Electrochemical Models of Multi-Unit Lithium Batteries Under Thermal Gradient

Temperature variations between multiple units of a lithium-ion battery affect electrochemical processes during its operation. Resolving each unit’s behavior with a physics-based model is computationally prohibitive, and conventional single-unit approximations fail to capture the battery performance, especially at high charge/discharge rates and/or low temperatures. Our surrogate model preserves multi-unit fidelity, while drastically reducing computational cost, and enables efficient real-time predictions of a battery’s state of charge under dynamic conditions. Comprising two convolutional neural networks and one multilayer perceptron, it emulates a standard pseudo-two-dimensional electrochemical model of an individual unit to sequentially advance state variables (lithium concentrations and electric potentials within active material and electrolyte) within each unit. Local temperature and electric current serve as inputs to the surrogate model of each unit, and inter-unit currents are computed via particle swarm optimization. Heat generation predicted by the surrogate is coupled with a one-dimensional thermal model to capture electrochemical–thermal interactions. Our simulations of commercial batteries NMC 21700 and 4680 show that the single-unit approximation overestimates effective discharge capacity, whereas the multi-unit model accurately captures the associated energy loss.