<p>Existing decomposition methods employ a divide-and-conquer strategy to extract diverse temporal frequency patterns but encounter challenges related to aleatoric uncertainty and epistemic uncertainty within machine learning frameworks. However, limited research has focused on quantitatively analyzing the irreducible stochastic nature of aleatoric uncertainty. To enhance the accuracy of wind speed prediction and effectively quantify its inherent uncertainty, this paper proposes an interval prediction method for wind speed by introducing a novel linear integral method. Firstly, leveraging seasonal characteristics, the raw wind speed data is partitioned into four distinct seasonal datasets (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">${S}_{spring}, {S}_{summer}, {S}_{autumn}, {S}_{winter}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msub> <mi>S</mi> <mrow> <mi mathvariant="italic">spring</mi> </mrow> </msub> <mo>,</mo> <msub> <mi>S</mi> <mrow> <mi mathvariant="italic">summer</mi> </mrow> </msub> <mo>,</mo> <msub> <mi>S</mi> <mrow> <mi mathvariant="italic">autumn</mi> </mrow> </msub> <mo>,</mo> <msub> <mi>S</mi> <mrow> <mi mathvariant="italic">winter</mi> </mrow> </msub> </mrow> </math></EquationSource> </InlineEquation>) to account for the impact of seasonal variations. Then, aleatoric uncertainty analysis is applied to wind speed temporal patterns by considering the decomposition and prediction process with adaptive noise method, extracting intrinsic mode functions (IMFs) to capture the time-series characteristics of wind speeds and capitalizing on the respective strengths of multiple machine learning algorithms. Through linear integral and fusion of the prediction results for all IMFs, prediction intervals with a 95% confidence level are constructed, thereby quantifying the prediction uncertainty. Finally, the optimal model combination for each season is determined through a comprehensive evaluation of the predictive performance across different model ensembles. Experimental results demonstrate that the proposed interval prediction method effectively quantifies the uncertainty of wind speed signals across different seasons.</p>

A novel linear integral method based on aleatoric uncertainty for short-term wind speed prediction

期刊论文
发表日期：

2025年10月18日

查看全文

Lei Zhou,
Jinxing Che,
Wei Dong,
Xiaoqing Wang,
Qinghua Zhang

摘要

Existing decomposition methods employ a divide-and-conquer strategy to extract diverse temporal frequency patterns but encounter challenges related to aleatoric uncertainty and epistemic uncertainty within machine learning frameworks. However, limited research has focused on quantitatively analyzing the irreducible stochastic nature of aleatoric uncertainty. To enhance the accuracy of wind speed prediction and effectively quantify its inherent uncertainty, this paper proposes an interval prediction method for wind speed by introducing a novel linear integral method. Firstly, leveraging seasonal characteristics, the raw wind speed data is partitioned into four distinct seasonal datasets ( ${S}_{spring}, {S}_{summer}, {S}_{autumn}, {S}_{winter}$ $S_{spring}, S_{summer}, S_{autumn}, S_{winter}$ ) to account for the impact of seasonal variations. Then, aleatoric uncertainty analysis is applied to wind speed temporal patterns by considering the decomposition and prediction process with adaptive noise method, extracting intrinsic mode functions (IMFs) to capture the time-series characteristics of wind speeds and capitalizing on the respective strengths of multiple machine learning algorithms. Through linear integral and fusion of the prediction results for all IMFs, prediction intervals with a 95% confidence level are constructed, thereby quantifying the prediction uncertainty. Finally, the optimal model combination for each season is determined through a comprehensive evaluation of the predictive performance across different model ensembles. Experimental results demonstrate that the proposed interval prediction method effectively quantifies the uncertainty of wind speed signals across different seasons.