基于小波消噪的参考作物腾发量RBF网络预测方法

RBF network forecast method for evapotranspiration of reference crops based on wavelet denoising

介绍小波消噪的原理和步骤。以北方某流域甲站2001—2005年逐日气象数据为基本资料,进行10阶Dmey小波消噪,然后构建预测ET0的前馈网络模型(RBF-ET0),用2001—2004年的资料作为训练样本,对2005年的ET0进行预测,并与Penman-Montieth公式计算值进行比较。结果为:预测值与目标值的相关系数为0.991 2,相对误差的平均值为6.56%,相对误差小于20%,15%,10%的合格率分别为93.88%,85.66%,73.51%,与未经小波消噪处理的RBF-ET0模型预测结果相比,预测精度有明显提高。

The principles and procedures of the wavelet denoising were introduced.By using the daily meteorological data of a station in a northern river basin from 2001 to 2005 as the basic information,the daily meteorological data were denoised by the tenth order Dmey wavelet.A feed-forward neural network forecast model for evaportranspiration of reference crops（RBF-ET0） was established.The meteorological data from 2001 to 2004 were taken as the training samples.The evaportranspiration of the reference crops in 2005 was predicted and compared with that calculated by the Penman-Montieth formula.The results show that the correlation coefficient of the predicted value and the target value is 0.991 2,and the average relative error is 6.56%.The qualified rates of the relative error less than 20%,15% and 10% are 93.88%,85.66% and 73.51% respectively.The prediction accuracy is obviously improved compared with that by the pure RBF-ET0 model.