Open Access

ARTICLE

Wind Power Aggregation Forecasting Method Considering Power Fluctuation Uncertainty

Gangui Yan, Lu Chen, Aolan Xing^*, Jianshu Li, Leiyujie Xiao
Key Laboratory of Modern Power System Simulation and Control & Renewable Energy Technology, Ministry of Education (Northeast Electric Power University), Jilin, China
* Corresponding Author: Aolan Xing. Email:
(This article belongs to the Special Issue: Advances in Grid Integration and Electrical Engineering of Wind Energy Systems: Innovations, Challenges, and Applications)

Energy Engineering https://doi.org/10.32604/ee.2026.076781

Received 26 November 2025; Accepted 27 January 2026; Published online 18 March 2026

Abstract

Wind power prediction is affected by seasonal meteorological conditions, exhibiting significant multi-scale spatio-temporal coupling characteristics and random volatility. Existing models still fail to consider the volatility and uncertainty of power aggregation under seasonal characteristics. Therefore, based on the data of 36 wind farms in three regions (Siping, Baicheng, and Songyuan) in Jilin Province, this paper proposes a wind power prediction method based on multi-level spatio-temporal feature fusion. In the time dimension, this method uses the temporal convolutional network and the seasonal embedding module to jointly model the multi-scale trends and fluctuation characteristics of the power sequence, achieving the extraction of power seasonal fluctuation characteristics based on convolution. In the spatial dimension, a regional power aggregation attenuation model based on the graph attention mechanism is constructed. A regional graph structure is established with wind farms as nodes, and the spatial dependence and power propagation laws between regions are adaptively learned in combination with the geographical attenuation function. In the prediction stage, Monte Carlo Dropout is introduced to jointly estimate the mean and variance of power prediction, thereby quantifying the prediction uncertainty and confidence interval under seasonal fluctuations. Experimental results show that the proposed method outperforms the comparison models in terms of prediction accuracy and credibility evaluation.

---

Keywords

Wind power prediction; power fluctuation; spatiotemporal feature fusion; graph attention mechanism; Monte Carlo simulation

---