Thermal conductivity of liquid siloxanes: accurate intelligent modeling and robust data assessment

Résumé fourni par la source

Selecting an appropriate working fluid with specific characteristics is critical for designing and operating an organic Rankine cycle (ORC) system, since it greatly influences the system’s thermal efficiency and heat transfer performance. This study creates advanced data-driven predictive models by applying multiple machine learning methods, including Random Forest (RF), Decision Tree (DT), AdaBoost (AB), Ensemble Learning (EL), K-nearest Neighbors (KNN), Multilayer Perceptron Artificial Neural Network (MLP-ANN), and Convolutional Neural Network (CNN), to estimate thermal conductivity of liquid siloxanes using inputs of molar mass, temperature, boiling point, and pressure. The models were trained using a comprehensive experimental dataset collected from previous studies. Statistical evaluation revealed that the Ensemble Learning (EL) model achieved the highest predictive accuracy with a determination coefficient (R2) of 0.98, a mean square error (MSE) of 3.2 × 10−6, and an average absolute relative error (AARE %) of 1.7% on the testing dataset, outperforming other algorithms. The analysis also indicated that pressure, molar mass, and boiling point positively influence thermal conductivity, while temperature exhibits a negative correlation. The developed EL model provides a reliable and user-friendly computational tool for estimating the thermal conductivity of liquid siloxanes, eliminating the need for time-consuming laboratory experiments.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.