Adaptive boundary extension in decomposition–ensemble forecasting of non-stationary wind speed time series

Abstract Accurate wind speed forecasting plays a crucial role in environmental modeling and risk evaluation, serving as a foundation for mitigating wind-induced hazards and ensuring the safety of critical infrastructures. In wind speed forecasting, the decomposition–ensemble strategy (DES) has been widely adopted as a preprocessing and fusion approach to improve model performance, yet its practical application remains limited. Two key issues hinder its effectiveness: (i) information leakage caused by improper data partitioning, and (ii) boundary effects (BES) that distort decomposed sub-sequences during rolling forecasting. To address these challenges, this study proposes an improved DES framework integrating an adaptive boundary extension algorithm based on historical similarity matching. The algorithm smooths signal discontinuities and preserves frequency-domain integrity during real-time decomposition. Using measured wind speed data, three groups of comparative experiments were conducted across different baseline models. Results demonstrate that the proposed method significantly reduces boundary-induced errors, enhances forecasting stability, and achieves accuracy levels suitable for practical deployment.