The potential of machine learning to predict melting response time of phase change materials in triplex-tube latent thermal energy storage systems

Accurate prediction of the melting response time is vital for optimizing thermal energy storage systems, which play a key role in addressing the temporal mismatch between thermal energy demand and supply in the built environment. This study aims to quantitatively predict the melting response time of a novel triplex-tube thermal energy storage system incorporating phase change materials and Y-shaped fins to enhance heat transfer. A numerical model based on the enthalpy-porosity method was developed to simulate the melting process, resulting in a dataset comprising 60 cases with melting response times ranging from 15 to 45 min under varying design and operational conditions. The key parameters investigated include fin angle (10°–30°), fin width (5–15 mm), and heat transfer fluid temperature (60 °C–80 °C). Prior to model development, variable independence was validated to ensure robust predictions. Four machine learning algorithms—polynomial regression, support vector regression, random forest regression, and extreme gradient boosting (XGBoost)—were employed, with hyperparameter optimization performed using a Bayesian approach. The XGBoost model demonstrated superior predictive capability, achieving an accuracy of 92 %. Feature importance analysis revealed that fin width and heat transfer fluid temperature were the dominant factors, contributing 51 % and 47 % to the prediction variance, respectively, whereas fin angle had a marginal influence of 2 %. This work provides a novel application of machine learning techniques to the design and optimization of thermal energy storage systems, offering valuable insights into improving their melting performance and operational efficiency. • Y-shaped fins to enhance phase change material charging performance in latent thermal energy storage systems. • Predicting melting response time using four machine learning methods. • Evaluation of model performance using mean square error and coefficient of determination. • Conducted Feature importance evaluation to guide fin improvements.