Scholarlink

<p>The accurate estimation of the soil permeability coefficient (<i>k</i>) is essential for civil engineering projects. This study employed an experimental dataset and effective parameters to model soil permeability. The specific particle sizes of soil, including d10, d50, and d60, and void ratio (e), were used as input parameters for the estimation of <i>k</i>. The artificial neural network (ANN) method was employed, along with two widely used metaheuristic algorithms, including particle swarm optimization (PSO) and genetic algorithm (GA), to enhance the accuracy of ANN. The proposed models were evaluated for accuracy and performance using statistical metrics as well as graphical analyses. The hybridization models outperformed the standalone ANN model. In addition, ANN-PSO was more accurate than ANN-GA. The ANN-PSO model demonstrated the highest accuracy during both the training and testing stages. In the testing stage, the ANN-PSO model achieved a coefficient of determination (R<sup>2</sup>) of 0.987, a root mean square error (RMSE) of 0.0003, and a performance index (PI) of 1.72. This was followed by the ANN-GA model, which recorded an R<sup>2</sup> of 0.977, an RMSE of 0.0004, and a PI of 1.60, and the ANN model with R<sup>2</sup> = 0.861, RMSE = 0.0010, and PI = 1.40. The d10 variable was also found to be the most important factor for estimating <i>k</i> by using cosine amplitude sensitivity analysis and the SHAP method based on the ANN-PSO model.</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

Prediction of permeability coefficient of soil using hybrid artificial neural network models

  • 期刊论文

  • 发表日期：

    2025年1月14日

查看全文
  • Majid M. Kharnoob,
  • Tarak Vora,
  • A. K. Dasarathy,
  • Ish Kapila,
  • Marwan Kheimi,
  • Srinivasa Rao Rapeti

摘要

The accurate estimation of the soil permeability coefficient (k) is essential for civil engineering projects. This study employed an experimental dataset and effective parameters to model soil permeability. The specific particle sizes of soil, including d10, d50, and d60, and void ratio (e), were used as input parameters for the estimation of k. The artificial neural network (ANN) method was employed, along with two widely used metaheuristic algorithms, including particle swarm optimization (PSO) and genetic algorithm (GA), to enhance the accuracy of ANN. The proposed models were evaluated for accuracy and performance using statistical metrics as well as graphical analyses. The hybridization models outperformed the standalone ANN model. In addition, ANN-PSO was more accurate than ANN-GA. The ANN-PSO model demonstrated the highest accuracy during both the training and testing stages. In the testing stage, the ANN-PSO model achieved a coefficient of determination (R2) of 0.987, a root mean square error (RMSE) of 0.0003, and a performance index (PI) of 1.72. This was followed by the ANN-GA model, which recorded an R2 of 0.977, an RMSE of 0.0004, and a PI of 1.60, and the ANN model with R2 = 0.861, RMSE = 0.0010, and PI = 1.40. The d10 variable was also found to be the most important factor for estimating k by using cosine amplitude sensitivity analysis and the SHAP method based on the ANN-PSO model.