基于Elman动态神经网络的土壤墒情预测研究  

Soil Moisture Prediction Based on Elman Recurrent Dynamic Neural Networks Model

作　　者：邵月红[1] 刘永和[2] 

机构地区：[1]南京信息工程大学应用水文气象研究院,江苏南京210044 [2]河南理工大学资源环境学院,河南焦作454000

出　　处：《水土保持通报》2012年第5期257-260,共4页Bulletin of Soil and Water Conservation

基　　金：水利部公益性行业专项"水文气象分区线性矩法在防洪规划中的应用"(201001047);南京信息工程大学科研基金项目"雷达定量估测降水在洪水预报中的应用研究"(20100402);国家青年科学基金项目"基于广义非线性模型和天气预报资料的降水统计降尺度模型优选研究"(41105074);江苏高校优势学科建设工程资助项目(PAPD)

摘　　要：土壤墒情是一个非线性、时空异质性和动态不确定过程,利用Elman动态神经网络对研究区临沂站和平邑站土壤水分含量进行了预测。结果表明,所建立的网络模型能够对土壤墒情进行成功模拟,预测的土壤水分值与观测值吻合得较好,模拟精度较高。临沂站和平邑站模拟土壤墒情的平均绝对误差分别为1.08%和1.07%,平均相对误差为10.2%和11.0%。Elman动态神经网络模型利用其独特的非线性、非凸性和适应时变特性的能力从时空变率复杂的土壤水分运移系统中找出一定的演变规律,为土壤水分预测提供了一种有效可靠的方法。为了更好地验证该方法的优越性,还需要更多的样本数据,更多的区域和更全面的敏感影响因素来验证,以及更深层次的理论研究和分析。By considering the nonlinearity,spatial and temporal heterogeneity and dynamic uncertainty of soil moisture process,Elman recurrent neural network model is applied to the prediction of soil moisture in Linyi and Pingyi stations.Results show that the model achieves a high accuracy of soil moisture simulation and the simulated values agree well with observed values in the whole process.The mean absolute errors of prediction precision for Linyi and Pingyi stations are 1.08% and 1.07%,and the mean relative errors,10.2% and 11.0%,respectively.Elman recurrent neural network model can be used to find some evolutional characters and regular patterns from complex soil moisture system by taking advantage of its nonlinearity,non-appendage and self-adaption capacity.Therefore,the model provides a simple and efficient method which provides high accuracy and reliable precision for soil moisture simulation.In order to further evaluate the superiority of this method,a longer series of data,more investigations in other regions and more comprehensive influence factors are needed to deepen theoretical study and analysis because of limited samples and the highly dynamic uncertainty of soil moisture process.

关 键 词：ELMAN神经网络 土壤 墒情预测 

分 类 号：S157[农业科学—土壤学] TP18[农业科学—农业基础科学]