An efficient Gaussian process-based optimization for resilience improvement of urban drainage systems considering changing climate

Abstract

Urban drainage systems (UDS) face significant challenges in managing rising stormwater loads because of changing climate conditions. Sustainable stormwater management solutions like green infrastructure together with existing UDS can be implemented optimally, but catchment-scale design is computationally difficult. To solve this problem, this study proposes and evaluates an efficient Gaussian process-based surrogate framework for optimizing green infrastructure and UDS under existing and future extreme rainfall conditions in order to minimize retrofit life cycle cost of infrastructures, building damage cost, outlet peak flow and maximize UDS's resilience. Furthermore, the computational performances of the surrogate-based optimization framework and traditional optimization algorithm (i.e., NSGA-II) were also compared. Rainfall will increase in future periods compared to baseline period, and the frequency will shift from low to high-intensity events. The surrogate-based optimal designs of sustainable stormwater infrastructures demonstrated a maximum increase of 760 % and 18.59 % in damage cost and outlet peak flow, and a decrease in UDS's resilience of -13.15 % for 50-year return periods in the far future period. Furthermore, the Gaussian process-based surrogate framework was 72.3 % faster than NSGA-II in computational performance evaluation, with maximum hypervolume improvement (i.e., convergence rate) for solving optimization problems in the first 100 iterations and limited improvement after that.