This paper presents a flexible and easy-to-use methodological approach to the forecasting of energy production in solar photovoltaic (PV) installations, using time series analysis and neural networks. The aim is to develop and validate a one day-ahead forecasting model by adopting an artificial neural network with tapped delay lines. The main novelty of our approach is the proposal of a general methodology, consisting of a sequence of steps to perform in order to find, based on heuristic criteria, the optimal structure of the neural network (particularly, number of hidden neurons and number of delay elements) and the best configuration of the neural predictor (namely, the training window width and the sampling frequency). The best experimental results have been obtained using as inputs the irradiation and the sampling hour to predict the daily accumulated energy. Considering a dataset of 15-minute measurements pertinent to one year, despite the presence of 77 missing days (scattered through the whole year in correspondence with system slowdown), we achieved seasonal mean absolute percentage errors ranging from a minimum of 12.2% (Spring) to a maximum of 26% (Autumn). Moreover the achieved results are significantly better than those obtained by the persistence method, a benchmark frequently used in this kind of applications.

One day-ahead forecasting of energy production in solar photovoltaic installations: an empirical study

COCOCCIONI, MARCO;D'ANDREA, ELEONORA;LAZZERINI, BEATRICE
2012

Abstract

This paper presents a flexible and easy-to-use methodological approach to the forecasting of energy production in solar photovoltaic (PV) installations, using time series analysis and neural networks. The aim is to develop and validate a one day-ahead forecasting model by adopting an artificial neural network with tapped delay lines. The main novelty of our approach is the proposal of a general methodology, consisting of a sequence of steps to perform in order to find, based on heuristic criteria, the optimal structure of the neural network (particularly, number of hidden neurons and number of delay elements) and the best configuration of the neural predictor (namely, the training window width and the sampling frequency). The best experimental results have been obtained using as inputs the irradiation and the sampling hour to predict the daily accumulated energy. Considering a dataset of 15-minute measurements pertinent to one year, despite the presence of 77 missing days (scattered through the whole year in correspondence with system slowdown), we achieved seasonal mean absolute percentage errors ranging from a minimum of 12.2% (Spring) to a maximum of 26% (Autumn). Moreover the achieved results are significantly better than those obtained by the persistence method, a benchmark frequently used in this kind of applications.
Cococcioni, Marco; D'Andrea, Eleonora; Lazzerini, Beatrice
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11568/205017
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 16
  • ???jsp.display-item.citation.isi??? 17
social impact