中山大学学报(自然科学版) ›› 2020, Vol. 59 ›› Issue (1): 105-113.doi: 10.13471/j.cnki.acta.snus.2020.01.013

• • 上一篇    下一篇

基于机器学习算法模型的巫山县洪水灾害研究

牟凤云,杨猛,林孝松,龙秋月,李梦梅,何勇   

  1. 重庆交通大学建筑与城市规划学院,重庆400074
  • 收稿日期:2019-02-27 出版日期:2020-01-25 发布日期:2020-02-28

The flood disasters in Wushan county based on machine learning algorithm model

MU Fengyun,YANG Meng,LIN Xiaosong,LONG Qiuyue,LI Mengmei,HE Yong   

  1. School of Architecture and Urban Planning, Chongqing Jiaotong University,Chongqing 400074,China
  • Received:2019-02-27 Online:2020-01-25 Published:2020-02-28

摘要: 利用机器学习算法RF模型、K-means模型与ARMA模型,对巫山县范围内12 369条径流河段进行分类预测,研究水文参数在时间序列上的变化规律,探究降雨-径流演变规律;并结合GIS空间可视化技术,综合研究区地理环境,基于RF模型预测洪水致灾范围,分析洪水灾害预测结果的空间特征。结果表明:① RF模型能有效预测降雨-径流演变过程中参数的重要性,当降雨强度为125 mm、150~175 mm时,预测出水位、流速变化率最大;降雨强度为100~175 mm时,预测出流速变化率最为剧烈;② 利用ARMA模型预测出河流比降、流量等水文参数回归性最好,较低等级河流所预测参数中,水位、流速变化率最为明显,流量无明显变化。相较于水位变化率,流速变化更为强烈,流速、水位变化率主要集中于等级较高河流;③ 机器学习算法能有效预测研究区洪水易发程度,在表征研究区水文参数时,水位变化主要集中于西北部、中南部,东北部与中南部水位变化率显著,预测出部分地区水位可升至20 m,处极危险状态。


关键词: 机器学习, RF, K-means, ARMA, 降雨-径流, 洪灾

Abstract: Kmeans model and ARMA model was applied to classify 12 369 runoff river reaches in Wushan County, and study the change of hydrological parameters in time series, and probes into the evolution of rainfall runoff process by using the machine learning algorithm RF model. Combined with the GIS spatial visualization technology and the geographical environment of the study are, the flood disaster scope is predicted and the spatial distribution of flood disasters is analyzed based on the RF model. The results show that: 1) The RF model can effectively predict the importance of parameters in the rainfallrunoff process. When the rainfall intensities are 125 mm and 150-175 mm, the change rate of water level and velocity is the largest; when the rainfall intensity is 100-175 mm, the change rate of velocity is the most intense. 2) The ARMA model is used to predict the river gradient, flow and other hydrological parameters with the best regression. Among the predicted parameters of lower level rivers, the change rate of water level and flow velocity is the most obvious, and the flow quantity has no obvious change. Compared with the change rate of water level, the change rate of flow rate is more intense, and the change rate of flow rate and water level is mainly concentrated in the river with higher grade. 3) The machine learning algorithm can effectively predict the flood prone degree in the study area. When characterizing the hydrological parameters of the study area, the change of water level is mainly concentrated in the northwest and the south central part, and the change rate of water level in the northeast and the south central part is significant. It is predicted that the water level in some areas can rise to 20 m in extremely dangerous condition.


Key words: machine learning, RF, K-means, ARMA, rainfall runoff, flood disaster

中图分类号: