Accurate modeling of nano-enhanced polyethylene glycol thermal conductivity using soft computing methods: application to thermal energy storage

Polyethylene glycol (PEG) is considered a renowned polymer with a semi-crystalline composition that, as a phase change material, is ideal for use in applications of heat energy storage. Recent research suggests that PEG's thermal conductivity can be significantly improved by incorporating nanoparticles. This study focuses on developing various robust machine learning methods like decision trees, adaptive boosting, ensemble learning, K-nearest neighbors, multilayer perceptron artificial neural networks, convolutional neural networks, and extra trees to accurately assess nano-enhanced PEG thermal conductivity based on PEG molecular weight, temperature, type of nanomaterial, and its concentration. The leverage technique is employed to identify potential outlier data within the collected dataset. Additionally, a sensitivity assessment will be performed to examine the relative impacts of each input parameter on the thermal conductivity. The K-fold cross-validation technique is used in every algorithm to mitigate the overfitting problem during model training. The results indicate that the extra trees (with R2 = 0.9368997, MSE = 0.0003628, AARE% = 3.7075695) and decision tree (with R2 = 0.9374603, MSE = 0.0003596, AARE% = 3.5759954) models are the most accurate in predicting the thermal conductivity of nano-enhanced PEG. These models achieve the highest coefficient of determination (R2) and the lowest error metrics (MSE and AARE%), highlighting their exceptional capacity to recognize intricate patterns and provide accurate forecasts, particularly for forecasting thermal conductivity. Also, it is implied that temperature, molecular weight of PEG, and nanoparticle concentration all tend to increase the thermal conductivity, with nanoparticle concentration being the most effective factor.