Skip to content
Conference Open access

NWPAnchor: Exogenous Gating and Weather Regime Reliability for Wind Power Forecasting

Sep 2026 · 2026 23rd International Computer Conference on Wavelet Active Media Technology and Information Processing (ICCWAMTIP) · pp. 1-6 · 0 citations · 25 references

Abstract

Accurate wind power forecasting relies on recent observations and issued numerical weather prediction (NWP), which obey different availability rules. Common multivariate models treat them as ordinary channels and often ignore that extreme weather changes weather–power relations. Sparse, overlapping physical labels also make direct weather routing unreliable. Building on XLinear’s temporal and variate gates, we propose NWPAnchor, an operational adaptation that separates measured target history from issued NWP, retains lead-specific NWP, and learns zero-start residual corrections around persistence. A separate evidence layer uses physical labels only for relation analysis, weather-subset evaluation, and validation-based interval calibration. On the public WindAI 2025 dataset, NWPAnchor reduces MAE by up to 16.92% against the strongest baseline at each horizon, with statistically supported gains at medium and long horizons. Group-specific calibration improves reliability at 24 h but not consistently at shorter horizons, and rare abrupt events remain difficult.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.