Short-term wind speed forecasting using multivariate pretreatment technique and correntropy loss-enhanced selective combination

• Published in: Journal of wind engineering and industrial aerodynamics Vol. 254; p. 105898
• Main Authors: Jiang, Yan, Liu, Shuoyu, Zhao, Ning, Liu, Duote
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.11.2024
• Subjects: Correntropy loss; Multiple predictors; Multivariate pretreatment technique; Selective combination; Wind speed prediction; Multiple predictors; Selective combination; Wind speed prediction; Multivariate pretreatment technique; Correntropy loss
• ISSN: 0167-6105
• DOI: 10.1016/j.jweia.2024.105898

Short-term wind speed prediction is an effective measure for the rational integration of wind energy into the grid system. Subject to the complex characteristics of natural winds, achieving accurate predictions often pose a significant challenge. For this purpose, this paper develops a new hybrid forecasting method based on multivariate variational mode decomposition (MVMD), four different predictors and correntropy loss-enhanced selective combination. Specifically, MVMD is first used to decompose the multi-height wind speed data into a number of subseries groups with a well mode-alignment attribute, thereby avoiding the problem of model aliasing to some extent. Then, four predictors with different design principles (i.e., the consideration of model diversity) are constructed for capturing multiple data features. Further, the correntropy loss is used to replace the conventional mean square error loss for reflecting the actual noise environment in a robust manner. On this basis, an improved group method of data handling with high practicability is developed to realize the selective combination prediction. Finally, numerical examples based on three groups of multi-channel datasets are employed to demonstrate the forecasting ability of the proposed method. The results indicate that this method is superior to the other concerned methods. For example, compared with VMD-based method, the average improvement realized via the proposed method in term of mean absolute error is 20.3343%. •MVMD is used for high-quality data processing.•Model diversity is considered for explaining more data characteristics.•MCC is employed as an optimization criterion or a robust loss function.•An improved GMDH is developed to consider the model practicability.