Scholarlink

<p>Accurate prediction of thermophysical properties is essential for the design and optimization of modern energy and industrial systems. Among these properties, the heat capacity of liquid siloxanes plays a critical role in thermal regulation, process efficiency, and the performance of energy conversion technologies such as Organic Rankine Cycles (ORCs). Conventional heat capacity measurements require significant time, resources, and expense, making them impractical for large-scale applications. This limitation highlights the need for robust, data-driven prediction methods capable of providing accurate results with greater efficiency. In this study, a fine-tuned Gradient Boosting Decision Tree (GBDT) model was developed and optimized using four advanced algorithms: Gaussian Processes Optimization (GPO), Bayesian Probability Improvement (BPI), Evolution Strategies (ES), and Batch Bayesian Optimization (BBO). The model utilized a dataset of 427 empirical measurements, divided into 90% for training and 10% for evaluation, with input features including temperature, pressure, molecular weight, and boiling point. To avoid overfitting, k-fold cross-validation was applied during training. The performance of the optimization methods was evaluated using statistical indicators like the mean squared error (MSE), coefficient of determination (R<sup>2</sup>) and average absolute relative error (AARE%), as well as computational runtime. Correlation analysis identified temperature as the most significant element (correlation coefficient = 0.76), followed by boiling point (0.56) and pressure (0.50), while molecular weight showed a negative correlation (–0.55). Among the optimization approaches, GBDT-ES demonstrated the greatest predictive precision (R<sup>2</sup> = 0.9199 on the test set), whereas GBDT-BPI exhibited superior computational efficiency (runtime = 152.8 s). Sensitivity analysis further confirmed the importance of each input variable, highlighting the robustness of the proposed framework. These findings demonstrate the value of data-driven methodologies for predicting the heat capacity of liquid siloxanes, reducing reliance on labor-intensive experiments and enabling more efficient applications in industrial and energy systems. The findings of this study can help achieve a better understanding of the thermophysical behavior of siloxanes under varying conditions and support the design of more efficient thermal management systems. In particular, these results provide practical guidance for applications such as Organic Rankine Cycles (ORCs), heat exchangers, and energy storage technologies, where accurate heat capacity predictions are essential for improved performance.</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

Development of fine-tuned hybrid gradient boosting decision tree models to reliably predict heat capacity of liquid siloxanes

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月3日

查看全文
  • Jamal I. Al-Nabulsi,
  • Zaid Ajzan Alsalami,
  • J. Deepak,
  • M. G. M. Johar,
  • Anupama Routray,
  • A. Karthikeyan,
  • Harjot Singh Gill,
  • Amanpreet Sandhu,
  • Abdolali Yarahmadi Kandahari

摘要

Accurate prediction of thermophysical properties is essential for the design and optimization of modern energy and industrial systems. Among these properties, the heat capacity of liquid siloxanes plays a critical role in thermal regulation, process efficiency, and the performance of energy conversion technologies such as Organic Rankine Cycles (ORCs). Conventional heat capacity measurements require significant time, resources, and expense, making them impractical for large-scale applications. This limitation highlights the need for robust, data-driven prediction methods capable of providing accurate results with greater efficiency. In this study, a fine-tuned Gradient Boosting Decision Tree (GBDT) model was developed and optimized using four advanced algorithms: Gaussian Processes Optimization (GPO), Bayesian Probability Improvement (BPI), Evolution Strategies (ES), and Batch Bayesian Optimization (BBO). The model utilized a dataset of 427 empirical measurements, divided into 90% for training and 10% for evaluation, with input features including temperature, pressure, molecular weight, and boiling point. To avoid overfitting, k-fold cross-validation was applied during training. The performance of the optimization methods was evaluated using statistical indicators like the mean squared error (MSE), coefficient of determination (R2) and average absolute relative error (AARE%), as well as computational runtime. Correlation analysis identified temperature as the most significant element (correlation coefficient = 0.76), followed by boiling point (0.56) and pressure (0.50), while molecular weight showed a negative correlation (–0.55). Among the optimization approaches, GBDT-ES demonstrated the greatest predictive precision (R2 = 0.9199 on the test set), whereas GBDT-BPI exhibited superior computational efficiency (runtime = 152.8 s). Sensitivity analysis further confirmed the importance of each input variable, highlighting the robustness of the proposed framework. These findings demonstrate the value of data-driven methodologies for predicting the heat capacity of liquid siloxanes, reducing reliance on labor-intensive experiments and enabling more efficient applications in industrial and energy systems. The findings of this study can help achieve a better understanding of the thermophysical behavior of siloxanes under varying conditions and support the design of more efficient thermal management systems. In particular, these results provide practical guidance for applications such as Organic Rankine Cycles (ORCs), heat exchangers, and energy storage technologies, where accurate heat capacity predictions are essential for improved performance.