Combining Reduced-Order Model With Data-Driven Model for Parameter Estimation of Lithium-Ion Battery

The parameters of a lithium-ion battery are important to construct an effective battery management system. Parameter estimation assisted by the pseudo-two-dimensional (P2D) model is much more cost-effective than direct measurement methods. However, this is a nontrivial task, because the simulation of the P2D model is time-consuming. Alternatively, surrogate models such as reduced-order/data-driven models are often used to accelerate the parameter estimation process. Each category of surrogate models has its own strengths and weaknesses. Traditionally, reduced-order models run faster than data-driven models, while data-driven models are more accurate than reduced-order models. To leverage the complementary advantages of these two kinds of surrogate models, we make an interesting attempt to combine them compactly, thus proposing a two-phase surrogate model-assisted parameter estimation algorithm (TPSMA-PEAL). In the first phase, a fast reduced-order model is designed for parameter prescreening. In the second phase, a high-fidelity data-driven model is developed for fine estimation. In TPSMA-PEAL, except the time-consuming simulation, the other two challenges (i.e., the overfitting problem and the low observability of some parameters) are also considered from the perspective of optimization. Note that TPSMA-PEAL takes advantage of differential evolution and parameter sensitivity analysis to address them. Simulations and experiments verify that TPSMA-PEAL is efficient and accurate.

Automated summary

The parameters of a lithium-ion battery are crucial for an effective battery management system. Parameter estimation using the pseudo-two-dimensional (P2D) model is more cost-effective than direct measurement methods, but the simulation of the P2D model is time-consuming. To overcome this, a two-phase surrogate model-assisted parameter estimation algorithm (TPSMA-PEAL) is proposed, combining the advantages of reduced-order and data-driven models. TPSMA-PEAL addresses challenges such as overfitting and low observability using differential evolution and parameter sensitivity analysis. Simulations and experiments demonstrate the efficiency and accuracy of TPSMA-PEAL.