TY - JOUR
T1 - A multi-layered gravitational search algorithm for function optimization and real-world problems
AU - Wang, Yirui
AU - Gao, Shangce
AU - Zhou, Mengchu
AU - Yu, Yang
N1 - Funding Information:
This research was partially supported by National Natural Science Foundation of China (61872271, 61673403, 61873105, 11972115), the Fundamental Research Funds for the Central Universities (22120190208), and JSPS KAKENHI (JP17K12751). Recommended by Associate Editor Long Chen.
Publisher Copyright:
© 2014 Chinese Association of Automation.
PY - 2021/1
Y1 - 2021/1
N2 - A gravitational search algorithm GSA uses gravitational force among individuals to evolve population. Though GSA is an effective population-based algorithm, it exhibits low search performance and premature convergence. To ameliorate these issues, this work proposes a multi-layered GSA called MLGSA. Inspired by the two-layered structure of GSA, four layers consisting of population, iteration-best, personal-best and global-best layers are constructed. Hierarchical interactions among four layers are dynamically implemented in different search stages to greatly improve both exploration and exploitation abilities of population. Performance comparison between MLGSA and nine existing GSA variants on twenty-nine CEC2017 test functions with low, medium and high dimensions demonstrates that MLGSA is the most competitive one. It is also compared with four particle swarm optimization variants to verify its excellent performance. Moreover, the analysis of hierarchical interactions is discussed to illustrate the influence of a complete hierarchy on its performance. The relationship between its population diversity and fitness diversity is analyzed to clarify its search performance. Its computational complexity is given to show its efficiency. Finally, it is applied to twenty-two CEC2011 real-world optimization problems to show its practicality.
AB - A gravitational search algorithm GSA uses gravitational force among individuals to evolve population. Though GSA is an effective population-based algorithm, it exhibits low search performance and premature convergence. To ameliorate these issues, this work proposes a multi-layered GSA called MLGSA. Inspired by the two-layered structure of GSA, four layers consisting of population, iteration-best, personal-best and global-best layers are constructed. Hierarchical interactions among four layers are dynamically implemented in different search stages to greatly improve both exploration and exploitation abilities of population. Performance comparison between MLGSA and nine existing GSA variants on twenty-nine CEC2017 test functions with low, medium and high dimensions demonstrates that MLGSA is the most competitive one. It is also compared with four particle swarm optimization variants to verify its excellent performance. Moreover, the analysis of hierarchical interactions is discussed to illustrate the influence of a complete hierarchy on its performance. The relationship between its population diversity and fitness diversity is analyzed to clarify its search performance. Its computational complexity is given to show its efficiency. Finally, it is applied to twenty-two CEC2011 real-world optimization problems to show its practicality.
UR - http://www.scopus.com/inward/record.url?scp=85097166815&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85097166815&partnerID=8YFLogxK
U2 - 10.1109/JAS.2020.1003462
DO - 10.1109/JAS.2020.1003462
M3 - Article
AN - SCOPUS:85097166815
SN - 2329-9266
VL - 8
SP - 94
EP - 109
JO - IEEE/CAA Journal of Automatica Sinica
JF - IEEE/CAA Journal of Automatica Sinica
IS - 1
M1 - 9272701
ER -