Uncertainty-Aware Quantile-GBRT Modeling and Robust Optimization of a Hybrid Air–Liquid Battery Thermal Management System

In recent years, with the advancement of technology and industry, electric vehicles have become widely used. These vehicles have evolved with the goals of environmental sustainability and energy efficiency, increasing demand for lithium-ion battery technologies that offer high energy density and long life. In this study, the maximum temperature and temperature increase of a lithium-ion battery module in a 4-series 3-parallel configuration were analyzed using experimental air and hybrid (air+water) battery thermal management system (BTMS) data with an artificial intelligence-based uncertainty modeling approach. Going beyond traditional deterministic regressions, quantile-based GBRT models were applied, and q10–q50–q90 probability bands were created for each experimental configuration. Thus, the system's temperature behavior was evaluated not only through average predictions but also with uncertainty intervals. Robust Pareto optimization conducted under the condition q90(Tmax) ≤ 30°C revealed safe design points that minimize both maximum temperature and temperature difference. The Pareto analysis determined the optimal region to be around q90(Tmax) ≈ 28.8°C and q90(ΔT) ≈ 2.1°C; even in this case, the temperature difference remained below 2.5°C. Explainable artificial intelligence tools such as SHAP and PDP analyses showed that air and refrigerant flow rates play a dominant role in the temperature response, and that hybrid cooling provides a significant improvement in both Tmax and ΔT values compared to air cooling. In conclusion, the study presents an uncertainty-aware and data-driven BTMS design framework that integrates quantile modeling, robust optimization, and explainable artificial intelligence approaches, contributing to the development of safe and efficient electric vehicle battery systems.