An improved adaptive surrogate model and application in thermal management system design

Abstract It is time-consuming to obtain the responses of real or high-precision simulation models in complex engineering problems. The surrogate model based on sample points can approximate the real model and thus greatly reduce the computational effort. The selection of sample points has a great influence on the accuracy of the surrogate model. Aiming at the problem of sample point selection in the process of establishing surrogate model, an adaptive sampling method based on distance density and local complexity is proposed. In this method, distance density is used to quantify the sparsity of new sample points, and local complexity is applied to quantify the change complexity of response values near new sample points. The high-quality of new sample point is added to improve the accuracy of the surrogate model. This method is compared with two other classical adaptive sampling methods through nine test functions. The results show that this method can make the new sample point more distributed in the key area of sample space, so that fewer sample points are used to establish a high-precision surrogate model. Finally, the effectiveness of this method is verified through an optimization of the thermal management system design for liquid-cooled cylindrical batteries.