基于缺省因子的BP-ANN土壤墒情预报简化模型  

Soil moisture forecast BP-ANN model and simulation based on sensitivity analysis

作　　者：黄令淼[1] 任树梅[1] 杨培岭[1] 税朋勃[2] 曹建武[2] 周嵘[2] 

机构地区：[1]中国农业大学水利与土木工程学院,北京100083 [2]北京市水利水电技术中心,北京100073

出　　处：《中国农业大学学报》2013年第5期166-172,共7页Journal of China Agricultural University

基　　金：北京市科学技术计划资助项目(pxm2009_035324_092070);北京市干旱风险评估项目

摘　　要：对影响土壤墒情的主要气象要素,平均气温、相对湿度、日照时数、平均风速、蒸降差和前一旬土壤墒情进行分析合并,建立BP-ANN土壤墒情预报6因子模型;通过缺省因子检验法,判断土壤墒情对6个因子敏感程度,简化冗余因子,构建BP-ANN的3因子(相对湿度、日照时数、前一旬土壤相对湿度)墒情预报模型。结果表明:3因子模型均方根误差3.55,具有数据收集和处理量小的优点,基本能够达到所需精度和拟合度。在北京市山区和平原区2个典型站点的模拟检验表明,3因子模型实测值与预测值的拟合关系均达到极显著相关水平,可操作性强的特点。Basing on the analysis and merging the major meteorological elements which affect soil moisture, the BP-ANN soil moisture forecast model with six factors was established. The soil moisture sensitivity on six factors was determined by sensitivity analysis. The moisture prediction BP-ANN model was built based on three factors （relative humidity, sunshine hours,average previous ten days of soil moisture）. The study showed that the three-factor model root mean square error was 3. 55, with the advantage of small data collection and less processing, which could achieve the required accuracy. The test that three-factor model was applied to mountains and the plains of Beijing showed the measured and predicted values reached a very significant level. It presented the strong characteristics of operability.

关 键 词：土壤墒情 预测预报 人工神经网络 缺省因子分析法 

分 类 号：S156.2[农业科学—土壤学] X592[农业科学—农业基础科学]