This is the authors' abstract. We don't add key points for this paper.
With the increasing integration of renewable energy into power systems, accurate short‐term wind power forecasting is essential for secure and economical operation. Existing data‐driven methods often remove the explicit organisation of multi‐height meteorological observations during feature flattening and have limited ability to represent direction‐dependent interactions among turbines. To address these issues, this paper proposes PSE‐WPF, a physics‐structured‐embedding‐based framework for spatiotemporal collaborative multi‐step wind power forecasting. A height‐aware vertical‐profile embedding is introduced to retain relative height information and facilitate the learning of cross‐height dependencies. A direction‐ and distance‐conditioned graph‐attention module is then developed to incorporate wind‐direction‐dependent and distance‐decayed structural information into turbine‐level feature aggregation. In addition, a power‐evolution consistency loss is integrated with a cascaded TCN–Informer backbone to regularise temporal variations while modelling local fluctuations and long‐range dependencies. Experiments on a private wind‐farm dataset show that PSE‐WPF achieves superior accuracy across 24‐, 48‐ and 96‐step horizons. Supplementary experiments on the public SDWPF dataset demonstrate that the adapted framework remains competitive under a different turbine configuration, sampling interval and variable structure. Because SDWPF lacks multi‐height NWP profiles, the public‐data experiment validates only the retained horizontal spatial and temporal components.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Zheng et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: