Machine learning-enhanced PID control optimization for battery thermal management: Integrating FiPy simulation with XGBoost surrogate modeling
Journal
International Communications in Heat and Mass Transfer
Journal Volume
178
Journal Issue
P2
Start Page
111678
ISSN
07351933
Date Issued
2026-09
Author(s)
Abstract
Low-temperature operation significantly degrades lithium-ion battery performance, making effective preheating control important for electric vehicle applications. This study develops a simulation-surrogate framework to investigate how proportional–integral–derivative (PID) gains influence battery preheating behavior. A two-dimensional finite-volume thermal model is implemented in FiPy for a simplified module-scale equivalent domain, where external film heating is represented as an equivalent volumetric heat input and heat dissipation is treated using an equivalent volumetric heat-loss formulation. A logarithmic parameter scan covering 12,500 combinations of Kp, Ki, and Kd is conducted to evaluate reaching time, cumulative heat input, cumulative heat loss, and success/failure behavior. XGBoost surrogate models are then developed as rapid interpolation tools within the sampled PID design space. Repeated random-split validation over 500 independent 80/20 partitions shows stable regression performance for Qin and Tfinal, with mean R2 > 0.9999 and mean MAPE below 0.07%. Ternary diagrams further reveal the trade-off between fast heating and heat loss, allowing favorable FAST & LOW-LOSS regions to be identified. The proposed framework provides an interpretable basis for PID-oriented battery preheating analysis under defined simulation conditions.
Subjects
Lithium-ion battery
Low-temperature preheating
PID control
Surrogate model
Ternary diagram
XGBoost
Publisher
Elsevier Ltd
Type
journal article