基于神经网络的中长期径流预报时间尺度研究  

Research on the Time-scale of Mid-to-Long-term Runoff Forecast Based on Neural Networks

作　　者：张素琼[1] 张艳军[1] 宋星原[1] 邹霞[1] 刘佳明[1] 袁迪[1] 

机构地区：[1]武汉大学水利水电学院,武汉430072

出　　处：《中国农村水利水电》2014年第8期110-114,共5页China Rural Water and Hydropower

基　　金：国家自然科学基金项目(51279143;51279140;51379149);国家自然科学基金青年科学基金项目(51209162)

摘　　要：为探究预报时间尺度选择对中长期径流预报精度的影响,以长江流域上游的寸滩站和宜昌站、中游的螺山站和汉口站以及下游的大通站各33年的流量资料为基础,采用人工神经网络方法,分别选取年、季、月、旬及3日五个预报时间尺度,以1981-2000年作为率定期,2001-2012年作为验证期,对5个水文站的径流进行模拟和预报,并比较其计算结果误差。研究结果表明,随着预报时间尺度的加大,径流模拟和预报的误差先增大后减小,从整体水平来看,月尺度的计算误差最大,3日尺度和年尺度的误差最小。研究结果对实际径流预报工作具有一定的指导意义。Based on the 33-year runoff data of Cuntan Station, Yichang Station, Luoshan Station,, Hankou Station, and Datong Sta- tion,, the influence of time scales on the accuracy of mid-to-long-term runoff forecast are explored. The set from 1981 to 2000 and from 2001 to 2012 is used as the calibration period and the validation period, respectively, scales of year, season, month, ten-day and three-day. Moreover, the calculation errors are compared. The results show that the average error of runoff simulation and pre- diction first increases and then decreases with the increase in the forecasting time scale. Overall, the maximal relative error appears on the monthly forecast and the minimum appears on the three-day and yearly forecast. The research results have certain practical guidance to the actual runoff forecasting work.

关 键 词：时间尺度 中长期水文预报 神经网络 长江流域 

分 类 号：P338.2[天文地球—水文科学]