In-context learning for nano-PCM thermal behavior prediction in battery thermal management via Lattice Boltzmann simulation

Effective thermal management is crucial for ensuring the safety and performance of lithium ion batteries in electric vehicles. While nano-enhanced phase change materials (nano-PCMs) offer excellent thermal regulation, their effectiveness is often limited by localized heat accumulation from natural convection. Existing ML-based studies mostly focus on temperature metrics alone, neglecting thermal uniformity and relying on black-box models with limited interpretability. This study proposed a zonal nanoparticle distribution strategy and utilized high-fidelity Lattice Boltzmann Method (LBM) simulations to investigate underlying heat transfer mechanisms in nano-enhanced PCM systems. The state-of-the-art Tabular Prior-data Fitted Network (TabPFN) was then employed to accurately predict key thermal indicators and was benchmarked against widely used models such as BPNNs, XGBoost, and CatBoost. Furthermore, SHapley Additive exPlanations (SHAP) analysis was applied to interpret TabPFN outputs, revealing key regional features and providing physical insights into system performance. The results demonstrated that with optimal non-uniform nanoparticle distribution pattern, the Nusselt number increased by 12.04 % and melting time was reduced by 13.05 %. TabPFN exhibited superior prediction accuracy compared to other popular machine learning models, with error bins generally lower in magnitude and reductions in MAE and RMSE by 8–92 % and 7–90 %. SHAP analysis further visualized quantitative correlation between training inputs and target variables and their influence on convective behavior and heat retention. The proposed explainable in-context learning framework based on TabPFN and SHAP is expected to provide valuable insights for guiding the design and optimization of advanced nano-PCM battery thermal management systems.