基于输入变量筛选与卷积神经网络的黄河流域高精度灌溉用水空间栅格研究  

Research on High-Precision Irrigation Water Data Product in the Yellow River Basin Based on Input Variable Screening and Convolutional Neural Network

作　　者：张敏 ZHANG Min(Architectural Design and Research Institute Co.,Ltd,Taiyuan University of Technology,Taiyuan 030000,China)

机构地区：[1]太原理工大学建筑设计研究院有限公司,太原030000

出　　处：《节水灌溉》2023年第10期49-54,共6页Water Saving Irrigation

摘　　要：以黄河流域为研究对象,基于地理变量相互关系法筛选了与灌溉用水强相关的地理变量,利用卷积神经网络实现了2000-2013年黄河流域内灌溉用水的空间分布式模拟,采用均方根误差、纳什效率系数与相对误差3个指标对模拟的灌溉用水量进行精度评价。结果表明:在空间集聚性上,2000-2013年均表现出正向的集聚性,且有随时间下降趋势;在时间演变规律上,不同地区的年均变化率不同,在山西晋城年增长率最高为12.14%,在甘肃甘南藏族自治州年减少率最高为-5.49%;在精度上,模型精度较高,误差在可接受范围内。Taking the Yellow River Basin as the research object,based on geographical variable interrelationships,this study selected the geographical variables which strongly correlated with irrigation water use,using the Convolutional Neural Networks,the spatial distribution of irrigation water use in the Yellow River Basin from 2000 to 2013 was generated,the accuracy of the simulated irrigation water use was evaluated using three metrics:root mean square error,Nash-Sutcliffe efficiency,and relative error.The results showed that,in terms of spatial agglomeration,positive agglomeration was observed from 2000 to 2013,with a decreasing trend over time;in terms of temporal evolution,the annual growth rates in different regions varied,with the highest rate of 12.14%in Jincheng,Shanxi and the highest annual decrease rate of-5.49%in Gannan Tibetan Autonomous Prefecture,Gansu.In terms of accuracy,this model had high accuracy,and the error was within an acceptable range.

关 键 词：灌溉用水 空间分布式模拟 黄河流域 地理变量 

分 类 号：S607[农业科学—园艺学] TV882.1[水利工程—水利水电工程]