Short-term wind power forecasting using model structure selection and data fusion techniques. (December 2016)

• Record Type: Journal Article
• Authors: Shao, Haijian; Deng, Xing
• Is Part Of: International journal of electrical power & energy systems. Volume 83(2016:Dec.)
• Page Start: 79
• Page End: 86
• Publication Date: 2016-12
• Subjects: Short-term wind power forecasting -- Model structure selection -- Manifold algorithm -- Data fusion; Electrical engineering -- Periodicals; Electric power systems -- Periodicals; Électrotechnique -- Périodiques; Réseaux électriques (Énergie) -- Périodiques; Electric power systems; Electrical engineering; Periodicals; 621.3
• DOI: 10.1016/j.ijepes.2016.03.059
• Languages: English
• ISSNs: 0142-0615

Abstract

Graphical abstract: Highlights: A solution for non-uniform data densities and 'overlapping' is given. An adaptive neighborhood selection method is presented to accelerate convergence. Fusion of dimensionality reduction combined with model structure selection is proposed. The diverse pattern of wind power and essential amount of information is evaluated. Abstract: This paper proposes a novel strategy to forecast the short-term wind power using model structure selection in combined with data fusion technique. The available inputs are usually treated as an integrated one for the predictive modeling, which ignores the fusion of the different types of inputs. This results the relationship between the multi-inputs and desired outputs are not effectively reflected in forecasting procedure. Moreover, the performances of the various types of data fusion methods are susceptible to the wind power distribution and critical factors such as smoothness, overlapping, the intrinsic amount of information evaluation and the number of neighbor points. These outlined factors can result the lower generality ability of the proposed methods due to the insufficient optimization of the model structure. So this paper presents short-term wind power forecasting method using model structure selection technique in combined with four representative fusions of dimensionality reduction methods to optimize the model structure, promote the computational efficiency and improve the forecasting accuracy. Experimental evaluation based on the real data from the wind farm in Jiangsu province is given to verify the effectiveness of the proposed method by comparing the traditional techniques.