发表机构
State Key Laboratory of Chemistry for NBC Hazards Protection（国家核生化防护化学重点实验室）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对极端风切变下风向预测的难题，提出可预测性引导的多尺度概率框架，通过分频建模与因果校准，在保持平静天气精度的同时显著降低极端转向误差，并揭示双峰分布以指导多模态学习。
AI 中文摘要
准确的多时域风向预测对于涡轮偏航控制和电网安全至关重要。快速方向切变（转向≥90°）通过S¹上的非欧几里得几何、多尺度动力学和依赖于状态的的不确定性对模型构成挑战。传统离散模型和基础模型在中频相位滞后和转向错位方面存在不足。我们表明，方向可预测性在不同频率子带中以不同速率衰减，使得单一机制模型次优。我们提出了一种可预测性引导的范式：缓慢的天气尺度漂移→确定性回归；中间转向→连续潜变量微分流；未解析的湍流→条件残差扩散；随后进行因果重新校准。在一个包含10,000个序列的多年基准测试中，我们的框架在保持平静天气精度（测试MCE 38.48°）的同时，降低了极端转向误差（案例1 MCE 60.69°，而零样本基础模型为70.42°）。圆形CRPS达到22.36°，在名义95%覆盖率下实际覆盖率为93.88%（分布外为91.01%）。密度估计进一步揭示了严重切变下近对映双峰结构（尾部质量≥135°占13.39%–15.43%），暴露了单中心校准覆盖不足（81.56%）的几何界限，从而推动了多模态圆形流形学习。
英文摘要
Accurate multi-horizon wind direction forecasting is critical for turbine yaw control and grid security. Rapid directional shear (turning $\ge 90^\circ$) challenges models via non-Euclidean geometry on $S^1$, multiscale dynamics, and regime-dependent uncertainty. Conventional discrete models and foundation models suffer from mid-frequency phase lag and turning misalignments. We show that directional predictability decays at disparate rates across frequency subbands, rendering monolithic mechanisms suboptimal. We propose a predictability-guided paradigm: slow synoptic drift $\to$ deterministic regression; intermediate turning $\to$ continuous latent differential flows; unresolved turbulence $\to$ conditional residual diffusion; followed by causal recalibration. On a 10,000-sequence multi-year benchmark, our framework maintains calm-weather accuracy (Test MCE $38.48^\circ$) while reducing extreme-turning error (Case 1 MCE $60.69^\circ$ vs $70.42^\circ$ for zero-shot foundation models). The circular CRPS reaches $22.36^\circ$, with 93.88\% coverage at nominal 95\% (91.01\% out-of-distribution). Density estimation further reveals near-antipodal bimodal structure under severe shear (13.39\%--15.43\% tail mass $\ge 135^\circ$), exposing a geometric bound where single-center calibration under-covers (81.56\%), motivating multimodal circular manifold learning.
Comments20 pages, 10 figures, 8 tables
