Non-uniform mode-pursuing sampling method based on multivariate multimodal distribution model

Abstract

A non-uniform distributing based mode-pursuing sampling method on black-box problem is proposed for the computation-intensive global optimization problem. The multivariate multimodal distribution model is established based on the expensive sample points of original optimization problem, which is controlled by the convergence principle of multiple correlation coefficient through variance of probability distribution. More design points are generated progressively around the current optimal regions and constitute the non-uniform discrete design space. Non-uniform sampling strategy changes the uniform distributing characteristic of design space, improves the utilization efficiency of design points and strengthens the distribution rationality of discrete design space. Analytical and numerical test results show that the improved method is more efficient and accurate than standard mode-pursuing sampling method and traditional algorithms, and has broad prospects for the expensive black-box problem.