面向城市内涝消减的蓄水池尺寸优化模拟

Detention tanks size optimization for urban flood mitigation

针对传统多目标优化算法存在计算复杂且效率偏低的问题,本文提出了一种基于替代模型的多目标优化方法.即建立人工神经网络模型代替城市水文水动力模拟模型,再耦合NSGA-Ⅱ优化算法估计最优Pareto解集.选择典型城市中心集水区作为研究区,选用反向传输神经网络构建替代模型,采用平均相对误差绝对值、相关系数、均方根误差和纳什效率系数对替代模型的预测性能进行评估,结果表明替代模型具有较高的精度(误差<5%,拟合精度>90%).在此基础上,研究了不同降雨重现期下最小化工程成本和最大化灾害减少比率的蓄水池多目标优化问题.优化结果表明:每个重现期下的最优方案都能够有效地降低最大积水深度(26%～41%)、平均积水深度(13%～33%),减少积水面积(11%～39%)以及洪峰流量(5%～16%).通过对比传统优化模型和基于替代模型的优化模型的计算时间成本,结果表明后者在保证精度的同时能够节省约99%的时间.本文提出的方法能够为洪水风险管理系统的优化设计提供参考思路并提高优化效率.

Urban design of flood control and drainage optimization is important in flood risk management. To address the common problems of high computational complexity and low efficiency in traditional multiobjective optimization algorithms,we propose a multi-objective optimization method based on surrogate models.Artificial neural networks are established to replace hydrologic-hydraulic simulation models,and then NSGA-Ⅱ algorithm was applied to estimate Pareto optimal solutions.A typical city catchment was studied, back propagation (BP)neural network was used to construct surrogate models.The performance of surrogate models was then evaluated by computing various statistical indices,including Mean Relative Absolute Error (MRAE),Coefficient of Correlation (CC),Root Mean Square Error (RMSE)and Nash-Sutcliffe efficiency (NSE).All surrogate models were found to have suitable performance (error<5%,simulation accuracy >90%).Multi-objective optimization designs of detention tanks were then examined to minimize costs and maximize disaster reduction ratio under different rainfall return periods.Simulation revealed that adding optimally designed water storage tanks could effectively reduce maximum water depth (26%-41%),average water depth (13%-33%),flood area (11%-39%),and peak flow (5%-16%)under different return periods.A comparison of computational time of traditional optimization model and of surrogate-based model found that the latter could save 99% of the running time while ensuring accuracy.This will provide useful information for optimal design of flood risk management system and improve efficiency of optimization.