Katsa: Knn Ameliorated Tree-Seed Algorithm for Complex Optimization Problems

作     者：Jiang, Jianhua Wu, Jiaqi Meng, Xianqiu Qian, Lize Luo, Jinmeng Li, Keqin 

作者机构：Center for Artificial Intelligence Jilin University of Finance and Economics Changchun130117 China Jilin Province Key Laboratory of Fintech Jilin University of Finance and Economics Changchun130117 China College of Computer Science and Technology Jilin University China Institute of Energy Research Jiangxi Academy of Sciences Nanchang330096 China Department of Computer Science State University of New York New PaltzNY12561 United States 

出 版 物：《SSRN》 

年 卷 期：2023年

核心收录：

主　　题：Swarm intelligence 

摘      要：Tree-Seed Algorithm (TSA) is an outstanding algorithm for optimization problems, but it inevitably falls into the local optimum and has a low convergence speed in solving complex problems. This paper aims to address the above defects. Inspired by efficient learning from neighbors, a K-Nearest Neighbor (KNN) mechanism is adopted to enhance the tree or seed generation methods for achieving the balance between exploitation and exploration. The proposed algorithm is named the KNN Ameliorated Tree-Seed Algorithm (KATSA). First, inspired by the KNN mechanism, based on the best tree, the search space is divided into best and non-best neighbor areas. Based on this division approach, the proposed seed generation strategy has a precise heuristic, and the convergence speed can be accelerated. Second, the proposed seed generation and tree migration methods integrate the proposed dynamic regulation mechanism, which reduces the possibility of falling into a local optimum. Third, the proposed feedback mechanism can effectively balance exploration and exploitation. With the enhancement from the KNN mechanism, KATSA can converge to the global optima more effectively during the search process. The results obtained from IEEE CEC 2014 benchmark functions verify the excellent performance of the KATSA when compared with some recent variants, including STSA, EST-TSA, fb\_TSA and MTSA. In addition, GWO, PSO, BOA, BA, GA, LSHADE and RSA are also adopted for some benchmark comparative experiments. The applicability of the proposed KATSA is demonstrated by 3 real complex and constrained problems compared to TSA, fb\_TSA, LSHADE, RSA, GWO, ABC and PSO. The experimental results show that the proposed KATSA can obtain stable and best results on these complex problems. © 2023, The Authors. All rights reserved.