Scholarlink

<p>Incipient motion denotes the critical condition when immobile particles at rest on the channel bed are set into motion. Understanding incipient motion is relevant for sediment transport estimates in alluvial channels, infrastructure protection, and channel design. This research applied two machine learning regression approaches for incipient motion, namely Convolutional Neural Network (CNN) and Deep Neural Network (DNN), to estimate the bed shear stress (τ<sub>b</sub>) required to initiate sediment motion. The features presented were water depth, water discharge, water surface slope, sediment size (<i>d₅₀</i>), specific gravity (<i>G</i>), and hydraulic radius. When assessing appropriate performance metrics for model predictions, the study used multiple metrics like R<sup>2</sup>, RMSE, NSE, PCC, and MSE. Overall, comparing the outputs of both models, the CNN had an R<sup>2</sup> of 0.9542 and a PCC of 0.9773, indicating relatively good, smooth convergence of bed shear stress (τ<sub>b</sub>) estimates compared to the original data. The DNN was representative, capturing the spatial relationships among the input variables; however, it yielded an R<sup>2</sup> of 0.9239, comparable to that of the CNN. The outputs of both models were presented as scatter plots and comparison plots. This study will broaden understanding of sediment transport dynamics, enhance channel stability analysis, and support decision-making in the design and management of riverine and hydraulic structures.</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

Deep Learning-Based Prediction of Incipient Motion in Alluvial Channel

  • 期刊论文

  • 发表日期：

    2026年7月14日

查看全文
  • Mun Mun Basumatary,
  • Soumen Maji,
  • Bimlesh Kumar

摘要

Incipient motion denotes the critical condition when immobile particles at rest on the channel bed are set into motion. Understanding incipient motion is relevant for sediment transport estimates in alluvial channels, infrastructure protection, and channel design. This research applied two machine learning regression approaches for incipient motion, namely Convolutional Neural Network (CNN) and Deep Neural Network (DNN), to estimate the bed shear stress (τb) required to initiate sediment motion. The features presented were water depth, water discharge, water surface slope, sediment size (d₅₀), specific gravity (G), and hydraulic radius. When assessing appropriate performance metrics for model predictions, the study used multiple metrics like R2, RMSE, NSE, PCC, and MSE. Overall, comparing the outputs of both models, the CNN had an R2 of 0.9542 and a PCC of 0.9773, indicating relatively good, smooth convergence of bed shear stress (τb) estimates compared to the original data. The DNN was representative, capturing the spatial relationships among the input variables; however, it yielded an R2 of 0.9239, comparable to that of the CNN. The outputs of both models were presented as scatter plots and comparison plots. This study will broaden understanding of sediment transport dynamics, enhance channel stability analysis, and support decision-making in the design and management of riverine and hydraulic structures.