Data-Driven Discrete Global Optimization Using an Information Mining Strategy for Costly Black-Box Problems

作     者：Wang, Shengfa Dong, Huachao Wang, Peng Zhu, Haijia Wen, Zhiwen Xiang, Junyu 

作者机构：School of Marine Science and Technology Northwestern Polytechnical University Xi’an710068 China University of Victoria Canada Xi’an Precision Machinery Research Institute Xi’an710077 China 

出 版 物：《SSRN》 

年 卷 期：2024年

核心收录：

主　　题：Optimization algorithms 

摘      要：Surrogate-assisted optimization algorithms are frequently applied to handle costly black-box optimization problems. However, most of the research is on continuous problems and there are fewer studies on solving discrete problems. In addition, there is underutilization of information from surrogate models and discrete data when solving discrete problems. To address these issues, this paper proposes a data-driven discrete global optimization algorithm (DDGO). The algorithm uses a two-stage optimization framework to incorporate efficient continuous optimization modules and fully exploit the information of the discrete problem itself and the surrogate model. In the first stage, a multiple starting points selection strategy is used to obtain valuable starting points for multi-start optimization. In the second stage, a discrete neighborhoods selection strategy achieves the selection from candidate continuous solutions to optimal discrete solutions. The DDGO algorithm is tested on 20 representative benchmark cases and can find better solutions compared to other 7 algorithms. Additionally, the algorithm is employed in the structural optimization design of a blended-wing-body underwater glider, which proves that the DDGO algorithm has a good capability for the solution of discrete black-box problems. © 2024, The Authors. All rights reserved.