Computational Fluid Dynamics-Generated Training Datasets for Artifical Intellingence-Assisted Thermal Storage System Optimization

This dataset contains the results of an extensive computational fluid dynamics (CFD) parametric study conducted to simulate the performance of advanced cascade-packed bed latent heat storage (PBLHS) systems. Using COMSOL Multiphysics, the study modeled fluid flow and heat transfer within a porous packed bed containing three layers of composite phase change materials (CPCMs), each layer with different melting points tailored for efficient waste heat recovery (25–600 °C). The CFD model incorporated detailed physical phenomena including non-Darcian flow using the Ergun equation, local thermal nonequilibrium (LTNE) heat transfer, dynamic variations of thermophysical properties with temperature, and heat losses through tank boundaries. A Latin Hypercube Sampling (LHS) approach was used to systematically vary key design parameters, including tank dimensions, particle sizes, insulation thickness, flow rates, charging/discharging times, and PCM melting points. The resulting dataset includes 620 validated simulation cases, with each case providing both charging and discharging thermal performance metrics, including stored and recovered heat quantities, total efficiencies, and relevant fluid/thermal field outputs. This dataset was employed to train deep learning models for rapid prediction and optimization of PBLHS performance, as presented in the article "An Integrated Machine Learning and Metaheuristic Approach for Advanced Packed Bed Latent Heat Storage System Design and Optimization" (Energy, 2024, https://doi.org/10.1016/j.energy.2024.131149). The simulations were conducted under the REDTHERM project, funded by the European Commission (Grant No. 101068507), aiming to advance medium-to-high temperature thermal storage technologies based on waste-derived materials.