A Decomposition–Graph Attention–Patch Transformer Framework with Conformal Calibration for Short-Term Wind Power Forecasting

Read full text Summarise with VIRAT
Authors: Yirui Wang, Yijia Liu, Zhaomeng Zhang
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Accurate short-term wind power forecasting is essential for power system dispatching because wind generation exhibits strong intermittency, non-stationarity, and nonlinear dependence on meteorological conditions. This study proposes a hybrid machine learning framework that integrates multi-resolution signal decomposition, spatio-temporal representation learning, and probabilistic uncertainty calibration for wind power forecasting. First, historical wind power and meteorological time series are decomposed into multiple components to alleviate non-stationary fluctuations and to enhance predictability across frequency bands. Second, spatial dependencies among wind turbines (or measurement sites) are modeled through a graph-based encoder using Graph Attention Network (GAT) to capture dynamic inter-node correlations. Third, temporal patterns are learned by a Patch-based Transformer that processes segmented sequence patches to efficiently represent long-range dependencies for multi-step forecasting horizons. Finally, probabilistic prediction intervals are produced via quantile regression and further calibrated using Conformal Prediction (CP) to improve coverage reliability under distribution shifts and missing observations. Experimental results on wind farm datasets demonstrate that the proposed framework achieves improved accuracy and robustness compared with classical statistical methods and representative deep learning baselines, while providing reliable uncertainty estimates for operational decision-making.

Fields of Study & Subjects

Energy Load and Power Forecasting Solar Radiation and Photovoltaics Traffic Prediction and Management Techniques

Publication details

Journal / source
Automation and Machine Learning
Volume
7
Issue
1
Publication date
1 January 2026
Language
EN