Quantifying dynamic sensitivity of optimization algorithm parameters to improve hydrological model calibration

•A framework for dynamically quantifying algorithm parameter impacts is developed.•Interactions among parameters have significant influence on algorithm performance.•Reflection parameter of SCE-UA can be more influential than complex number. It is widely recognized that optimization algorithm parame...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Journal of hydrology (Amsterdam) Jg. 533; S. 213 - 223
Hauptverfasser: Qi, Wei, Zhang, Chi, Fu, Guangtao, Zhou, Huicheng
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier B.V 01.02.2016
Schlagworte:
Algorithm
Algorithms
Analysis of variance
Arizona
Dynamics
hydrologic models
Interactive
Mathematical models
Optimization
SCE-UA
Search process
Sensitivity
Sensitivity analysis
TOPMODEL
variance
Variance decomposition
USA, Arizona
Arizona
TOPMODEL
Sensitivity
Variance decomposition
Algorithm
SCE-UA
Optimization
ISSN:0022-1694, 1879-2707
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:•A framework for dynamically quantifying algorithm parameter impacts is developed.•Interactions among parameters have significant influence on algorithm performance.•Reflection parameter of SCE-UA can be more influential than complex number. It is widely recognized that optimization algorithm parameters have significant impacts on algorithm performance, but quantifying the influence is very complex and difficult due to high computational demands and dynamic nature of search parameters. The overall aim of this paper is to develop a global sensitivity analysis based framework to dynamically quantify the individual and interactive influence of algorithm parameters on algorithm performance. A variance decomposition sensitivity analysis method, Analysis of Variance (ANOVA), is used for sensitivity quantification, because it is capable of handling small samples and more computationally efficient compared with other approaches. The Shuffled Complex Evolution method developed at the University of Arizona algorithm (SCE-UA) is selected as an optimization algorithm for investigation, and two criteria, i.e., convergence speed and success rate, are used to measure the performance of SCE-UA. Results show the proposed framework can effectively reveal the dynamic sensitivity of algorithm parameters in the search processes, including individual influences of parameters and their interactive impacts. Interactions between algorithm parameters have significant impacts on SCE-UA performance, which has not been reported in previous research. The proposed framework provides a means to understand the dynamics of algorithm parameter influence, and highlights the significance of considering interactive parameter influence to improve algorithm performance in the search processes.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:0022-1694
1879-2707
DOI:10.1016/j.jhydrol.2015.11.052