水文序列展延的MTSE法和WA-BP法研究  

MTSE Method and WA-BP Method of Hydrological Series Extension

作　　者：宋胜丽[1] 王栋[1] 吴吉春[1] 朱庆平 王玲[3] 

机构地区：[1]南京大学地球科学与工程学院水科学系,江苏南京210093 [2]中水新华国际工程咨询公司,北京100053 [3]水利部黄河水利委员会水文局,河南郑州450001

出　　处：《水文》2011年第1期1-7,共7页Journal of China Hydrology

基　　金：国家自然科学基金(40725010;40730635);水利公益项目(200701024);南京大学青年骨干教师和优秀中青年学科带头人培养计划项目

摘　　要：传统相关展延(CE)法较多地考虑了序列总体的变化特征,并且两序列需呈线性关系。但是不同的水文变量,具有不同的时空分布特性,可以提供插补展延的信息资源是不同的。因此,本文对该法进行了改进,把小波分析和BP神经网络应用于水文序列的展延,提出了两种改进方法:多时间尺度展延(MTSE)法和小波-BP网络展延(WA-BP)法,力图既考虑水文序列的多时间尺度特征,又突破线性关系的束缚,充分利用水文变量提供的信息资源。通过实例对比分析表明:(1)CE法、MTSE法和WA-BP法的展延精度依次提高。(2)当水文序列的波动变化不显著时,CE法、MTSE法和WA-BP法都满足相对误差的均值小于或等于0.2的精度要求,且WA-BP法稍优越于MTSE法。(3)当水文序列的波动变化显著时,CE法和MTSE法不是很适用,选择WA-BP法仍然可以得到较精确的展延结果。The Correlation Extension （CE） method considers more of the whole variation characteristics of series, and the two series need to be linear. But the various hydrological series have different space-time distribution characteristics, and the information resource that provided to interpolate and extend is also different. So, some improvements were made for this method, the wavelet analysis and BP neural network were applied to hydrological series extension, and the new methods of multiple time scale extension （MTSE） and wavelet analysis-BP neural network extension （WA-BP） were advanced, thus, it was not only to consider the multiple time scale characteristics of the hydrological series, but also broke the linear restriction, and made full use of the information resource provided by hydrological series. The analysis of some examples show that （1） The extension accuracy of CE method, MTSE method and WA-BP method get higher in sequence; （2）When fluctuation change of hydrological series are not significant, the mean relative errors of the CE method, MTSE method and WA-BP method are less than or equal to 0.2, and the WA-BP method slightly superior to MTSE method; （3）When fluctuation change of hydrological series are significant, the CE method and MTSE method are not very applicable, select the WA-BP method can still get more accurate extension result.

关 键 词：水文序列 小波分析 BP神经网络 相关展延 多时间尺度展延 WA—BP展延 

分 类 号：P333.9[天文地球—水文科学]