基于大数据的城市洪涝灾害预报预警理论与方法研究

Aiming at the urgent problems of short forecast period and low precision in the field of urban flood forecasting and warning, we propose a scheme to further study the theory and method of flood forecasting and warning in the wide expansion of big data, which combines the massive data and efficient processing advantages of big data with the principles of urban flood simulation. Firstly, the proposed method based on the semantic perception can establish the unified expression or index between urban flood process and big data，which deeply reveals the response mechanism between urban flood disaster process and big data. Secondly, with adoption of correlation, clustering and principal component analysis, we have a rigorous research on the data mining techniques for different objects, such as urban storm forecast, parameters of runoff, flood and disaster loss. Then, depending on the data mining technology, this paper provides the paths to unified downscaling of multi-source rainstorm forecast data, assimilation of model parameters and water information. Based on this, we build a urban flood forecasting model and real time correction method driven by large data, which is linkage by urban storm forecasting model, urban flood dynamic simulation model and evaluation model of urban flood disaster loss. Finally, we put forward the index system of urban flood disaster loss and the evaluation method of disaster grades. Simultaneously, we construct the dynamic warning mechanism of urban flood disaster. In conclusion, the scheme in this paper breaks through the traditional thought-ways and expands the theoretical principle of urban hydrology. The author sincerely hopes that this thesis could serve as a development in the research of urban hydrology and provides technical support for decision making of urban flood disaster.

针对城市洪涝灾害预报预警领域亟待解决的预见期短和精度低等问题，将大数据的海量信息和高效处理的优势与城市洪涝灾害模拟原理融合，研究大数据背景下城市洪涝灾害预报预警理论和方法。基于语义感知原理，建立城市洪涝灾害过程与大数据的统一表达与索引，揭示城市洪涝灾害过程与城市大数据间的响应机制；采用关联、聚类、主成分分析等方法，研究面向城市暴雨预报、产汇流参数、洪水、灾害损失等不同对象的数据挖掘技术；基于数据挖掘技术，提出多源暴雨预报数据统一降尺度、模型参数与水情数据同化等方法，构建大数据驱动下的城市暴雨预报模型、城市洪涝动态模拟模型和城市洪涝灾害损失评估模型耦合的城市洪涝灾害预报模型与实时校正方法；提出城市洪涝灾害损失指标体系和灾害等级评价方法，构建城市洪涝灾害动态预警机制。该项目突破传统研究思路，拓展了城市水文学的理论基础，发展了城市水文学的研究途径，可为城市洪涝灾害防控决策提供技术支撑。

本项目利用大数据分析技术与城市水文学交叉融合的优势，提出一套基于大数据的城市洪涝灾害预报预警理论与方法体系，包括灾害损失对城市洪涝灾害过程数据的响应关系、大数据驱动的城市洪 涝灾害动态预报与实时校正模型，以及洪涝灾害损失驱动的城市洪涝灾害多尺度动态预警模式等。本项目提出的理论与方法以郑州市为验证与实例应用典型区。主要研究成果如下：（1）构建了基于大数据的城市洪涝灾害过程数据本体模型，定量揭示了城市洪涝灾害过程数据对城市洪涝灾害损失的影响关系。研究发现，相同的要素在市域尺度和街区尺度上影响尺度不同，降雨在市域尺度和街区尺度对洪涝灾害的影响程度分别为0.281和0.649，市域尺度上，孕灾环境指标对灾害损失的影响高于承灾体指标，街区尺度上承灾体指标对灾害损失的影响比孕灾环境指标高。（2）将城市洪涝预报过程与和城市洪涝模拟模型结合，提出了模型敏感性参数快速识别与自适应调整方法，实现洪涝过程中模型参数快速校正和和预报过程结果实时修正。验证结果表明，参数识别响应速度到达秒级，而且准确率超过95%和，提高了城市洪涝预报的精度和时效性。（3）基于深度学习方法，研发了数据驱动的城市洪涝积水过程动态预报模型，该模型通过捕捉驱动因素与积水过程的非线性关系，建立数据之间的潜在联系，避免了物理模型中模型结构和参数不确定性。模型预报合格率在82%，峰值平均相对误差为5.74%。（4）基于数据挖掘，提出了缺灾情资料城市洪涝灾害经济损失和交通损失的量化方法，以经济-交通-人口综合损失为预警指标，点-线-面三层次为预警空间尺度，耦合洪涝预报过程，构建了“洪涝预报-损失评估-多尺度预警”的动态预警模式。本项目研究实现了大数据分析方法与城市洪涝预报预警过程的深度融合，拓展了城市洪涝灾害预报预警的研究途径与研究领域，推动学科交叉融合，为城市洪涝防灾减灾提供科技方法。