Physics-Informed Neural Network-Enhanced Model Predictive Temperature Balancing Control for Li-Ion Battery Modules

Nonuniform temperatures in lithium-ion battery modules, caused by manufacturing inconsistencies, vibrations, and unequal line resistances, lead to uneven current distribution and accelerated degradation of the battery. Existing thermal management methods face challenges in achieving real-time cell-level balancing due to limited intercell modeling, high computational cost, and lack of closed-loop control. This article proposes a model predictive temperature balancing control (MPTBC) strategy based on a scalable 2-D thermal network model (TNM) that captures intercell thermal coupling and enables real-time prediction with reduced computational cost. A physics-informed neural network (PINN) models the nonlinear internal resistance, with Bayesian optimization (BO) used to efficiently identify optimal parameters. The MPTBC is implemented on a four-module, high-power-density, single-input multioutput (SIMO) switched-capacitor (SC) converter. Experiments validate the TNM accuracy and demonstrate that MPTBC effectively minimizes cell-to-cell temperature differences.