectively, offering hope for a greener future.
Leder av disputas
Prof. Øivind Kure, Universitetet i Oslo
Bedømmelseskomité
1. Opponent: Prof. Georges Kariniotakis, Mines Paris
2. Opponent: Dr. Corinna Möhrlen, WEPROG
Komitéleder: Prof. Josef Noll, Universitetet i Oslo
Veiledere
Hovedveileder: Paal Engelstad, Universitetet i Oslo
Medveileder: Narada Dilp Warakagoda, Universitetet i Oslo
Medveileder: Roy Stenbro, IFE
Kontaktinformasjon til kandidat
LinkedIn: Lars' LinkedIn
E-post: lars.nbe@hotmail.com
Mob.nr: +47 97139979
For mer informasjon
PhD-koordinator: Yvonne Baade
Anmodning om tilgang til avhandlingen (pdf)