The absence or inaccessibility of high-quality Urban Underground Stormwater Networks (UUSNs) data hinders precise modeling and analysis of urban flood. This study proposed an algorithm for deriving UUSNs based on urban road networks and UUSNs design standards by using domain adversarial neural networks and complex network analysis. In data-supported region, the algorithm successfully captured 92% of the existing UUSNs. Based on the proposed algorithm, we predicted the UUSNs data in the urban area of Zhengzhou City and simulated the inundation scenario of the extraordinary Zhengzhou flood of 7/20 by incorporating derived UUSNs, the prediction accuracy of inundation depth achieved a threefold improvement compared to conventional generalized infiltration parameterization method. Our study proposed a new algorithm for generating UUSNs in areas lacking labeled data and demonstrated excellent migration capability of cross-region prediction, which facilitated an enhanced urban flooding simulation.

中文翻译：

---

一种用于生成城市地下雨水管网的新算法及其在增强城市洪水模拟中的应用

缺乏或无法访问高质量的城市地下雨水网络 （UUSN） 数据阻碍了城市洪水的精确建模和分析。本研究利用域对抗神经网络和复杂网络分析，提出了一种基于城市道路网络和 UUSNs 设计标准的 UUSN 推导算法。在数据支持区域，该算法成功捕获了 92% 的现有 UUSN。基于所提算法，预测了郑州市市区的 UUSNs 数据，并结合推导的 UUSNs 模拟了 7/20 郑州特大洪水的洪水泛滥情景，洪水深度的预测精度相比传统的广义渗透参数化方法提高了 3 倍。我们的研究提出了一种在缺乏标记数据的区域生成 UUSN 的新算法，并展示了出色的跨区域预测迁移能力，这有助于增强城市洪水模拟。