Regresar

Surrogate model of liquid cooling system for lithium-ion battery using extreme gradient boosting

Abstract:

Battery thermal management system (BTMS) has been widely regarded as an effective means to mitigate the sensitivities of lithium ion batteries (LIBs) to temperature, as to uphold the safe and high-performance operation of LIBs. On the premise that the volume and cost are limited, optimizing the structure of BTMS becomes relevant. Owing to the complex flow and heat transfer processes, the objective functions of BTMS structural optimization cannot be expressed analytically, but has to be evaluated using finite element method (FEM), which is time-consuming and therefore not scalable. The well-known solution to speed up the evaluation of objective functions is to engage a surrogate model, which offers a fast but slightly inaccurate mapping from the input variable space to the response. In this article, two innovations on surrogate modeling for BTMS structural optimization are proposed. One of them is in terms of input variables of surrogate model, which takes into consideration not only the structural parameters, but also the operating conditions. The other is on the surrogate model itself, in that, extreme gradient boosting (XGBoost) is used to replace the conventional kriging and response-surface methods, which is able to reduce the maximum absolute error of the model by 50%. The errors of maximum temperature and temperature difference decrease 0.179°C and 0.255°C, respectively. Moreover, the computational time of the XGBoost-based model is just 2 ms. The method proposed in this article can be applied on a wide range of structural optimization tasks in real driving cycle of electric vehicles.