摘要
为了提高具有非线性和非稳定性特征的参考作物腾发量(ET0)时间序列的预测精度,提出基于经验模态分解(EMD)的BP神经网络预测模型。以大连地区1970~2006年间逐月ET0序列为例,首先应用经验模态分解(EMD)方法将ET0序列分解为具有不同尺度特征的本征模态函数(IMF),然后运用BP神经网络对ET0序列和分解得到的IMF进行训练,得到ET0序列的预测模型,并对ET0进行预测,最后将预测值及单纯的BP神经网络预测值分别与真实值进行对比分析。结果表明:EMD-BP模型预测值的平均绝对百分误差(MAPE)、均方根误差(RMSE)、平均绝对差(MAD)及判定系数(R2)分别为1.32%,0.0327,0.0278,0.9967;而BP模型相应的指标值分别为8.50%,0.2583,0.1839,0.8967。显然,EMD-BP模型的MAPE、RMSE、MAD值较小且R2值较大。因此,其预测精度及稳定性均优于单纯的BP模型,可作为ET0预测的一种参考。
In order to improve the prediction a ccuracy of the ET0 characterized by its nonlinear and unstable, the BP neural network prediction model based on empirical mode decomposition(EMD) was proposed. Taking the monthly ET0 series in Dalian area during 1970-2006 as example, by using the EMD method, we decomposed the ET0 series into intrinsic mode function of different scales. The ET0 series and IMF were trained by the BP neural network and the prediction model to forecast the ET0 was obtained. The prediction model was applied to forecast the ET0 value. Comparison analysis between the predicted value of the two models and the acture value was made. The result showed that the mean absolute percentage error(MAPE), root mean square error(RMSE), mean absolute difference(MAD) and the determination coefficient(R2) predicted by EMD-BP model was1.32%, 0.0327, 0.0278 and 0.9967, respectively. While the corresponding index value of the pure BP model was 8.50%,0.2583,0.1839 and 0.8967, respectively. Obviously, the MAPE, RMSE and MAD value of the EMD-BP model was smaller than the pure BP model, while the R2 value was higher than the pure BP model. Hence, the prediction accuracy and stability of it is superior to the pure BP model. Tt can be taken as a reference to the prediction of ET0.
出处
《沈阳农业大学学报》
CAS
CSCD
北大核心
2015年第1期67-73,共7页
Journal of Shenyang Agricultural University
基金
国家公益性行业(农业)科研专项项目(201303125)
高等学校博士学科点专项科研基金联合资助项目(20112103110007)
辽宁省优秀人才支持计划项目(2012)
辽宁省特聘教授基金项目(2013)
关键词
参考作物腾发量
经验模态分解
BP神经网络
预测精度
reference crop evapotranspiration
empirical mode decomposition
back propogation neural network
prediction accuracy
