Nowadays, all the global energy policies defined to mitigate the world climate changes have promoted the development of renewable energies which have reached a total installed capacity worldwide equal to 2378 GW (2018), including an higher number of countries generating more 15% of electricity from renewable sources. In the last decade, wind has experienced a strong expansion reaching 591 GW (2018)of installed capacity worldwide, with more than 101 countries participating in the related energy market. The higher penetration of variable renewable energy sources, (wind and solar), has led to a growing demand for predictive methods of the power produced, requested especially by the transmission system operators in order to properly guarantee the increasingly complex physical balance between the power introduced and extracted from the electrical network. Regarding the economic aspect, the benefits deriving from an accurate predictive method allow to maximize the extracted energy content from a given energy resource,decreasing the costs of electricity production and making these technologies competitive for a comparison with fossil fuel-based power systems. Given the benefits deriving from a correct prediction of the variability of the wind, the present work proposes a wind power predictive model based on the implementation of artificial neural networks, which have now found application in a wide range of scientific problems. The time series used are composed of hourly measurements covering a period of two years, collected by measuring instruments of a wind farm located in Chieti (Italy),consisting of 24 wind turbines with a nominal power of 0.66 MW.In order to ensure an adequate reliability and robustness of the results obtained from the performance evaluation, it was chosen to use eight different error metrics and to evaluate the performance considering two different predictive situations (annual and daily), using the persistence model as the reference one. The evaluations of predictive performances, regarding both the analyses, revealed very satisfactory results which confirm the superiority of data-driven approaches in the daily wind power prediction. In more detail, such results include a reduction (compared to persistence model) of the MAE, the NRMSE and the SS by 50%, 47.82%and 47.68%, respectively.
Al giorno d’oggi, tutte le politiche energetiche mondiali definite per mitigare i cambiamenti climatici hanno incentivato lo sviluppo delle energie rinnovabili che hanno raggiunto una capacità installata totale pari a 2378 GW (2018), con un numero sempre più grande di paesi che generano più del 15% della propria elettricità da fonti rinnovabili. Nell'ultimo decennio, l’eolico ha beneficiato di una forte espansione raggiungendo 591 GW (2018) di capacità installata in tutto il mondo, con più di 101 paesi partecipanti nel relativo mercato energetico.L’elevata penetrazione delle fonti energetiche rinnovabili variabili (eolico e solare),ha portato ad una crescente domanda di modelli predittivi della potenza prodotta,richiesti soprattutto dagli operatori della rete di trasmissione per garantire il sempre più complesso equilibrio fisico tra potenza immessa ed estratta dalla rete elettrica.Per quanto riguarda l’aspetto economico, i benefici derivanti da un accurato metodo predittivo permettono di massimizzare l’estrazione del contenuto energetico di una data risorsa, diminuendo il costo di produzione dell’elettricità e rendendo queste tecnologie competitive per un confronto diretto con i sistemi basati su combustibili fossili (anche in assenza di supporti finanziari).Dati i benefici derivanti da una corretta predizione della variabilità del vento, il presente lavoro propone un modello predittivo della potenza eolica basato sull'utilizzo delle reti neurali artificiali, che hanno ormai trovato applicazione in un vastissimo range di problemi scientifici.Le serie storiche utilizzate sono composte da misurazioni orarie ricoprenti un periodo di due anni, raccolte dagli strumenti di misura di un campo eolico situato a Chieti(Italia), composto da 24 turbine eoliche con una potenza nominale pari a 0.66 MW. Per garantire un’adeguata affidabilità e robustezza dei risultati ottenuti dalla valutazione delle performance, è stato scelto di utilizzare otto differenti metriche di errore e di valutare le prestazioni dei modelli in due situazioni diverse (annuale e giornaliera),utilizzando il modello della persistenza come riferimento.Le valutazioni delle prestazioni predittive, ottenute da entrambe le analisi, hanno rivelato risultati molto soddisfacenti che confermano la superiorità dei modelli "data-driven" nella predizione della potenza eolica giornaliera. Più in dettaglio, tali risultati comprendono una riduzione (rispetto al modello della persistenza) del MAE, del NRMSE e del SS rispettivamente del 50%, 47.82% e 47.68%.
Wind power 24h ahead forecast. Predictive models analysis and comparison
Giglio, Alessandro
2019/2020
Abstract
Nowadays, all the global energy policies defined to mitigate the world climate changes have promoted the development of renewable energies which have reached a total installed capacity worldwide equal to 2378 GW (2018), including an higher number of countries generating more 15% of electricity from renewable sources. In the last decade, wind has experienced a strong expansion reaching 591 GW (2018)of installed capacity worldwide, with more than 101 countries participating in the related energy market. The higher penetration of variable renewable energy sources, (wind and solar), has led to a growing demand for predictive methods of the power produced, requested especially by the transmission system operators in order to properly guarantee the increasingly complex physical balance between the power introduced and extracted from the electrical network. Regarding the economic aspect, the benefits deriving from an accurate predictive method allow to maximize the extracted energy content from a given energy resource,decreasing the costs of electricity production and making these technologies competitive for a comparison with fossil fuel-based power systems. Given the benefits deriving from a correct prediction of the variability of the wind, the present work proposes a wind power predictive model based on the implementation of artificial neural networks, which have now found application in a wide range of scientific problems. The time series used are composed of hourly measurements covering a period of two years, collected by measuring instruments of a wind farm located in Chieti (Italy),consisting of 24 wind turbines with a nominal power of 0.66 MW.In order to ensure an adequate reliability and robustness of the results obtained from the performance evaluation, it was chosen to use eight different error metrics and to evaluate the performance considering two different predictive situations (annual and daily), using the persistence model as the reference one. The evaluations of predictive performances, regarding both the analyses, revealed very satisfactory results which confirm the superiority of data-driven approaches in the daily wind power prediction. In more detail, such results include a reduction (compared to persistence model) of the MAE, the NRMSE and the SS by 50%, 47.82%and 47.68%, respectively.| File | Dimensione | Formato | |
|---|---|---|---|
|
Giglio_876173_Versione_Finale.pdf
non accessibile
Dimensione
12.27 MB
Formato
Adobe PDF
|
12.27 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/10589/164359