[关键词]
[摘要]
基于丹江口水库1956—2016年逐月平均流量资料,采用数理统计手段分析丹江口水库年径流量的周期性和趋势性;从百项气候系统指数集和太阳黑子数中筛选出预测因子,构建月平均流量与预测因子间的多元线性回归模型和随机森林模型,实现丹江口水库月径流预测。结果表明,丹江口水库年径流量呈显著的减小趋势,并伴随有6~8a、18~21a两类尺度的周期振荡特性;以2017年逐月平均流量为例,随机森林模型和多元线性回归模型的预报合格率分别为83.3%、75.0%,预报精度均较好,且随机森林模型的预测精度优于多元线性回归模型,可用于丹江口水库月径流预测。
[Key word]
[Abstract]
The water inflow of Hanjiang River basin (HRB) continues to decline in recent years.The imbalance between supply and demand of water resources becomes more acute.Hence,it is necessary to study the middle and longterm water inflow prediction of HRB.If the annual and monthly runoff of the Danjiangkou reservoir can be predicted accurately,it would may ensure water supply safety of the Middle Route of the South-to-North Water Transfer Project for the northern region.Meanwhile,it can also provide technical support for making an annual water regulation plan of the Middle Route of the South-to-North Water Transfer Project. The basin above the Danjiangkou reservoir was selected as a research object.Water inflow of the Danjiangkou reservoir was predicted from qualitative and quantitative perspectives,respectively.The trend and periodicity characteristics of annual runoff were analyzed based on the monthly runoff data from 1956 to 2016.Characteristics of the annual water inflow were analyzed from qualitative perspectives in the Danjiangkou reservoir.The predictive factors which closely related to the runoff were selected based on hundredterm climate system indices set and sunspot number.The multiple linear regression model and random forest model were constructed for the monthly runoff prediction.The monthly runoff was also predicted from quantitative perspective. The results showed that the annual runoff exhibited a significant decreasing trend,with periodic oscillation characteristics of 6-8 years and 18-21 years.The monthly runoff of 2017 was predicted with a passing rate of 83.3% and 75.0% for the random forest model and multiple linear regression model,respectively.The high prediction accuracies indicated that they can be applied for the monthly runoff prediction in the Danjiangkou reservoir.By contrast,the random forest model was better. Presently,the runoff of the Danjiangkou reservoir is low.It is also predicted that water inflow less than normal runoff is expected in the near future.The multiple linear regression model and random forest model were constructed.Both simulation precisions were good based on calibration and validation results.They can be applied for the runoff prediction of the Danjiangkou reservoir.By contrast,the simulation performance of the random forest model is slightly better.Besides,the prediction precision of runoff in dry seasons is poor.The main reason is that prediction permissible error depends on the variation range of annual runoff.
[中图分类号]
[基金项目]