A Surrogate-Based Optimization Approach for Sustainable Drainage Design in Large Urban Areas

The design of conventional and sustainable urban drainage systems is a complex task that requires consideration of several design objectives and decision variables. Simulation-based optimization models allow exploring the decision space and identify design options that best meet the design criteria. However, existing approaches generally require simulation of the system hydraulics for each function evaluation, which leads to prohibitive computational cost when applied to large drainage networks.

In this work, a disaggregation-emulation approach is proposed which allows sequential optimization of multiple sub-catchments in an urban area without having to simulate the full system dynamics. This is achieved by using artificial neural networks (ANN) to represent the boundary condition at the interface between neighboring sub-catchments. The approach is demonstrated with an application to a many-objective optimization problem in which sustainable drainage systems are used to expand the capacity of an existing drainage network. The evaluation of the objective function using the emulation model is found to be 22 times faster than using the physically based model, resulting in a significant speed-up of the optimization process. Unlike previously proposed optimization approaches that rely on surrogate models to emulate the objective functions, the proposed approach remains physically based for the individual sub-catchments, thus reducing the chance of bias in the optimization results.