A novel implementation of kNN classifier based on multi-tupled meteorological input data for wind power prediction. (1st March 2017)

• Record Type: Journal Article
• Authors: Yesilbudak, Mehmet; Sagiroglu, Seref; Colak, Ilhami
• Is Part Of: Energy conversion and management. Volume 135(2017)
• Page Start: 434
• Page End: 444
• Publication Date: 2017-03-01
• Subjects: Wind power production -- Very short-term prediction -- Multidimensional meteorological data -- k-nearest neighbor classifier; 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.2016.12.094
• Languages: English
• ISSNs: 0196-8904

Abstract

Highlights: An accurate wind power prediction model is proposed for very short-term horizon. The k-nearest neighbor classifier is implemented based on the multi-tupled inputs. The variation of wind power prediction errors is evaluated in various aspects. Our approach shows the superior prediction performance over the persistence method. Abstract: With the growing share of wind power production in the electric power grids, many critical challenges to the grid operators have been emerged in terms of the power balance, power quality, voltage support, frequency stability, load scheduling, unit commitment and spinning reserve calculations. To overcome such problems, numerous studies have been conducted to predict the wind power production, but a small number of them have attempted to improve the prediction accuracy by employing the multidimensional meteorological input data. The novelties of this study lie in the proposal of an efficient and easy to implement very short-term wind power prediction model based on the k-nearest neighbor classifier (kNN), in the usage of wind speed, wind direction, barometric pressure and air temperature parameters as the multi-tupled meteorological inputs and in the comparison of wind power prediction results with respect to the persistence reference model. As a result of the achieved patterns, we characterize the variation of wind power prediction errors according to the input tuples, distance measures and neighbor numbers, and uncover the most influential and the most ineffective meteorological parameters on the optimization of wind power prediction results.