Cross-validation of numerical methods for investigating the heating process in large-volume thermal storage tanks

Next-generation district heating relies on large hot-water storage tanks, which in turn require accurate, computationally efficient full-scale modelling for resolving stratification and thermocline dynamics and supporting design, control, and long-horizon planning. This work cross-validates three numerical modelling approaches on the same 1900 m 3 vertical tank during a 10 h charge: a high-fidelity CFD, a custom finite-difference (FDM) model, and a system-level nodal (lumped-parameter) simulation. The CFD setup (2-D axisymmetric, buoyancy-dominated laminar interior) underwent five-level space–time refinement; the adopted mesh/time step delivered stable advancement and mesh-independent thermocline thickness. Against experiment, CFD reproduces profiles with MD ≤ 1.25 °C, MBE ≤2.91 %, and RMSE = 1.4–3.9 % over 0–10 h. FDM accuracy improves markedly with resolution; at 500 control volumes it matches CFD in the hot/cold layers and captures thermocline evolution within the experimental uncertainty band, while requiring hours rather than days of runtime. The system-level nodal simulation executes in seconds and suits long-horizon studies, but with ≤40 nodes in the present implementation it cannot resolve short-term thermocline sharpening. A clear tradeoff between accuracy and computational cost was observed: CFD suits short, detail-focused analyses; finite-difference models handle day-scale studies efficiently; nodal models serve seasonal planning. Most prior work on hot-water storage tanks focuses on small or medium volumes and a single modelling approach, rarely reporting uncertainty. To address this, three approaches—CFD, finite difference, and nodal models—were compared under identical conditions on the same dataset, with guidance on method selection and expected errors. • Full-scale hot-water tanks need accurate yet fast numerical modelling. • Fluid CFD numerical simulation matches measured profiles with 1.4–3.9 % error. • Finite difference (500 layers) tracks the thermocline within 0.5 m. • Nodal model runs in seconds but misses short-term stratification detail. • Findings guide method choice by trading accuracy against runtime.