The strong development of renewable sources that has been witnessed over the last decade all over the world, has created the need to integrate these sources more and more efficiently within the electricity system. Referring in particular to the solar source and the wind source, since they are aleatory sources whose production it is not possible to determine exactly their production, the development of those forecasting methods that allow a good forecast of the power produced is of fundamental importance; precisely, a more active participation of these sources within the electricity system makes it possible to reduce the critical issues related to renewable sources in the management of the electricity system, facilitating the balancing of the network by the appropriate grid operators, reducing the costs of imbalance to which the owners of the plants are subject to and therefore reducing the cost for the whole community. In particular, in this master thesis, a forecast model has been implemented based on feedforward artificial neural networks for the production of the hourly power of a photovoltaic system, whose application is not precluded to other aleatory sources. The developed forecasting scheme presents the originality with respect to the methods previously described in the literature, to be able to free itself completely from the weather forecast and to use exclusively the time series of the real power measured in the previous days, with all the advantages deriving from it; a sensitivity analysis was carried out to optimize the model with respect to the main parameters on which it depends, based on the evaluation of the appropriate performance indicators, including the innovative EMAE. Finally, through a post-processing phase, it was considered appropriate to generate a suitable correction coefficient of the power expected to optimize the final results obtained and to make the proposed forecasting model more reliable.
Il forte sviluppo delle fonti rinnovabili al quale si è assistito nel corso dell’ultimo decennio in tutto il mondo, ha creato la necessità di integrare queste fonti in modo sempre più efficiente all’interno del sistema elettrico. Facendo riferimento in particolare alla fonte solare e alla fonte eolica, trattandosi di fonti aleatorie delle quali non è possibile determinare con esattezza la produzione, risulta di fondamentale importanza lo sviluppo di quei metodi previsionali che consentano una buona previsione della potenza prodotta; per l’appunto una partecipazione più attiva di queste fonti all’interno del sistema elettrico consente di ridurre le relative criticità nella gestione del sistema elettrico, facilitando il bilanciamento della rete da parte degli opportuni enti regolatori, riducendo i costi di sbilanciamento a cui sono soggetti i proprietari degli impianti e quindi riducendo la spesa per l’intera collettività. In particolare nel presente lavoro di tesi è stato implementato un modello previsionale basato sulle reti neurali artificiali feedforward per la produzione della potenza oraria di un impianto fotovoltaico, la cui applicazione non è preclusa alle altre fonti aleatorie. Lo schema previsionale sviluppato presenta l’originalità rispetto ai metodi precedentemente esposti in letteratura, di potersi svincolare dalle grandezze meteorologiche previste e di utilizzare esclusivamente la serie temporale della potenza misurata nei giorni precedenti, con tutti i vantaggi che ne derivano; è stata condotta un’analisi di sensitività per ottimizzare il modello rispetto ai parametri principali dai quali dipende, basandosi sulla valutazione degli opportuni indicatori di performance, tra cui l’innovativo EMAE. Infine attraverso una fase di post-processing, si è ritenuto opportuno generare un idoneo coefficiente correttivo della potenza prevista per ottimizzare i risultati finali ottenuti e rendere più affidabile il modello previsionale proposto.
Analisi di serie temporali tramite reti neurali artificiali per la previsione della produzione di un impianto fotovoltaico
TRIZZINO, FABIO
2016/2017
Abstract
The strong development of renewable sources that has been witnessed over the last decade all over the world, has created the need to integrate these sources more and more efficiently within the electricity system. Referring in particular to the solar source and the wind source, since they are aleatory sources whose production it is not possible to determine exactly their production, the development of those forecasting methods that allow a good forecast of the power produced is of fundamental importance; precisely, a more active participation of these sources within the electricity system makes it possible to reduce the critical issues related to renewable sources in the management of the electricity system, facilitating the balancing of the network by the appropriate grid operators, reducing the costs of imbalance to which the owners of the plants are subject to and therefore reducing the cost for the whole community. In particular, in this master thesis, a forecast model has been implemented based on feedforward artificial neural networks for the production of the hourly power of a photovoltaic system, whose application is not precluded to other aleatory sources. The developed forecasting scheme presents the originality with respect to the methods previously described in the literature, to be able to free itself completely from the weather forecast and to use exclusively the time series of the real power measured in the previous days, with all the advantages deriving from it; a sensitivity analysis was carried out to optimize the model with respect to the main parameters on which it depends, based on the evaluation of the appropriate performance indicators, including the innovative EMAE. Finally, through a post-processing phase, it was considered appropriate to generate a suitable correction coefficient of the power expected to optimize the final results obtained and to make the proposed forecasting model more reliable.File | Dimensione | Formato | |
---|---|---|---|
2018_4_Trizzino.pdf
non accessibile
Descrizione: Testo della tesi
Dimensione
23.13 MB
Formato
Adobe PDF
|
23.13 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/10589/139712