Hybridizing Optimization Method and Artificial Neural Network for Urban Drainage System Design

Abstract

Urban Drainage System (UDS) design has been performed to determine the pipe layout and sizes mostly by optimization-based models. However, function evaluations of the UDS design problem are often very time-consuming because of the computation time for rainfall-runoff and network hydraulics of Strom Water Management Model (SWMM). Therefore, various surrogate models have been proposed during the last decade to overcome the limitation. This study first proposes a hybrid models that combine an optimization method and a surrogate model based on Artificial Neural Network (ANN) for UDS pipe layout and size optimization. Then, various versions of the proposed model are applied to the design problem of a real large urban drainage network: (1) Version 1 determines the optimal pipe sizes by the optimization module whereas the pipe layout is derived from the ANN module; (2) Version 2 determines the former by the ANN module, and the latter by the optimization module; and (3) Version 3 produces the optimal pipe sizes and layout solely by the ANN module. We compared the three different versions with respect to their computational efforts and times and estimation accuracy (e.g., Mean Square Error and R2). Finally, this study will provide guidelines for using the hybrid models in the UDS design.