Scholarlink

<p>This study presents a comprehensive data-driven approach for predicting and optimizing CO<sub>2</sub> reduction efficiency in nano-modified bitumen systems by integrating machine learning, dimensionality reduction, and optimization technique. Using data sourced from experimental evaluations, key features influencing CO<sub>2</sub> reduction, including additive dosage, bitumen dosage, and gas mixture temperature, are analyzed and transformed using Principal Component Analysis (PCA) to capture the most critical information while reducing dimensionality. Among the evaluated models-Linear Regression, Random Forest, XGBoost, LightGBM Gradient Boosting Machine (GBM), Expected Mean Squared Error (EMSE) and Support Vector Regression-Linear Regression outperforms the others, with the lowest cross-validated RMSE, demonstrating that linear relationships dominate the underlying structure of the dataset. Feature Importance analysis further identified principal components, which are influencing CO<sub>2</sub> reduction the most. The hybrid optimization framework, combining Genetic Algorithm (GA) for global exploration and Particle Swarm Optimization (PSO) for fine-tuning, effectively minimize prediction errors and provide optimal feature scaling parameters. The optimization landscape visualization reveals stable regions of minimal RMSE, aiding in robust decision-making for material formulation. This integrated methodology not only improves model accuracy and robustness but also supports sustainable engineering by guiding the design of environmentally efficient bitumen systems. The findings offer practical implications for the construction industry in reducing CO<sub>2</sub> emissions and developing eco-friendly materials. </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

AI-Driven Optimization of Nano-modified Bitumen: CO2 reduction Efficiency Through Machine Learning and Optimization Framework

  • 期刊论文

  • 发表日期：

    2025年7月15日

查看全文
  • Priyam Nath Bhowmik,
  • Kezia Saini,
  • Pradyut Anand

摘要

This study presents a comprehensive data-driven approach for predicting and optimizing CO2 reduction efficiency in nano-modified bitumen systems by integrating machine learning, dimensionality reduction, and optimization technique. Using data sourced from experimental evaluations, key features influencing CO2 reduction, including additive dosage, bitumen dosage, and gas mixture temperature, are analyzed and transformed using Principal Component Analysis (PCA) to capture the most critical information while reducing dimensionality. Among the evaluated models-Linear Regression, Random Forest, XGBoost, LightGBM Gradient Boosting Machine (GBM), Expected Mean Squared Error (EMSE) and Support Vector Regression-Linear Regression outperforms the others, with the lowest cross-validated RMSE, demonstrating that linear relationships dominate the underlying structure of the dataset. Feature Importance analysis further identified principal components, which are influencing CO2 reduction the most. The hybrid optimization framework, combining Genetic Algorithm (GA) for global exploration and Particle Swarm Optimization (PSO) for fine-tuning, effectively minimize prediction errors and provide optimal feature scaling parameters. The optimization landscape visualization reveals stable regions of minimal RMSE, aiding in robust decision-making for material formulation. This integrated methodology not only improves model accuracy and robustness but also supports sustainable engineering by guiding the design of environmentally efficient bitumen systems. The findings offer practical implications for the construction industry in reducing CO2 emissions and developing eco-friendly materials.