Scholarlink

<p>The Storm Water Management Model (SWMM) is widely applied to simulate urban flood disasters and water resources management. Accurate parameter calibration is crucial for model performance. To enhance calibration efficiency and accuracy, this study combines the advantages of kernel principal component analysis (KPCA), sparrow search algorithm (SSA), and back propagation neural network (BPNN) to propose a new framework for SWMM model parameter calibration. KPCA extracts key features from high-dimensional calibration data, creating a reduced-dimension training dataset for BPNN. SSA optimizes the initial weights and thresholds of BPNN, leading to a KPCA-SSA-BPNN surrogate model that accurately captures the relationship between SWMM parameters and simulation results. This approach facilitates rapid and precise parameter calibration. A sub-catchment in Zhuzhou City, Hunan Province, China, was selected as the study area, using seven rainfall events for calibration and validation. Compared with manual calibration, standalone BPNN, and other hybrid methods (PSO-BPNN, SSA-BPNN, PCA-PSO-BPNN, PCA-SSA-BPNN, and KPCA-PSO-BPNN), the KPCA-SSA-BPNN method demonstrated superior performance, achieving an effective calibration probability of 93.8%, significantly higher than BPNN’s 40.3%. Furthermore, KPCA-SSA-BPNN consistently yielded higher Nash–Sutcliffe efficiency (<i>NSE</i>) and relative error of node peak water depth (<i>RE</i><sub><i>p</i></sub>) in most scenarios. With an average <i>NSE</i> of 0.913, <i>RE</i><sub><i>p</i></sub> of 2.847%, and a calibration time of 22.7 s, this method demonstrates considerable feasibility and practicality for real-world engineering applications.</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 Automatic Calibration Framework of Storm Water Management Model Based on KPCA-SSA-BPNN and its Application in Urban Stormwater Flood Simulation

  • 期刊论文

  • 发表日期：

    2025年3月21日

查看全文
  • Shuichang Liu,
  • Zhihao Xue,
  • Xin Chen,
  • Renhui Pan,
  • Yong Zhang,
  • Zelin Zhong

摘要

The Storm Water Management Model (SWMM) is widely applied to simulate urban flood disasters and water resources management. Accurate parameter calibration is crucial for model performance. To enhance calibration efficiency and accuracy, this study combines the advantages of kernel principal component analysis (KPCA), sparrow search algorithm (SSA), and back propagation neural network (BPNN) to propose a new framework for SWMM model parameter calibration. KPCA extracts key features from high-dimensional calibration data, creating a reduced-dimension training dataset for BPNN. SSA optimizes the initial weights and thresholds of BPNN, leading to a KPCA-SSA-BPNN surrogate model that accurately captures the relationship between SWMM parameters and simulation results. This approach facilitates rapid and precise parameter calibration. A sub-catchment in Zhuzhou City, Hunan Province, China, was selected as the study area, using seven rainfall events for calibration and validation. Compared with manual calibration, standalone BPNN, and other hybrid methods (PSO-BPNN, SSA-BPNN, PCA-PSO-BPNN, PCA-SSA-BPNN, and KPCA-PSO-BPNN), the KPCA-SSA-BPNN method demonstrated superior performance, achieving an effective calibration probability of 93.8%, significantly higher than BPNN’s 40.3%. Furthermore, KPCA-SSA-BPNN consistently yielded higher Nash–Sutcliffe efficiency (NSE) and relative error of node peak water depth (REp) in most scenarios. With an average NSE of 0.913, REp of 2.847%, and a calibration time of 22.7 s, this method demonstrates considerable feasibility and practicality for real-world engineering applications.