Keywords
multi-objective optimization, solution diversity, decision space, model calibration, parameter uncertainty
Start Date
26-6-2018 2:20 PM
End Date
26-6-2018 2:40 PM
Abstract
Modern multi-objective optimization algorithms should be able to converge to an efficient and diverse set of tradeoff solutions in each and every optimization trial. Maintaining the solution diversity in the objective space has been explored extensively in the literature. The vast majority of multi-objective optimization algorithms use a measure of solution diversity in the objective space, such as the niching distance, crowding distance, or hypervolume contribution, to guide their search. Nonetheless, minor attention has been paid to the solution diversity in the decision space. In this study, a modified version of Pareto Archived-Dynamically Dimensioned Search with the density-based spatial clustering approach in the decision space is applied to calibrate a multi-objective hydrologic model calibration problem. Results show that each optimization trial of the modified algorithm can identify high quality solutions that are noticeably distant from each other in the decision space. Such solutions can be used to estimate the model parameter uncertainty.
Diverse solutions of environmental and water resources engineering problems provide the decision makers with meaningfully distinct options/policies to operate the system and/or to estimate the model parameter uncertainty. Each trial of the optimization algorithm identifies distinct clusters of solutions in the decision space and performs an independent optimization in each cluster. High quality solutions from these clusters are used to estimate the model parameter uncertainty.
Enhancing Solution Diversity in Multi-Objective Optimization: Application to Model Calibration
Modern multi-objective optimization algorithms should be able to converge to an efficient and diverse set of tradeoff solutions in each and every optimization trial. Maintaining the solution diversity in the objective space has been explored extensively in the literature. The vast majority of multi-objective optimization algorithms use a measure of solution diversity in the objective space, such as the niching distance, crowding distance, or hypervolume contribution, to guide their search. Nonetheless, minor attention has been paid to the solution diversity in the decision space. In this study, a modified version of Pareto Archived-Dynamically Dimensioned Search with the density-based spatial clustering approach in the decision space is applied to calibrate a multi-objective hydrologic model calibration problem. Results show that each optimization trial of the modified algorithm can identify high quality solutions that are noticeably distant from each other in the decision space. Such solutions can be used to estimate the model parameter uncertainty.
Diverse solutions of environmental and water resources engineering problems provide the decision makers with meaningfully distinct options/policies to operate the system and/or to estimate the model parameter uncertainty. Each trial of the optimization algorithm identifies distinct clusters of solutions in the decision space and performs an independent optimization in each cluster. High quality solutions from these clusters are used to estimate the model parameter uncertainty.
Stream and Session
F3: Modelling and Decision Making Under Uncertainty