Intelligence-Enabled Predictive Thermal Management System for Enhancing the Performance, Safety, And Longevity of Electric Vehicle Batteries
TL;DR
The findings show that AI can support proactive cooling, but deployment should include leakage checks, external validation, uncertainty estimation, and safety-oriented class balancing.
Abstract
Efficient battery thermal management is essential for safe, reliable, and durable electric-vehicle operation. This study develops an artificial-intelligence-enabled predictive thermal management system (PTMS) that combines sensor-driven data processing, machine-learning prediction, thermal-risk classification, and an interactive decision interface. The source project contains two complementary analytical stages. First, a general thermal-management dataset containing electrical load, ambient conditions, cooling variables, component temperature, and an overheating label was used to screen sequence and ensemble models. The long short-term memory model achieved a test accuracy of 99.66%, while Random Forest and Gradient Boosting each reached 100.00% and XGBoost reached 99.72%. The overheating class represented 24.9% of observations. Second, an electric-vehicle battery dataset containing battery temperature, state of charge, state of health, current, voltage, charging and discharging power, ambient temperature, vehicle speed, driving mode, and cooling mode was used to construct an EV-specific PTMS. Random-Forest regression was applied to battery-temperature and lifespan-impact prediction, and Random-Forest classification was used for cooling-adjustment and thermal-runaway-risk decisions. From the reported confusion matrices, cooling-adjustment accuracy was 81.91% with a macro-F1 of 78.45%, whereas thermal-runaway-risk accuracy was 87.23% with a macro-F1 of 66.83%. The latter result reflects strong performance for low- and medium-risk states but weak recall for the rare high-risk state, revealing the need for cost-sensitive learning and additional hazardous-event data. A Flask-based graphical interface demonstrates real-time data entry and prediction delivery. The findings show that AI can support proactive cooling, but deployment should include leakage checks, external validation, uncertainty estimation, and safety-oriented class balancing.