This study develops a novel deep learning (DL)-coupled multiobjective optimization framework for large-scale, high-resolution, and time-efficient low impact development (LID) planning in decentralized drainage systems. A DL surrogate model is used to replace the Stormwater Management Model (SWMM) to predict the outfall peak flow in varied LID scenarios efficiently and accurately, with statistically quantified uncertainties. An innovative K-means algorithm is developed for postoptimization analysis, enabling prioritization of customized preferences. This framework is validated through a real-world case study. By combining high-dimensional neural networks with Nondominated Sorting Genetic Algorithm II (NSGA-II), Pareto-optimal LID solutions are generated under 5, 10, and 20-yr rainfalls to minimize peak flow and investment cost. The clustering results of Pareto fronts facilitate decision prioritization considering stakeholders’ preferences. The optimal solutions are validated to mitigate peak flows cost-effectively. In the climate change scenario, the mitigation effect of selected LID solutions is found to be more significant. This framework supports decision-makers to achieve hydraulic and socioeconomic goals efficiently for sustainable and resilient urban drainage management.
Chen et al. (Tue,) studied this question.