论文标题
基于动态系统理论的实用数据驱动的洪水预测：日本的案例研究
Practical data-driven flood forecasting based on dynamical systems theory: Case studies from Japan
论文作者
论文摘要
数据驱动的洪水预测方法很有用，尤其是对于缺乏水文信息来构建物理模型的河流。尽管这些以前的方法可以仅使用过去的水位和降雨数据来预测河流阶段，但它们不能轻易处理以前没有经验的水位，并且需要大量数据来构建准确的模型。在这里，我们专注于相空间重建方法，并开发一种实用数据驱动的预测方法来克服现有问题。所提出的方法可以处理未经验的水位，并仅使用少量的水上事件提供预测。我们将方法应用于实际河流的数据，并在现有方法中实现了最佳的预测性能，包括物理径流模型，数据驱动的多层感知器以及基于相位空间重建的常规方法。此外，拟议的方法还预测了陡峭河流早6小时的疏散警告的超出性。鉴于其性能和可维护性，可以将提出的方法应用于许多实际河流以进行早期疏散。
Data-driven flood forecasting methods are useful, especially for the rivers that lack hydrological information to build physical models. Although these former methods can forecast river stages using only past water levels and rainfall data, they cannot handle previously unexperienced water levels easily, and require a large amount of data to build accurate models. Here, we focus on phase-space reconstruction approaches, and develop a practical data-driven forecasting method to overcome the existing problems. The proposed method can handle the unexperienced water levels and provide forecasts using only a small number of water rise events. We apply the method to data from actual rivers, and it achieved the best forecast performance among existing methods, including a physical runoff model, a data-driven multi-layer perceptron, and a conventional method based on phase-space reconstruction. In addition, the proposed method also forecasted the exceedance of the evacuation warning level 6 h earlier for steep rivers. Given its performance and maintainability, the proposed method can be applied to many actual rivers for early evacuation.
