Scholarlink

Background <p>Pervious concrete is widely recognized for its role in sustainable urban development, offering benefits such as stormwater management and groundwater recharge. Accurate prediction of its mechanical and permeability-related properties is essential for optimizing its performance in green infrastructure applications.</p> Problem <p>Despite its advantages, pervious concrete presents challenges in performance prediction due to its complex pore structure and variable strength characteristics. Conventional predictive models often fail to deliver sufficient accuracy across key properties such as compressive strength, tensile strength, flexural strength, permeability, and porosity, particularly over different curing periods.</p> Methods <p>This research introduces a novel deep learning-based predictive framework that employs an Improved Deep Belief Network (IDBN) optimized through a Self-Improved Randomized Population Optimization (SI-RPO) algorithm. The model is developed using experimentally obtained data to predict key properties of pervious concrete such as compressive strength, split tensile strength, flexural strength, permeability, and porosity at 7 and 28&#xa0;days of curing. The SI-RPO algorithm is specifically designed to fine-tune the network’s weights and biases, thereby improving convergence speed, avoiding local minima, and enhancing the model’s generalization ability. This hybrid approach ensures robust and accurate prediction of pervious concrete performance metrics, demonstrating superior effectiveness compared to conventional optimization techniques.</p> Result <p>The proposed SI-RPO optimized IDBN model outperforms conventional approaches, specifically, for 7-day compressive strength, the model achieved a minimum Mean Squared Error (MSE) of 0.021, and for 28-day strength, it yielded the lowest Mean Absolute Percentage Error (MAPE) of 0.034. These results highlight the effectiveness of the proposed method in accurately predicting pervious concrete properties and support its application in sustainable construction practices.</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 Strength and Permeability of Pervious Concrete Using an Enhanced IDBN Model

  • 期刊论文

  • 发表日期：

    2025年11月19日

查看全文
  • Somdeep Chakraborty,
  • Kaberi Majumdar,
  • Manish Pal,
  • Debashish Karmakar,
  • Pankaj Kumar Roy,
  • Dipankar Sarkar

摘要

Background

Pervious concrete is widely recognized for its role in sustainable urban development, offering benefits such as stormwater management and groundwater recharge. Accurate prediction of its mechanical and permeability-related properties is essential for optimizing its performance in green infrastructure applications.

Problem

Despite its advantages, pervious concrete presents challenges in performance prediction due to its complex pore structure and variable strength characteristics. Conventional predictive models often fail to deliver sufficient accuracy across key properties such as compressive strength, tensile strength, flexural strength, permeability, and porosity, particularly over different curing periods.

Methods

This research introduces a novel deep learning-based predictive framework that employs an Improved Deep Belief Network (IDBN) optimized through a Self-Improved Randomized Population Optimization (SI-RPO) algorithm. The model is developed using experimentally obtained data to predict key properties of pervious concrete such as compressive strength, split tensile strength, flexural strength, permeability, and porosity at 7 and 28 days of curing. The SI-RPO algorithm is specifically designed to fine-tune the network’s weights and biases, thereby improving convergence speed, avoiding local minima, and enhancing the model’s generalization ability. This hybrid approach ensures robust and accurate prediction of pervious concrete performance metrics, demonstrating superior effectiveness compared to conventional optimization techniques.

Result

The proposed SI-RPO optimized IDBN model outperforms conventional approaches, specifically, for 7-day compressive strength, the model achieved a minimum Mean Squared Error (MSE) of 0.021, and for 28-day strength, it yielded the lowest Mean Absolute Percentage Error (MAPE) of 0.034. These results highlight the effectiveness of the proposed method in accurately predicting pervious concrete properties and support its application in sustainable construction practices.