Wind Farm Flow Control has been a prominent research topic in the wind energy community for several years. This field focuses on controlling wind turbines at the farm level to achieve various objectives. In this study, the control strategy aims to maximize the wind farm's power output by optimizing the yaw control of turbines to steer wakes away from downstream turbines. This strategy has been extensively studied, with most approaches relying on open-loop optimization, where the control law is computed beforehand and then applied to the operating wind farm. Theoretically, under ideal conditions, this approach can achieve power output increases of up to 10%. However, in real-world applications, the open-loop method encounters significant challenges. Wind farms operate under a wide range of conditions, such as turbine shutdowns or blade degradation, which alter their response in unpredictable ways. Furthermore, yaw optimization relies on models that never perfectly replicate real-world dynamics. Environmental conditions often used as inputs for optimization, are either unavailable (eg. turbulence intensity) or prone to errors. To address these limitations, a shift toward closed-loop wind farm control has emerged. This method overcomes the drawbacks of open-loop control by adapting to real-time conditions. In this thesis, a closed-loop, model-free control approach based on recursive least squares adaptive filters and policy gradient methods is presented. This algorithm estimates optimal control points in real-time by evaluating the wind farm's response directly, without relying on pre-existing models. While the resulting power gains are comparable to those achieved with open-loop methods, the optimization process is fundamentally different.
Il wind farm flow control è stato un argomento di ricerca di grande interesse nella comunità dell'energia eolica per diversi anni. Questo campo si concentra sul controllo delle singole turbine per ottimizzare dei parametri al livello del parco eolico. In questo studio, la strategia di controllo mira a massimizzare la produzione di energia ottimizzando il controllo di imbardata delle turbine, in modo da deviare le scie dalle turbine sottovento. Questa strategia è stata ampiamente studiata, con la maggior parte degli approcci basati sull'ottimizzazione in anello aperto, in cui la legge di controllo viene calcolata a priori e poi applicata al parco eolico in funzione. Teoricamente, in condizioni ideali, questo approccio può portare a un incremento della produzione di potenza fino al 10%. Tuttavia, nelle applicazioni reali, il metodo in anello aperto presenta significative criticità. I parchi eolici operano in una vasta gamma di condizioni, come arresti delle turbine o degradazione delle pale, che alterano in modo imprevedibile la loro risposta aerodinamica. Inoltre, l'ottimizzazione del controllo di imbardata si basa su modelli che non replicano mai perfettamente le dinamiche del mondo reale. Infine, Le condizioni ambientali, spesso utilizzate come input per l’ottimizzazione, sono o indisponibili (ad esempio l'intensità della turbolenza) o soggette a errori di misurazione. Per affrontare queste limitazioni, è emerso un nuovo approccio, il controllo in anello chiuso. Questo metodo supera le criticità del controllo in anello aperto adattandosi alle condizioni in tempo reale. In questa tesi, viene presentato un approccio di controllo in anello chiuso, senza l'utilizzo di alcun modello e basato su filtri adattivi recursive least square e metodi di discesa del gradiente. Questo algoritmo stima i punti di controllo ottimali in tempo reale valutando direttamente la risposta del parco eolico. Sebbene i guadagni di potenza risultanti siano comparabili a quelli ottenuti con i metodi in anello aperto, il processo di ottimizzazione è fondamentalmente diverso.
Reinforcement learning-based wind farm flow control
Bassalti, Alessandro
2023/2024
Abstract
Wind Farm Flow Control has been a prominent research topic in the wind energy community for several years. This field focuses on controlling wind turbines at the farm level to achieve various objectives. In this study, the control strategy aims to maximize the wind farm's power output by optimizing the yaw control of turbines to steer wakes away from downstream turbines. This strategy has been extensively studied, with most approaches relying on open-loop optimization, where the control law is computed beforehand and then applied to the operating wind farm. Theoretically, under ideal conditions, this approach can achieve power output increases of up to 10%. However, in real-world applications, the open-loop method encounters significant challenges. Wind farms operate under a wide range of conditions, such as turbine shutdowns or blade degradation, which alter their response in unpredictable ways. Furthermore, yaw optimization relies on models that never perfectly replicate real-world dynamics. Environmental conditions often used as inputs for optimization, are either unavailable (eg. turbulence intensity) or prone to errors. To address these limitations, a shift toward closed-loop wind farm control has emerged. This method overcomes the drawbacks of open-loop control by adapting to real-time conditions. In this thesis, a closed-loop, model-free control approach based on recursive least squares adaptive filters and policy gradient methods is presented. This algorithm estimates optimal control points in real-time by evaluating the wind farm's response directly, without relying on pre-existing models. While the resulting power gains are comparable to those achieved with open-loop methods, the optimization process is fundamentally different.File | Dimensione | Formato | |
---|---|---|---|
Bassalti_Alessandro_Thesis_18_11_2024.pdf
non accessibile
Descrizione: Thesis
Dimensione
5.53 MB
Formato
Adobe PDF
|
5.53 MB | Adobe PDF | Visualizza/Apri |
Bassalti_Alessandro_Executive_Summary_18_11_2024.pdf
non accessibile
Descrizione: Executive Summary
Dimensione
1.85 MB
Formato
Adobe PDF
|
1.85 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/231114