由于城市化的快速发展和气候变化的加剧，城市内涝问题日益凸显，给城市居民的生命财产安全带来了巨大威胁。为缓解城市内涝问题，推进海绵城市和智慧城市的建设，需要建立准确、高效的城市雨洪模型。本研究以摩尔邻域为基础，考虑地表产流、地表漫流、与雨水管网的交互，通过矩阵化计算参数的方式，利用python语言实现了城市雨洪模型的构建，并使用GPU (Graphic Processing Unit) 进行加速。在此基础上，将开发模型的漫流模块、地下管网模块以及地表对产流过程的影响与常用的MIKE 21模型进行了对比分析。最后，应用该模型在北京通州海绵试点区的建成区进行了内涝风险评估和LID (Low Impact Development) 措施控制效果评估。研究得到了以下结论：（1）研究基于元胞自动机开发了涵盖地表产流、地下管网、地表漫流、LID设施的城市雨洪模拟与分析模型，并通过对比成熟商业模型证明了模型的精度和计算速度，开发模型使用的输入文件采用了与GIS (Geographic Information System) 相同的格式，便于数据前处理和后处理。（2）研究使用模型对北京通州海绵试点区建成区进行了内涝风险评估，模拟了降雨重现期从10年到1000年8个不同情景下研究区的最大水深和积水范围，综合考虑积水时间、积水深度、危险性划分了内涝风险分区。重现期为200年时，区域内最大积水深度开始超过0.40 m并出现I级红色风险区。重现期为1000年时，超过15 cm的积水面积为41.42 hm2，I级红色风险区的面积为0.42 hm2。（3）对绿色屋顶、下沉式绿地、透水铺装三种常用的LID控制措施的评估显示，研究区内，对于中小降雨事件，透水铺装具有最明显的控制效果。降雨重现期为1年、2年、5年时，15%的透水铺装削减15 cm以上的积水范围的比例均超过90%。在不同的LID源头设施布局方式中，能阻断水流沿高程流动的布局方式和随机分散式布局具有更好的效果。
Due to the rapid urbanization and climate change, the problem of urban flooding has become a severe problem, threatening the safety of life and property of urban residents. In order to alleviate urban flooding and promote the construction of sponge cities and smart cities, it is of great significance to establish an accurate, efficient urban stormwater model to make urban stormwater management more "smart". Based on the Moore neighborhood, this study considers surface runoff, surface overflow, interaction with the sewer network to build a stormwater model. The construction of the model uses matrix calculations and is coded using python language. GPU (Graphic Processing Unit) is introduced to accelerate the model computation. Then, the model is compared with MIKE 21 model in a benchmark case to analyse the surface overflow module. The two simulation methods of the underground pipe network module of the model are compared, and the influence of the surface on the runoff process was analyzed. Finally, the model is used to evaluate the flooding risk and the control effect of LID (Low Impact Development) measures in the built-up area of the Tongzhou sponge pilot region in Beijing. The control effect of green roof, sunken greenbelt and permeable pavement are compared. The main results are as follows: (1) Based on cellular automata, the research developed an urban stormwater simulation and analysis model covering surface runoff, underground pipe network, surface overflow, and LID measures, and proved the accuracy and speed of the model by comparing with mature commercial models. The input files are in the same format as GIS (Geographic Information System) for easy data pre- and post-processing.(2) The research used the model to assess the flooding risk in the built-up area of the Tongzhou sponge pilot area in Beijing, and simulated the maximum water depth and the range of flooding in the study area under 8 different scenarios with a rainfall return period from 10 years to 1000 years. Flooding time, water depth, and danger are combined to divide the flooding risk zone. When the return period comes to 200 years, the maximum ponding depth in the area begins to exceed 0.40 m and red flooding risk area appears. When the return period comes to 1000 years, the area where water depth exceeding 15 cm is 41.42 hm2, and the area of the red flooding risk area is 0.42 hm2.(3) The evaluation of three commonly used LID control measures, i.e., green roof, sunken greenbelt and permeable pavement shows that in the study area, permeable pavement has the best control effect. For light to moderate rainfall events, i.e., return periods of 1 year, 2 years and 5 years, change 15% pavement into permeable pavement can reduce area where water depth exceeding 15 cm to less than 10%. Among different LID layouts, the layout that can block the flow of water along the elevation and the random distributed layout have better effects.
