ISSN 1000-1239 CN 11-1777/TP

• 人工智能 •

### 实值优化问题的非对称负相关搜索算法

1. 1(大数据分析与应用安徽省重点实验室(中国科学技术大学) 合肥 230027);2(天津大学管理与经济学部 天津 300072) (yrunl@mail.ustc.edu.cn)
• 出版日期: 2019-08-01
• 基金资助:
国家自然科学基金项目(61672483，U1605251)；中国科学院青年创新促进会优秀会员专项(2014299)；安徽省科技创新战略与软科学研究专项(201806a02020055)

### Negatively Correlated Search with Asymmetry for Real-Parameter Optimization Problems

Yu Runlong1, Zhao Hongke2, Wang Zhong1, Ye Yuyang1, Zhang Peining1, Liu Qi1, Chen Enhong1

1. 1(Anhui Province Key Laboratory of Big Data Analysis and Application (University of Science and Technology of China), Hefei 230027);2(College of Management and Economics, Tianjin University, Tianjin 300072)
• Online: 2019-08-01

Abstract: As many real-world applications are closely related to complex real-parameter optimization problems, some metaheuristic assumptions are employed to help design search strategies and have been shown to be powerful tools. The balance between exploration (diversification) of new areas of the search space and exploitation (intensification) of good solutions accomplished by this kind of algorithms is one of the key factors for their high performance with respect to other metaheuristics. In particular, negatively correlated search (NCS) improves the search performance of parallel hill climbing by introducing negative correlation of search trends between search processes, which contributes greatly to the diversity maintenance of solutions. NCS models the search behaviors of individual search processes as probability distributions. On this basis, we further divide the search behaviors of a couple of search processes into global search behavior and local search behavior according to the size of the coverage of each search process. Then we present a new metaheuristic, namely negatively correlated search with asymmetry (NSA), which assumes that the search process with global search behavior should be away from the search process with local search behavior. Due to the asymmetry of the negative correlation between search processes, the efficiency of NSA has been greatly improved compared with NCS. The experimental results show that NSA is competitive to well-established search methods in the sense that NSA achieves the best overall performance on 20 multimodal real-parameter optimization problems.