Non-parametric hybrid models for wind speed forecasting. (15th September 2017)

• Record Type: Journal Article
• Authors: Han, Qinkai; Meng, Fanman; Hu, Tao; Chu, Fulei
• Is Part Of: Energy conversion and management. Volume 148(2017)
• Page Start: 554
• Page End: 568
• Publication Date: 2017-09-15
• Subjects: Wind speed forecasting -- Hybrid algorithm -- Non-parametric modeling -- Autoregressive moving average; Direct energy conversion -- Periodicals; Energy storage -- Periodicals; Energy transfer -- Periodicals; Énergie -- Conversion directe -- Périodiques; Direct energy conversion; Periodicals; 621.3105
• DOI: 10.1016/j.enconman.2017.06.021
• Languages: English
• ISSNs: 0196-8904

Abstract

Graphical abstract: Highlights: Two hybrid models are proposed by combining non-parametric and ARMA models. Various tests on the real data from six wind sites of China are conducted. Performance of both single and hybrid models is compared and evaluated. NP hybrid models generally outperform the other models and have robust performances. Introduction of ARMA model for NP fitting residuals could greatly improve accuracy. Abstract: It is essential to predict the wind speed accurately in order for protecting the security of wind power integration. The aim of this study is to develop non-parametric hybrid models for probabilistic wind speed forecasting. By adopting the non-parametric models, two hybrid models, namely the hybrid autoregressive moving average/non-parametric and hybrid non-parametric/autoregressive moving average models, are proposed and their performance is compared. In the hybrid autoregressive moving average/non-parametric models, the residuals obtained after fitting with the autoregressive moving average model are studied by the non-parametric model to extract the nonlinear part of the data. In the hybrid non-parametric/autoregressive moving average models, the residuals obtained from the non-parametric model are fitted by the autoregressive moving average model. In order for comparisons, the artificial neural network with back propagation, support vector machine and random forest models are also introduced for hybrid modeling. Through conducting various tests on the real hourly wind speed time series, the prediction performance of both single and hybrid models is compared and evaluated in detail. The results of this study are to show that non-parametric based hybrid models generally outperform the other models and have more robust forecast performances. When the single autoregressive moving average model basically outperforms the single non-parametric models, the introduction of the autoregressive moving average model for the residuals from the non-parametric fitting is possible to obtain better prediction accuracy.