Abstract
Purpose
This paper aims to quantify the dependence relationship of bat algorithm’s (BA) behaviour on the factors that could possibly affect the outputs, and rank the importance of the various uncertain factors thus suggesting research priorities.
Design/methodology/approach
This paper conducts a sensitivity analysis based on variance decomposition of factors in both of original and improved BA. The data sets for sensitivity analysis are generated by optimal Latin hyper sampling in the design of experiment. The optimal factor sets are screened by stochastic error bar measures for the effective and robust implementation of BA.
Findings
The paper reveals the inner dependent relationship between factors and output in both of original and improved BA. It figures out the weakness in original BA and improves that. It suggests that uncertainty brought about by factors are mainly caused by the interaction effect and all the higher-order term in sensitivity indices for both of original and improved BA. It ranks the main effect and the total effect of factors and screens out some optimal factor sets for BA.
Originality/value
This paper quantifies the dependence relationship of BA’s behaviour on the factors that could affect outputs using sensitivity analysis based on variance decomposition.
Keywords
Acknowledgements
We want to express our appreciation to Prof Dr Xin-She Yang (Middlesex University, London) for providing us with the valuable materials on metaheuristic algorithm. Meanwhile, we are grateful to the anonymous reviewers who considerably helped in improving the manuscript.
Citation
Yin, S. and Zhu, M. (2019), "Sensitivity analysis based on variance decomposition for factors in bat algorithm", Engineering Computations, Vol. 36 No. 5, pp. 1608-1625. https://doi.org/10.1108/EC-09-2018-0402
Publisher
:Emerald Publishing Limited
Copyright © 2019, Emerald Publishing Limited