Switzerland is currently producing about the 60% of electricity from hydropower. This percentage is doomed to increase in the following years, since the recent decision of nuclear phase-out is opening new challenges for balancing the national energy demand. Markets where renewable sources produce a relevant share of the total energy production turn out to be much more flexible in adjusting the production than those with conventional plants. Therefore, an accurate prediction of price trajectory in the short term can help suppliers to maximize their utility by properly scheduling the production, as well as helping consumers to have successful trade of this commodity. In modern competitive electricity markets, the prediction of energy prices is thus considered a key information for both suppliers and consumers. The complex nature of electricity prices makes this information affected by high level of uncertainty, mainly due to their non-linearity, non-stationarity and time-variancy. The aim of this thesis is identifying a mathematical model capable of making reliable predictions in a short-term horizon (some hours up to some days), in order to maximize the benefits of players involved and provide them with a robust tool for planning in the near future. First, a pool of statistical models with increasing complexit have been considered: ARs, ARMAs and SARIMAs. Second, results from these models have been compared with models having the inclusion of exogenous parts, e.g. temperature, foreign price trajectories and cross-boundary exchanges of energy. Finally, the best-performing model resulting for the Swiss case study is applied to some neighbouring countries (Italy, France, Germany and Austria) in order to assess its goodness and flexibility. First, our results show the great explanatory capability of seasonal models, able to accuratly reproduce the trajectories (daily RMSE is, on average, 4.80 e/MWh) when no relevant anomaly takes place. Second, the inclusion of exogenous variables seems not to lead to an improvement of predictive performances. Conversly, the application to neighbouring countries reveals their great ability to adapt also to different contexts, outperforming most of the models from existing literature. Being the first attempt made to model the Swiss market, this work opens new opportunities for to future developments, representing a valid benchmark for further improvements.
Quasi il 60% della produzione attuale di energia elettrica in Svizzera deriva dall’idroelettrico. Questa percentuale é ulteriormente destinata a crescere nei prossimi anni a seguito della decisione di abbandonare la produzione da fonti nucleari (che copre quasi il restante 40%) per investire su risorse rinnovabili, aprendo cosí nuove sfide per il soddisfacimento della domanda energetica nazionale. Questi mercati energetici caratterizzati da un’elevata percentuale di produzione da fonti rinnovabili risultano essere piú flessibili nella gestione della produzione elettrica rispetto a quelli fortemente basati sui combustibili fossili. Per questa ragione, poter disporre di una previsione a breve termine dei prezzi futuri costituisce un’informazione preziosa sia per i produttori, i quali possono massimizzare i loro ritorni economici programmando al meglio la produzione elettrica, sia per gli stessi consumatori, che possono acquistare energia a prezzi convenienti. Le dinamiche complesse del mercato elettrico rendono il suo commercio affetto da grandi livelli di incertezza, principalmente dovuti a fenomeni quali non-stazionarietá, varianza non costante nel tempo e presenza di significative anomalie nelle serie temporali. L’obiettivo di questa tesi é dunque quello di identificare un modello matematico in grado di fornire una previsione a breve termine il piú accurata possibile delle traiettorie orarie dei prezzi elettrici, in modo da permettere agli utenti coinvolti di avere uno strumento che consenta di massimizzare i loro ritorni economici. Per questo scopo, sono stati prima di tutto analizzati una serie di modelli statistici di complessitá crescente: AR, ARMA e SARIMA. In secondo luogo, i risultati di questi modelli sono stati comparati con quelli di modelli ottenuti con l’aggiunta di variabili esogene. Infine, il migliore modello trovato per il caso svizzero é stato implementato in alcuni Stati esteri (Austria, Germania, Francia e Italia), in modo da poterne testare la bontá di adattamento e flessibilitá ad altri mercati. I risultati di questo lavoro mostrano un’ottima capacitá esplicativa da parte dei modelli stagionali (mediamente, il RMSE giornaliero é pari a 4.80 e/MWh), che sono in grado di riprodurre accuratamente l’andamento dei prezzi storici osservati quando questi non presentano significative anomalie. L’aggiunta di variabili esogene a questi modelli non ha tuttavia portato ad un miglioramento delle performances. Diversamente, l’implementazione in altri casi di studio ha mostrato una grande flessibilitá di questi modelli anche a contesti diversi, migliorando in molte situazioni i risultati precedentemente ottenuti in letteratura. Questo lavoro, poiché rappresenta il primo tentativo di modellizzazione del mercato energetico svizzero, costituisce un riferimento per ulteriori miglioramenti, ponendo dunque le basi per ampi sviluppi futuri.
Forecasting hourly energy price in the Swiss market. Model identification and application for accurate short-term forecasts
BORDIGNON, CARLO ALBERTO
2017/2018
Abstract
Switzerland is currently producing about the 60% of electricity from hydropower. This percentage is doomed to increase in the following years, since the recent decision of nuclear phase-out is opening new challenges for balancing the national energy demand. Markets where renewable sources produce a relevant share of the total energy production turn out to be much more flexible in adjusting the production than those with conventional plants. Therefore, an accurate prediction of price trajectory in the short term can help suppliers to maximize their utility by properly scheduling the production, as well as helping consumers to have successful trade of this commodity. In modern competitive electricity markets, the prediction of energy prices is thus considered a key information for both suppliers and consumers. The complex nature of electricity prices makes this information affected by high level of uncertainty, mainly due to their non-linearity, non-stationarity and time-variancy. The aim of this thesis is identifying a mathematical model capable of making reliable predictions in a short-term horizon (some hours up to some days), in order to maximize the benefits of players involved and provide them with a robust tool for planning in the near future. First, a pool of statistical models with increasing complexit have been considered: ARs, ARMAs and SARIMAs. Second, results from these models have been compared with models having the inclusion of exogenous parts, e.g. temperature, foreign price trajectories and cross-boundary exchanges of energy. Finally, the best-performing model resulting for the Swiss case study is applied to some neighbouring countries (Italy, France, Germany and Austria) in order to assess its goodness and flexibility. First, our results show the great explanatory capability of seasonal models, able to accuratly reproduce the trajectories (daily RMSE is, on average, 4.80 e/MWh) when no relevant anomaly takes place. Second, the inclusion of exogenous variables seems not to lead to an improvement of predictive performances. Conversly, the application to neighbouring countries reveals their great ability to adapt also to different contexts, outperforming most of the models from existing literature. Being the first attempt made to model the Swiss market, this work opens new opportunities for to future developments, representing a valid benchmark for further improvements.File | Dimensione | Formato | |
---|---|---|---|
thesis.pdf
non accessibile
Descrizione: testo della tesi
Dimensione
6.05 MB
Formato
Adobe PDF
|
6.05 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/10589/142587