Scholarlink

<p>With frequent extreme weather and increasing urban waterlogging damage, establishing fast and accurate prediction models is crucial. However, uneven monitoring data quality due to delayed urban smartization presents a major challenge. Thus, developing robust models that effectively utilize limited data is essential. This paper proposes a Physics-based Time Lag Correlation (PTLC) analysis method, which, combined with spatial decoupling, provides prior analysis for model prediction. Additionally, a Parameter Optimization Automatic Rolling Long Short-Term Memory network (Poar_LSTM) is developed and coupled with a hydrodynamic model to form the Physically-guided and Optimization-based Automatic Rolling Deep Hydrodynamic Coupled model (Poar_DHC) framework. Using Fuzhou’s Doumen area as a case study (2021–2022 rainfall events), results show Poar_LSTM accurately predicts river levels, achieving Nash efficiency coefficients of 0.969 and 0.971. While different data-driven models slightly affect coupling performance, Poar_DHC achieves the highest accuracy in predicting underground water levels. Furthermore, PTLC effectively guides both prior analysis and posterior evaluation across various rainfall scenarios. This study provides a scientific reference for rapid and accurate storm-flood forecasting.</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

Urban Waterlogging Prediction Based on Time Lag Correlation Analysis and Multi-Model Coupling

  • 期刊论文

  • 发表日期：

    2026年4月23日

查看全文
  • Xiaohui Lei,
  • Dongkun Liu,
  • Yan Long,
  • Haocheng Huang

摘要

With frequent extreme weather and increasing urban waterlogging damage, establishing fast and accurate prediction models is crucial. However, uneven monitoring data quality due to delayed urban smartization presents a major challenge. Thus, developing robust models that effectively utilize limited data is essential. This paper proposes a Physics-based Time Lag Correlation (PTLC) analysis method, which, combined with spatial decoupling, provides prior analysis for model prediction. Additionally, a Parameter Optimization Automatic Rolling Long Short-Term Memory network (Poar_LSTM) is developed and coupled with a hydrodynamic model to form the Physically-guided and Optimization-based Automatic Rolling Deep Hydrodynamic Coupled model (Poar_DHC) framework. Using Fuzhou’s Doumen area as a case study (2021–2022 rainfall events), results show Poar_LSTM accurately predicts river levels, achieving Nash efficiency coefficients of 0.969 and 0.971. While different data-driven models slightly affect coupling performance, Poar_DHC achieves the highest accuracy in predicting underground water levels. Furthermore, PTLC effectively guides both prior analysis and posterior evaluation across various rainfall scenarios. This study provides a scientific reference for rapid and accurate storm-flood forecasting.