Dynamic mesh technology based on support vector regression

Dynamic mesh generation is one of the key technologies for solving moving boundary problems. In order to improve the computation efficiency of dynamic mesh generation, the widely used dynamic mesh method based on radial basis functions is analyzed in-depth, and an efficient approach based on support vector regression is proposed to further relax the constraints of the greedy choice of points. Moreover, an adaptation scheme for this machine learning algorithm applying to mesh motion is given. According to three different grids and four types of motions in the benchmark cases, a comprehensive analysis of the fitting performance of different radial basis kernel functions is carried out. Using previous Gaussian kernel based method as the benchmark, the selected kernel functions are compared through typical dynamic grid cases, for their performance in mesh deformation quality, computational efficiency and parameter setting. The results show that based on the CP C² and IMQB kernel functions, the proposed dynamic mesh method has good deformability and computational efficiency.