Comparison of Optimization Techniques for Lithium-Ion Battery Model Parameter Estimation 2014-01-1851

Due to rising fuel prices and environmental concerns, Electric Vehicles (EVs) and Hybrid Electric Vehicles (HEVs) have been gaining market share as fuel-efficient, environmentally friendly alternatives. Lithium-ion batteries are commonly used in EV and HEV applications because of their high power and energy densities. During controls development of HEVs and EVs, hardware-in-the-loop simulations involving real-time battery models are commonly used to simulate a battery response in place of a real battery. One physics-based model which solves in real-time is the reduced-order battery model developed by Dao et al. [1], which is based on the isothermal model by Newman [2] incorporating concentrated solution theory and porous electrode theory [3].

The battery models must be accurate for effective control; however, if the battery parameters are unknown or change due to degradation, a method for estimating the battery parameters to update the model is required. A set of manufacturer recommended battery parameters were evaluated using a numerical sensitivity analysis to evaluate their identifiability. The parameters chosen to be identified were ε_p, ε_s and brugg. The optimization algorithms that were evaluated for parameter estimation were: Self-Adaptive Evolution, Efficient Global Optimization, Differential Evolution, and Simulated Annealing. These algorithms were evaluated based on how many simulation calls were required to converge to an accuracy of 1e-4. Differential Evolution was shown to have the best performance in estimating the parameters, requiring an average of 1485 simulations to converge.