Surrogate-based Global Sequential Sampling Algorithm

To improve research efficiency of engineering problems, Surrogate model has gained its popularity in replacing real engineering model. This paper proposes a kind of Global Sequential Sampling Algorithm (GSSA) based on surrogate model. With the process of iteration, GSSA can sample both in unexplored region and large-error region, then iteratively update the samples. OLHS is used as initial sampling method. Crossover operator which is employed in Genetic Algorithm (GA) is adopted to generate candidate sample assembly each iteration; Candidate sample with maximum weight product of cross validation error and minimum distance from existing samples will be chosen as newly added sample. At last a global surrogate model is built with all samples. GSSA is compared to MSE approach, CV-Voronoi Algorithm, and OLHS method on several test functions and results validate its effectiveness.