A computationally efficient electricity price forecasting model for real time energy markets. (1st April 2016)

• Record Type: Journal Article
• Authors: Feijoo, Felipe; Silva, Walter; Das, Tapas K.
• Is Part Of: Energy conversion and management. Volume 113(2016)
• Page Start: 27
• Page End: 35
• Publication Date: 2016-04-01
• Subjects: Electricity price forecasting -- Real time electricity markets -- Support vector machine -- Support vector regression; 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.2016.01.043
• Languages: English
• ISSNs: 0196-8904

Abstract

Highlights: A fast hybrid forecast model for electricity prices. Accurate forecast model that combines K-means and machine learning techniques. Low computational effort by elimination of feature selection techniques. New benchmark results by using market data for year 2012 and 2015. Abstract: Increased significance of demand response and proliferation of distributed energy resources will continue to demand faster and more accurate models for forecasting locational marginal prices. This paper presents such a model (named K-SVR). While yielding prediction accuracy comparable with the best known models in the literature, K-SVR requires a significantly reduced computational time. The computational reduction is attained by eliminating the use of a feature selection process, which is commonly used by the existing models in the literature. K-SVR is a hybrid model that combines clustering algorithms, support vector machine, and support vector regression. K-SVR is tested using Pennsylvania–New Jersey–Maryland market data from the periods 2005–6, 2011–12, and 2014–15. Market data from 2006 has been used to measure performance of many of the existing models. Authors chose these models to compare performance and demonstrate strengths of K-SVR. Results obtained from K-SVR using the market data from 2012 and 2015 are new, and will serve as benchmark for future models.