基于BIC-KMeans和SWMM的城市雨洪快速模拟方法

Urban rainstorm flood rapid simulation method based on BIC-KMeans and SWMM

针对城市雨洪模型率定过程过于烦琐复杂的问题,耦合贝叶斯信息准则(BIC)和K均值聚类机器学习算法(KMeans)构建了BIC-KMeans算法。根据样本参数在不同城市功能区的分布规律,结合暴雨洪水管理模型(SWMM)提出了城市雨洪快速模拟方法。分别在郑州大学主校区、郑州市金水区南部、郑州市中心城区3个集水区选取了6、3、4场历史观测洪水事件进行了验证。结果表明:提出的城市雨洪快速模拟方法适用性较好;洼蓄量、地表曼宁系数、入渗率和衰减系数等不确定性参数在不同城市功能区的取值由小到大依次为工商业区、居民区、公共用地区;3个集水区雨洪模拟洪水流量相对误差均小于12%、纳什效率系数均大于0.75、决定系数均大于0.80,各项指标均优于传统调参法;SWMM模拟精度会随着集水区空间尺度增大而变小,峰现时间误差在较小尺度集水区雨洪模拟中是分钟级,在较大尺度为小时级。

In response to the problem of overly cumbersome and complex calibration process of urban rainstorm flood models,a BIC-KMeans algorithm was constructed by coupling Bayesian information criterion(BIC)and K-means clustering machine learning algorithm(KMeans).According to the distribution law of sample parameters in different urban functional areas,combined with the storm water management model(SWMM),an urban rainstorm flood rapid simulation method was proposed.Six,three,and four historical observation flood events were selected for verification in the main campus of Zhengzhou University,the southern part of Jinshui District in Zhengzhou City,and the central urban area of Zhengzhou.The results show that the proppsed urban rainstorm flood rapid simulation method has good applicability.The values of uncertainty parameters such as depression storage capacity,surface Manning coefficient,infiltration rate,and attenuation coefficient vary from small to large in different urban functional areas,with the order of industrial and commercial areas,residential areas,and public use areas.The relative error of simulated flood discharge in the three catchment areas is less than 12%,the Nash efficiency coefficient is greater than 0.75,and the determination coefficient is greater than 0.80.All indicators are better than traditional parameter adjustment methods.The simulation accuracy of SWMM will decrease with the increase of the spatial scale of the catchment area.The peak time error is on the minute level in small-scale rainwater and flood simulation,and on the hour level in larger scale.