The determination of locations and sizes for such a system is important in a drainage master plan or a storm-water management system. However, the distribution of detentions in the upstream and midstream is often more dispersed using many combinations of volume scales. This paper uses the non-dominated sorting genetic algorithm combined with the Storm Water Management Model to explore and calculate the optimal layout scheme for decentralized rainwater detention. The purpose is to find a design and planning method that can achieve the optimal balance of decentralized detention considering the aspects of flood disaster control, peak flow reduction, and investment cost. The optimal results of Pareto in applied case show that among the five most unfavourable nodes, the detentions with different layout volumes and relatively smaller size can control water logging from rainstorm. The project cost is effectively reduced and the standard of the return period of the regional rainwater system is enhanced from 2 to 20 years.