Middle-term wind power forecasting method based on long-span NWP and microscale terrain fusion correction

As the increase of renewable energy in the power grid, the impact of its random volatility on the grid becomes more and more significant. Power forecasting technology, especially on the scale of more than 7 days, is crucial for enhancing the stability of renewable energy integration into the grid. However, current wind power forecasting technologies mainly cover 1–7 days or 1–6 h due to our limited understanding of weather patterns. Consequently, the growing need for ultra-long forecast horizons exceeding the traditional 7-day limit poses unprecedented challenges to wind power prediction technology. To meet this challenge, we propose a novel approach for medium-to-long-term wind power forecasting (WPF). The method combines a new numerical weather prediction (NWP) correction technique and a multi-resolution, long-sequence data feature extraction algorithm. For NWP correction, the Static-Dynamic Spatio-Temporal Mixture Network (SDSTMN) is proposed, which integrates mesoscale NWP data with microscale terrain information from various forecast horizons and height layers. This integration addresses the issues of accuracy degradation in long-span NWP forecasts and the limited adaptability of mesoscale models in complex terrain. For data feature extraction, the Multi-Info-Feature Fusion Network (MIFFN) is proposed as the medium-to-long-term forecasting (MLTF) model. By extracting periodicity and volatility features from long time-series sequences with varying resolutions, addressing the key challenge in tracking trends in long-term forecasting. A case study conducted at a real wind farm (WF) in China demonstrates the effectiveness of our proposed method. Specifically, we achieved an accuracy improvement of 0.66%–4.03 % on the 10th day of forecasting compared to benchmark methods.