Scholarlink

Trajectory control for stratospheric balloons relies on accurate modeling of complex wind fields.However, stratospheric winds exhibit complex multi-scale dynamics and spatio-temporal non-stationarity, posing significant modeling challenges. While existing high-fidelity numerical models are computationally expensive and unsuitable for developing control strategies like reinforcement learning (RL), purely data-driven models often lack physical realism. We propose a Wavelet-Fourier Physics-informed Latent Diffusion Model (WF-PLDM). It combines Wavelet and Fourier transforms to capture the wind field’s multi-scale characteristics and integrates Navier-Stokes equations to ensure physical consistency. Quantitative evaluations on the ERA5 reanalysis dataset demonstrate that the proposed WF-PLDM outperforms several baseline generative models in generation accuracy and physical consistency metrics. Our approach provides a reliable simulation environment conducive to training advanced RL-based control strategies for stratospheric balloons.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Stratospheric Wind Field Simulation Using Physics-Constrained Latent Diffusion Model

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Jianquan Ouyang,
  • Zexiang Zi

摘要

Trajectory control for stratospheric balloons relies on accurate modeling of complex wind fields.However, stratospheric winds exhibit complex multi-scale dynamics and spatio-temporal non-stationarity, posing significant modeling challenges. While existing high-fidelity numerical models are computationally expensive and unsuitable for developing control strategies like reinforcement learning (RL), purely data-driven models often lack physical realism. We propose a Wavelet-Fourier Physics-informed Latent Diffusion Model (WF-PLDM). It combines Wavelet and Fourier transforms to capture the wind field’s multi-scale characteristics and integrates Navier-Stokes equations to ensure physical consistency. Quantitative evaluations on the ERA5 reanalysis dataset demonstrate that the proposed WF-PLDM outperforms several baseline generative models in generation accuracy and physical consistency metrics. Our approach provides a reliable simulation environment conducive to training advanced RL-based control strategies for stratospheric balloons.