Machine Learning Driven Predictive Thermal Management of Liquid Cooled EV Battery Packs under Dynamic Drive Cycles

Abstract

Electric vehicle (EV) battery packs undergo high rates of temperature increase due to aggressive dynamic drive cycles like the US06, WLTP and FTP-75 battery pack, which cause thermal non-uniformity, enhanced degradation and lower safety margins. Traditional thermal management techniques that mainly include PI and FLC-based liquid-cooling controllers have a negative temperament of slow transient response, high overshoot, and inefficient pump operation, leading to the additional auxiliary power consumption of approximately 6-12%. In order to overcome these shortcomings, this paper suggests the ML-Optimized Liquid-Cooling Thermal Management Strategy, which incorporates the use of the Random Forest Regression, Support Vector Regression (SVR), and the lightweight Artificial Neural Network (ANN) to generate heat in real-time and to control the coolant flow. Using the predictive machine-learning module, the key operating variables including current, SOC, ambient temperature, and drive-cycle dynamics are used to predict cell temperature increase and consequently, coolant mass-flow rate is adjusted. The outcomes of simulation show that there is a great improvement in comparison with the traditional approaches. The ML-based controller minimizes the maximum cell temperature by 4.2°C, temperature gradient across the module by 38 and the pump power consumption by 11.5%. The prediction model has a high level of accuracy, with MAE = 0.28°C and RMSE = 0.41°C, allowing responsive and energy efficient cooling. The suggested plan improves thermal homogeneity, increases battery pack duration, and offers a scalable architecture of the next-generation EV thermal managing systems.