Short-term density forecasting of wave energy using ARMA-GARCH models and Kernel density estimation

Jooyoung Jeon, James W. Taylor

Research output: Contribution to journalArticle

  • 4 Citations

Abstract

Wave energy has great potential as a renewable source of electricity. Installed capacity is increasing, and developments in technology mean that wave energy is likely to play an important role in the future mix of electricity generation. Short-term forecasts of wave energy are required for the efficient operation of wave farms and power grids, as well as for energy trading. The intermittent nature of wave energy motivates the use of probabilistic forecasting. In this paper, we evaluate the accuracy of probabilistic forecasts of wave energy flux from a variety of methods, including unconditional and conditional kernel density estimation, univariate and bivariate autoregressive moving average generalised autoregressive conditional heteroskedasticity (ARMA-GARCH) models, and a regression-based method. The bivariate ARMA-GARCH models are implemented with different pairs of variables, such as (1) wave height and wave period, and (2) wave energy flux and wind speed. Our empirical analysis uses hourly data from the FINO1 research platform in the North Sea to evaluate density and point forecasts, up to 24 h ahead, for the wave energy flux. The empirical study indicates that a bivariate ARMA-GARCH model for wave height and wave period led to the greatest accuracy overall for wave energy flux density forecasting, but its usefulness for point forecasting decreases as the lead time increases. The model also performed well for wave power data that had been generated from wave height and wave period observations using a conversion matrix.
LanguageEnglish
Pages991-1004
JournalInternational Journal of Forecasting
Volume32
Issue number3
Early online date19 Dec 2015
DOIs
StatusPublished - Jul 2016

Fingerprint

Energy
Kernel density estimation
Autoregressive conditional heteroskedasticity
Autoregressive moving average
Density forecasting
Electricity generation
Lead time
Density forecasts
Empirical study
Grid
Farm
North Sea
Usefulness
Electricity
Empirical analysis
Point forecasts

Keywords

  • Sustainability
  • Wave energy
  • Probability density
  • ARMA-GARCH
  • Kernel density estimation

Cite this

Short-term density forecasting of wave energy using ARMA-GARCH models and Kernel density estimation. / Jeon, Jooyoung; Taylor, James W.

In: International Journal of Forecasting, Vol. 32, No. 3, 07.2016, p. 991-1004.

Research output: Contribution to journalArticle

@article{c0cd6f88827149a4bac4069f69e1d22e,
title = "Short-term density forecasting of wave energy using ARMA-GARCH models and Kernel density estimation",
abstract = "Wave energy has great potential as a renewable source of electricity. Installed capacity is increasing, and developments in technology mean that wave energy is likely to play an important role in the future mix of electricity generation. Short-term forecasts of wave energy are required for the efficient operation of wave farms and power grids, as well as for energy trading. The intermittent nature of wave energy motivates the use of probabilistic forecasting. In this paper, we evaluate the accuracy of probabilistic forecasts of wave energy flux from a variety of methods, including unconditional and conditional kernel density estimation, univariate and bivariate autoregressive moving average generalised autoregressive conditional heteroskedasticity (ARMA-GARCH) models, and a regression-based method. The bivariate ARMA-GARCH models are implemented with different pairs of variables, such as (1) wave height and wave period, and (2) wave energy flux and wind speed. Our empirical analysis uses hourly data from the FINO1 research platform in the North Sea to evaluate density and point forecasts, up to 24 h ahead, for the wave energy flux. The empirical study indicates that a bivariate ARMA-GARCH model for wave height and wave period led to the greatest accuracy overall for wave energy flux density forecasting, but its usefulness for point forecasting decreases as the lead time increases. The model also performed well for wave power data that had been generated from wave height and wave period observations using a conversion matrix.",
keywords = "Sustainability, Wave energy, Probability density, ARMA-GARCH, Kernel density estimation",
author = "Jooyoung Jeon and Taylor, {James W.}",
year = "2016",
month = "7",
doi = "10.1016/j.ijforecast.2015.11.003",
language = "English",
volume = "32",
pages = "991--1004",
journal = "International Journal of Forecasting",
issn = "0169-2070",
publisher = "Elsevier",
number = "3",

}

TY - JOUR

T1 - Short-term density forecasting of wave energy using ARMA-GARCH models and Kernel density estimation

AU - Jeon,Jooyoung

AU - Taylor,James W.

PY - 2016/7

Y1 - 2016/7

N2 - Wave energy has great potential as a renewable source of electricity. Installed capacity is increasing, and developments in technology mean that wave energy is likely to play an important role in the future mix of electricity generation. Short-term forecasts of wave energy are required for the efficient operation of wave farms and power grids, as well as for energy trading. The intermittent nature of wave energy motivates the use of probabilistic forecasting. In this paper, we evaluate the accuracy of probabilistic forecasts of wave energy flux from a variety of methods, including unconditional and conditional kernel density estimation, univariate and bivariate autoregressive moving average generalised autoregressive conditional heteroskedasticity (ARMA-GARCH) models, and a regression-based method. The bivariate ARMA-GARCH models are implemented with different pairs of variables, such as (1) wave height and wave period, and (2) wave energy flux and wind speed. Our empirical analysis uses hourly data from the FINO1 research platform in the North Sea to evaluate density and point forecasts, up to 24 h ahead, for the wave energy flux. The empirical study indicates that a bivariate ARMA-GARCH model for wave height and wave period led to the greatest accuracy overall for wave energy flux density forecasting, but its usefulness for point forecasting decreases as the lead time increases. The model also performed well for wave power data that had been generated from wave height and wave period observations using a conversion matrix.

AB - Wave energy has great potential as a renewable source of electricity. Installed capacity is increasing, and developments in technology mean that wave energy is likely to play an important role in the future mix of electricity generation. Short-term forecasts of wave energy are required for the efficient operation of wave farms and power grids, as well as for energy trading. The intermittent nature of wave energy motivates the use of probabilistic forecasting. In this paper, we evaluate the accuracy of probabilistic forecasts of wave energy flux from a variety of methods, including unconditional and conditional kernel density estimation, univariate and bivariate autoregressive moving average generalised autoregressive conditional heteroskedasticity (ARMA-GARCH) models, and a regression-based method. The bivariate ARMA-GARCH models are implemented with different pairs of variables, such as (1) wave height and wave period, and (2) wave energy flux and wind speed. Our empirical analysis uses hourly data from the FINO1 research platform in the North Sea to evaluate density and point forecasts, up to 24 h ahead, for the wave energy flux. The empirical study indicates that a bivariate ARMA-GARCH model for wave height and wave period led to the greatest accuracy overall for wave energy flux density forecasting, but its usefulness for point forecasting decreases as the lead time increases. The model also performed well for wave power data that had been generated from wave height and wave period observations using a conversion matrix.

KW - Sustainability

KW - Wave energy

KW - Probability density

KW - ARMA-GARCH

KW - Kernel density estimation

UR - http://dx.doi.org/10.1016/j.ijforecast.2015.11.003

U2 - 10.1016/j.ijforecast.2015.11.003

DO - 10.1016/j.ijforecast.2015.11.003

M3 - Article

VL - 32

SP - 991

EP - 1004

JO - International Journal of Forecasting

T2 - International Journal of Forecasting

JF - International Journal of Forecasting

SN - 0169-2070

IS - 3

ER -