Scholarlink

<p>Accurate low-visibility prediction and real-time fog detection remain challenging, especially for sudden localized events where current methods often respond slowly, offer limited spatial resolution, and produce frequent false alarms. This study presents a hybrid machine learning framework that integrates video-based fog density estimation (using MVG modeling) with real-time atmospheric observations to bridge spatial and temporal data gaps. The framework includes a data preparation module and a training module combining LSTM and XGBoost. Experimentally, it achieves strong reconstruction performance with a test RMSE of 121.48&#xa0;m and an R2 of 0.935, improving R2 by 5.67% over other LSTM hybrids. The optimized LSTM–XGBoost model also outperforms both baseline models and unoptimized variants. These results confirm that the framework effectively utilizes video-derived fog density to dynamically calibrate visibility and deliver fast, accurate fog impact reconstruction.</p>

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

An intelligent hybrid machine learning framework for low-visibility reconstruction for airport transportation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月10日

查看全文
  • Fang-wei Liu,
  • Ren-jie Wu

摘要

Accurate low-visibility prediction and real-time fog detection remain challenging, especially for sudden localized events where current methods often respond slowly, offer limited spatial resolution, and produce frequent false alarms. This study presents a hybrid machine learning framework that integrates video-based fog density estimation (using MVG modeling) with real-time atmospheric observations to bridge spatial and temporal data gaps. The framework includes a data preparation module and a training module combining LSTM and XGBoost. Experimentally, it achieves strong reconstruction performance with a test RMSE of 121.48 m and an R2 of 0.935, improving R2 by 5.67% over other LSTM hybrids. The optimized LSTM–XGBoost model also outperforms both baseline models and unoptimized variants. These results confirm that the framework effectively utilizes video-derived fog density to dynamically calibrate visibility and deliver fast, accurate fog impact reconstruction.