Scholarlink

<p>Accurate liquid-level monitoring is vital for the safe operation of Liquefied Petroleum Gas (LPG) carriers. Conventional radar gauges are expensive and vulnerable to performance degradation under dynamic conditions. This paper proposes a Robust Graph Transformer Networks (RGTNs)–based Soft Sensor to enhance reliability and accuracy in radar level estimation. The RGTNs integrates a graph attention network for spatial dependency modeling and a Transformer for temporal feature extraction, enabling effective spatiotemporal fusion. Real-world experiments on an LPG carrier demonstrate that RGTNs achieves superior performance, with an R<sup>2</sup> of 0.9704, MAE of 151.12&#xa0;mm, and RMSE of 198.33&#xa0;mm, outperforming existing models. By fusing radar measurements with thermodynamically coupled auxiliary sensors through learned spatial and temporal attention, the proposed framework provides a scalable and physically interpretable solution for liquid-level monitoring in LPG storage applications under dynamic marine operating conditions.</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

Robust graph-transformer soft sensor for radar level estimation in dynamic LPG storage systems

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月24日

查看全文
  • Songqiao Bai,
  • Shidong Fan,
  • Pengcheng Wu,
  • Bin Wang

摘要

Accurate liquid-level monitoring is vital for the safe operation of Liquefied Petroleum Gas (LPG) carriers. Conventional radar gauges are expensive and vulnerable to performance degradation under dynamic conditions. This paper proposes a Robust Graph Transformer Networks (RGTNs)–based Soft Sensor to enhance reliability and accuracy in radar level estimation. The RGTNs integrates a graph attention network for spatial dependency modeling and a Transformer for temporal feature extraction, enabling effective spatiotemporal fusion. Real-world experiments on an LPG carrier demonstrate that RGTNs achieves superior performance, with an R2 of 0.9704, MAE of 151.12 mm, and RMSE of 198.33 mm, outperforming existing models. By fusing radar measurements with thermodynamically coupled auxiliary sensors through learned spatial and temporal attention, the proposed framework provides a scalable and physically interpretable solution for liquid-level monitoring in LPG storage applications under dynamic marine operating conditions.