A Surrogate-Based Optimization Design and Uncertainty Analysis for Urban Flood Mitigation

Abstract

This study proposes a surrogate-based optimization framework (SBO) to help analyze the tradeoff between flood damages and investment while considering uncertainty originating from surrogates. The surrogate models were constructed based on the relationship between drainage specifications and simulated flood information and used to replace the numerical model in optimization, thereby reducing the computational burden. The bootstrapping approach was employed to quantify the uncertainty originating from surrogate models, which were incorporated into the NSGA-II optimization algorithm to seek the interval of optimal solutions. Through a case study, the results showed that the uncertainties caused by surrogate models have a significant influence on the reliability of the optimal solutions, but require lower computational efforts. Moreover, the local design conditions (i.e., various designed rainfalls) had an impact on the design and performance of the detention tanks. The proposed framework will facilitate cost-effective planning of flood mitigation systems with an awareness of associated uncertainty in order to resolve tradeoffs, particularly for large-scale problems.