Modeling transient temperature in phase change materials using a hybrid convolutional neural network and long short-term memory approach for melting process analysis

Abstract

This study presents a hybrid deep learning framework combining Convolutional Neural Networks and Long Short-Term Memory networks to predict the transient temperature evolution in phase change materials during the melting process. Trained on over 38000 experimental temperature data points collected from 18 thermocouples under three wall boundary temperatures (45, 55, and 70 °C), the model effectively captures spatial and temporal dependencies governing the coupled conduction–convection heat transfer. The hybrid model achieved a root mean square error of 0.2275 °C and a coefficient of determination of 0.9994 on test data, confirming its high accuracy and generalization ability. Comparative validation with experimental and numerical results revealed that the model outperforms physics-based simulations by over one order of magnitude in accuracy while requiring substantially less computational effort. An ablation study confirmed the complementary roles of the convolutional and recurrent components, while sensitivity analysis demonstrated the model’s robustness against boundary and spatial perturbations, with all variations in error below 1.5 %. These findings establish the proposed data-driven approach as a reliable and computationally efficient alternative to conventional numerical methods for modeling transient heat transfer and melting dynamics in thermal energy storage applications. Copyright © 2025. Published by Elsevier Ltd.