Short-term wind power combined forecasting based on error forecast correction. (1st July 2016)

• Record Type: Journal Article
• Authors: Liang, Zhengtang; Liang, Jun; Wang, Chengfu; Dong, Xiaoming; Miao, Xiaofeng
• Is Part Of: Energy conversion and management. Volume 119(2016)
• Page Start: 215
• Page End: 226
• Publication Date: 2016-07-01
• Subjects: Wind power -- Short-term multi-step forecasting -- Combined forecasting -- Error forecast correction -- Support vector machine -- Extreme learning machine; 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.2016.04.036
• Languages: English
• ISSNs: 0196-8904

Abstract

Graphical abstract: Highlights: The correlation relationships of short-term wind power forecast errors are studied. The correlation analysis method of the multi-step forecast errors is proposed. A strategy selecting the input variables for the error forecast models is proposed. Several novel combined models based on error forecast correction are proposed. The combined models have improved the short-term wind power forecasting accuracy. Abstract: With the increasing contribution of wind power to electric power grids, accurate forecasting of short-term wind power has become particularly valuable for wind farm operators, utility operators and customers. The aim of this study is to investigate the interdependence structure of errors in short-term wind power forecasting that is crucial for building error forecast models with regression learning algorithms to correct predictions and improve final forecasting accuracy. In this paper, several novel short-term wind power combined forecasting models based on error forecast correction are proposed in the one-step ahead, continuous and discontinuous multi-step ahead forecasting modes. First, the correlation relationships of forecast errors of the autoregressive model, the persistence method and the support vector machine model in various forecasting modes have been investigated to determine whether the error forecast models can be established by regression learning algorithms. Second, according to the results of the correlation analysis, the range of input variables is defined and an efficient strategy for selecting the input variables for the error forecast models is proposed. Finally, several combined forecasting models are proposed, in which the error forecast models are based on support vector machine/extreme learning machine, and correct the short-term wind power forecast values. The data collected from a wind farm in Hebei Province, China, are selected as a case study to demonstrate the effectiveness of the proposed combined models. The simulation results show that: (1) the autocorrelation function of the one-step ahead forecast errors of support vector machine shows more significant tailing than those of the autoregressive and persistence models and the correlation relationships of the multi-step ahead forecast errors of support vector machine do significantly exist, but in the case of the autoregressive and persistence models, they do not; (2) the proposed combined models have significantly enhanced the short-term wind power forecasting accuracy in the three forecasting modes. In particular, the one-step ahead forecasting accuracies of combined models show little difference; for the continuous multi-step ahead forecasting, the improvements of the proposed combined models compared with a certain individual model increase with increasing prediction steps.