干旱区生态环境知识资源中心(Arid): 基于改进BP神经网络的干旱区芦苇腾发量预测模型

Arid

	基于改进BP神经网络的干旱区芦苇腾发量预测模型
其他题名	Prediction modeling for evapotranspiration of Phragmites australis base on the modified BP neural network in arid regions
	苏里坦 1; 玉米提 2; 宋郁东 1
来源期刊	干旱区地理
ISSN	1000-6060
出版年	2011
卷号	34 期号:4 页码:551-557
中文摘要	植被蒸腾与地表蒸发研究是土壤-植物-大气连续体系统研究的核心部分,也是干旱区自然植被需水量研究的前提。计算植被蒸腾与地表蒸发的方法众多,但需要大量参数的输入,其中水分特征曲线、非饱和导水率、水分扩散率、比水容重、导热率、比热容等土壤水热特性参数很难获取。针对这些繁琐问题,在传统BP网络中加入有动量的梯度下降法,建立了基于有动量梯度的改进BP网络模型,提出了通过易获取的气象、植物、土壤数据预测植被腾发量的新方法,并对该模型的有效性进行了初步验证。研究结果表明:应用有动量梯度的改进BP网络模型可以很好的反映环境因子（气象因子、植物因子、土壤因子）与芦苇腾发量（芦苇蒸腾量、地表蒸发量）之间的非线性函数映射关系。在无长序列气象资料的条件下,利用改进BP神经网络,以日为时间尺度来预测芦苇蒸腾量和地表蒸发量可以取得很好的效果,克服了以往的BP算法收敛速度慢和预报精度低的缺欠。与传统回归模型相比,改进BP网络模型中的输入参数（太阳净辐射、大气温度、大气相对湿度、冠层顶风速、芦苇盖度和土壤含水率）比较容易测定,且模型也便于应用,能够更好刻画植被腾发量的复杂非线性特性,为干旱区自然植被腾发量估算和生态需水量计算提供了新的思路和方法。
英文摘要	Evapotranspiration evaluation is the hard core on SPAC research; it is also the precondition for research of natural vegetation water requirements.There are many quantification methods of vegetation evapotranspiration,which require a large quantity of parameters.The method of neural network is an ideal tool to deal with the problem of non-linearity,but serious correlation of input data makes the network unsteady.To predict the vegetation evapotranspiration on arid land,an attempt has been made in this paper to investigate its application possibility with gradient descent method combining neural network.The modified BP network model has been established based on the adding gradient descent method of momentum to the traditional BP network in this study.A new method was tested for predicting vegetation evapotranspiration by taking meteorologic data,plant data and soil data,and the validity of model.Results show that non-linear function reflection relation between environmental factors such as meteorological factor,plant factor and soil factor,and evapotranspiration of Phragmites australis can be reflected with the help of combining the gradient descent method and the BP neural network.The Matlab software was used to predict vegetation evapotranspiration in arid area of Xinjiang.The results of prediction indicate that the maximum relative errors of the evaporation and the transpiration were 7.62% and 11.63%,respectively,and the maximum correlation coefficients were 0.936 9 and 0.957 4,respectively,representing a high prediction precision.Therefore the combination method is better than un-combination one,which eliminates the correlation index of samples and reduces the input dimension of neural network.It makes node numbers of generalized regression neural network input layer cut down to six from eight,and has the effects of simplifying structure and strengthening stability of neural network.The combination method indicates that the proposed modelling is more reliable and its performance is better than that of conventional method.The combination model can resolve the problem that multi-collinearity in factors can not be effectively identified and eliminated by regression of the least square method.The simulated results of modified BP network model show that the prediction precision is high under the condition of having no long-term climatic data.The model provides a newly effective and feasible way for the evaluation of natural vegetation evapotranspiration and the calculation of ecological water requirements.
中文关键词	改进BP神经网络 ; 干旱区 ; 芦苇蒸腾 ; 地表蒸发 ; 预测模型
英文关键词	modified BP neural network arid regions transpiration of Phragmites australis ground surface evaporation prediction model
语种	中文
国家	中国
收录类别	CSCD
WOS类目	AGRICULTURE MULTIDISCIPLINARY
WOS研究方向	Agriculture
CSCD记录号	CSCD:4271773
来源机构	中国科学院新疆生态与地理研究所
资源类型	期刊论文
条目标识符	http://119.78.100.177/qdio/handle/2XILL650/226464
作者单位	1.中国科学院新疆生态与地理研究所, 荒漠与绿洲生态国家重点实验室, 乌鲁木齐, 新疆 830011, 中国; 2.新疆维吾尔自治区水利厅水土保持技术推广中心, 乌鲁木齐, 新疆 830000, 中国
推荐引用方式 GB/T 7714	苏里坦,玉米提,宋郁东. 基于改进BP神经网络的干旱区芦苇腾发量预测模型[J]. 中国科学院新疆生态与地理研究所,2011,34(4):551-557.
APA	苏里坦,玉米提,&宋郁东.(2011).基于改进BP神经网络的干旱区芦苇腾发量预测模型.干旱区地理,34(4),551-557.
MLA	苏里坦,et al."基于改进BP神经网络的干旱区芦苇腾发量预测模型".干旱区地理 34.4(2011):551-557.