Scholarlink

<p>This study introduced a new approach to predicting slope factors of safety using a feedforward backpropagation artificial neural network (<i>ANN</i>). Based on a database of 540 cases, this paper investigated the effect of significant parameters such as slope angle, soil friction angle, soil cohesion, slope height, and soil unit weight. The two-hidden-layer <i>ANN</i> architecture consists of ten neurons in each hidden layer utilizing the rectified linear unit activation function for nonlinearity and a linear activation function for the output layer. This architecture could capture complex interactions among input parameters with excellent predictive capability, as seen from <i>R</i> values of 0.95806, 0.9556, 0.94566, and 0.95566 for training, validation, testing, and overall datasets. The model is trained by implementing the backpropagation algorithm with the Adam optimizer and assessed using mean squared error as the loss function. Notably, the findings indicated that <i>ANN</i> cut computational time by about 90% relative to traditional numerical techniques. Enhanced performance of the <i>ANN</i> model not only eases slope stability analysis but also yields a powerful tool for efficient slope stability prediction under different 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

ANN Insight in Forecasting Slope Stability by Analyzing the Influence of Strength Parameters in 2D and 3D Scenarios

  • 期刊论文

  • 发表日期：

    2025年3月31日

查看全文
  • Masoud Nasiri,
  • Ehsan Amiri

摘要

This study introduced a new approach to predicting slope factors of safety using a feedforward backpropagation artificial neural network (ANN). Based on a database of 540 cases, this paper investigated the effect of significant parameters such as slope angle, soil friction angle, soil cohesion, slope height, and soil unit weight. The two-hidden-layer ANN architecture consists of ten neurons in each hidden layer utilizing the rectified linear unit activation function for nonlinearity and a linear activation function for the output layer. This architecture could capture complex interactions among input parameters with excellent predictive capability, as seen from R values of 0.95806, 0.9556, 0.94566, and 0.95566 for training, validation, testing, and overall datasets. The model is trained by implementing the backpropagation algorithm with the Adam optimizer and assessed using mean squared error as the loss function. Notably, the findings indicated that ANN cut computational time by about 90% relative to traditional numerical techniques. Enhanced performance of the ANN model not only eases slope stability analysis but also yields a powerful tool for efficient slope stability prediction under different conditions.