A novel probabilistic wind power forecasting framework integrating similar curve matching mechanism and an enhanced conditional diffusion model

Achieving carbon neutrality is a global priority, with wind power playing a key role in the sustainable energy transition. However, its inherent uncertainty poses challenges to power grid stability. To address this, we propose a novel probabilistic wind power forecasting framework that integrates a Similar Curve Matching Mechanism (SCMM) with an enhanced conditional diffusion model to improve uncertainty quantification. We introduce a Conditional-Denoising Joint Network (CDJN) that dynamically processes condition information, enhancing model expressiveness compared to static encoding. To account for the heterogeneity of wind power generation, SCMM leverages K-Medoids clustering and Fast-Dynamic Time Warping (Fast-DTW) to generate customized training sets, enabling the model to capture diverse wind power patterns with greater accuracy. The trained conditional diffusion model generates a large set of wind power scenarios, from which a Gated Recurrent Unit-based Autoencoder (GRUA) and Kernel Density Estimation (KDE) are applied to produce reliable probabilistic forecasts with time dependencies. This novel approach effectively enhances forecasting precision and adaptability. Case studies using real wind farm data from Ordos, Inner Mongolia, China, demonstrate that the proposed framework significantly outperforms state-of-the-art models across multiple evaluation metrics. Notably, the ablation study indicates that removing key components caused a 469.21 % increase in average loss, highlighting the critical role of each module. The results highlight its improved ability to measure wind power uncertainty and improve forecasting accuracy, hence being a promising solution to incorporate wind power into power systems more reliably.