摘要
尝试利用BP神经网络由常规地面气象观测要素估算土壤湿度.首先利用主成分分析确定少数与土壤湿度显著相关的特征气象要素,进而以这些特征气象要素为输入变量、以土壤湿度为输出变量建立BP神经网络.通过比较网络的性能选定了适用的训练方法及隐层神经元的个数,实现了对土壤湿度的估算.
This paper attempts to estimate soil moisture with routine meteorologic parameters. With principal component analysis we first determined some representative parameters like the shallow ground temperature, air temperature, deep ground temperature, wind and so on. Then with these parameters the input variables and soil moisture of 10 cm the output, we designed a BP neural network for the problem of making good relationship between meteorologic parameters and soil moisture. By comparing the network's performance, the "Levenberg-Marquardt" training algorithm was selected and a reasonable number for hidden layer neurons was chosen, finally achieved good estimate results.
出处
《华中师范大学学报(自然科学版)》
CAS
CSCD
北大核心
2010年第3期527-530,共4页
Journal of Central China Normal University:Natural Sciences
基金
教育部下一代互联网应用示范资助项目
关键词
气象观测要素
土壤湿度
主成分分析
BP神经网络
meteorological parameters
soil moisture
principal component analysis
BP neural network