Advanced Gaussian process modelling for probabilistic wind power forecasting
The demand for more sustainable social and economic development has resulted in a rapid growth in wind power generation largely due to the highly available of wind resource worldwide. Despite that various approaches have been proposed to improve the accuracy and to overcome the uncertainties associa...
Main Author: | |
---|---|
Published: |
Queen's University Belfast
2016
|
Subjects: | |
Online Access: | http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.709864 |
id |
ndltd-bl.uk-oai-ethos.bl.uk-709864 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-bl.uk-oai-ethos.bl.uk-7098642017-07-25T03:44:07ZAdvanced Gaussian process modelling for probabilistic wind power forecastingYan, Juan2016The demand for more sustainable social and economic development has resulted in a rapid growth in wind power generation largely due to the highly available of wind resource worldwide. Despite that various approaches have been proposed to improve the accuracy and to overcome the uncertainties associated with traditional wind power methods, the stochastic nature of wind still remains the most challenging issue. A temporally local Gaussian process (TLGP) for time series forecasting is proposed to enhance the time varying adaptation and overcome the computation complexity problem. The iterative techniques are employed to achieve multi-step forecasting. Statistical analysis of the application results in real wind farms shows that the method generates smaller residuals which better follows Gaussian distribution. Such a method overcomes the strict assumption made in standard Gaussian process and reduces the computation complexity involved. Further, the uncertainty propagation for iterative multi-step forecasting is analysed based on Taylor expansion. The key finding is that TLGP not only gives better accuracy but also is less sensitive to the multi-step uncertainty propagation. The probability of successful predictions in each confidence region of multi-step TLGP better fits a Gaussian distribution. Under some unique circumstances such as ramping events, the nearby local data will contribute little to predict new generations. A hybrid Gaussian process (HGP) is proposed, combining both recently collected data and historical similar patterns. The application of HGP to real wind farm shows improved prediction accuracy and response swiftness. In addition to accurate forecasting, efficient modelling and calculation techniques are investigated. Meta heuristic methods for nonlinear optimization have been studied and TLBO is deployed. Furthermore, efficient matrix operation is achieved by combining proper matrix decomposition and least square solution methods. Therefore, advanced Gaussian process modelling for probabilistic wind power forecasting under different circumstances is accomplished.621.31Queen's University Belfasthttp://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.709864Electronic Thesis or Dissertation |
collection |
NDLTD |
sources |
NDLTD |
topic |
621.31 |
spellingShingle |
621.31 Yan, Juan Advanced Gaussian process modelling for probabilistic wind power forecasting |
description |
The demand for more sustainable social and economic development has resulted in a rapid growth in wind power generation largely due to the highly available of wind resource worldwide. Despite that various approaches have been proposed to improve the accuracy and to overcome the uncertainties associated with traditional wind power methods, the stochastic nature of wind still remains the most challenging issue. A temporally local Gaussian process (TLGP) for time series forecasting is proposed to enhance the time varying adaptation and overcome the computation complexity problem. The iterative techniques are employed to achieve multi-step forecasting. Statistical analysis of the application results in real wind farms shows that the method generates smaller residuals which better follows Gaussian distribution. Such a method overcomes the strict assumption made in standard Gaussian process and reduces the computation complexity involved. Further, the uncertainty propagation for iterative multi-step forecasting is analysed based on Taylor expansion. The key finding is that TLGP not only gives better accuracy but also is less sensitive to the multi-step uncertainty propagation. The probability of successful predictions in each confidence region of multi-step TLGP better fits a Gaussian distribution. Under some unique circumstances such as ramping events, the nearby local data will contribute little to predict new generations. A hybrid Gaussian process (HGP) is proposed, combining both recently collected data and historical similar patterns. The application of HGP to real wind farm shows improved prediction accuracy and response swiftness. In addition to accurate forecasting, efficient modelling and calculation techniques are investigated. Meta heuristic methods for nonlinear optimization have been studied and TLBO is deployed. Furthermore, efficient matrix operation is achieved by combining proper matrix decomposition and least square solution methods. Therefore, advanced Gaussian process modelling for probabilistic wind power forecasting under different circumstances is accomplished. |
author |
Yan, Juan |
author_facet |
Yan, Juan |
author_sort |
Yan, Juan |
title |
Advanced Gaussian process modelling for probabilistic wind power forecasting |
title_short |
Advanced Gaussian process modelling for probabilistic wind power forecasting |
title_full |
Advanced Gaussian process modelling for probabilistic wind power forecasting |
title_fullStr |
Advanced Gaussian process modelling for probabilistic wind power forecasting |
title_full_unstemmed |
Advanced Gaussian process modelling for probabilistic wind power forecasting |
title_sort |
advanced gaussian process modelling for probabilistic wind power forecasting |
publisher |
Queen's University Belfast |
publishDate |
2016 |
url |
http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.709864 |
work_keys_str_mv |
AT yanjuan advancedgaussianprocessmodellingforprobabilisticwindpowerforecasting |
_version_ |
1718506788290035712 |