两河口水库中长期入库径流概率预报研究  

Study on the Medium and Long-term Streamflow Probability Forecast of Lianghekou Reservoir

作　　者：刘治理 刘源 夏远洋 陈平 LIU Zhi-li;LIU Yuan;XIA Yuan-yang;CHEN Ping(Yalong River Hydropower Development Co.,Ltd.,Chengdu 610051,Sichuan Province,China)

机构地区：[1]雅砻江流域水电开发有限公司,四川成都610051

出　　处：《中国农村水利水电》2024年第12期27-34,共8页China Rural Water and Hydropower

基　　金：国家自然科学基金项目(52179016)。

摘　　要：针对确定性径流预报不能提供径流预报不确定性度量,难以充分发挥预报价值的问题,提出基于Vine copu⁃la函数的贝叶斯转移预报(Vine Copula based Bayesian Transition Forecast,VCBTF)模型,对确定性径流预报结果进行后处理得到概率预报结果。首先,通过多种水文预报经验模型获得入库径流预报的集合结果;其次,采用集合Kalman滤波技术将入库径流预报的集合结果融合为后验预报结果;最后,采用VCBTF方法对融合预报结果进行后处理以量化径流预报的不确定性并得到入库径流的概率预报结果。以雅砻江流域具有多年调节能力的两河口水库为研究对象开展中长期入库径流概率预报研究,考虑以月为预见期,研究结果表明:①VCBTF模型在均方根误差、平均绝对误差和纳什效率系数评价指标上均优于确定性径流预报模型,且至少增加2.7%的预报合格率;②VCBTF模型的连续概率排位分数指标的平均值为104.54 m^(3)/s优于GPR模型的106.92 m^(3)/s,VCBTF模型的α-index指标均高于0.89且优于GPR模型的相应结果,表明基于VCBTF模型的概率预报结果具有更高的可靠性;③VCBTF模型的单位平均相对区间宽度所包含的实测点据比例指标的平均值为2.2优于GPR模型的1.87,表明VCBTF模型的概率预报结果具有更高的集中度。因此VCBTF模型可以更有效降低径流预报的不确定性,能够提供更可靠的预报区间信息,可为开展流域梯级水库优化调度提供有力的技术支持。To solve the problem that deterministic streamflow forecast can not provide the uncertainty measurement of streamflow forecast and is difficult to give full play to the forecast value,this paper proposes a Vine copula-based Bayesian Transition Forecast(VCBTF)model to obtain the probability forecast result by post-processing the deterministic runoff streamflow results.Firstly,the ensemble results of the streamflow forecast were obtained through a variety of hydrological forecasting empirical models.Secondly,the ensemble Kalman filtering technology was used to fuse the ensemble results of the forecasted streamflow into the posterior forecast results.Finally,the VCBTF was used to post-process the fused streamflow forecast results to quantify the uncertainty of the runoff forecast and obtain the probabilistic forecast results of the flow.Taking Lianghekou Reservoir with multi-year regulation capacity in the Yalong River Basin as a case study,the medium and long-term streamflow probability forecast is carried out.The results show that:①The VCBTF model is superior to the deterministic run⁃off forecast model in terms of Root Mean Square Error,Mean Absolute Error and Nash-Sutcliffe efficiency index,and the Qualified Rate of forecast is increased by at least 2.7%;②The average value of continuous probability ranking score index of the VCBTF model is 104.54 m^(3)/s,which is better than that of the GPR model with 106.92 m^(3)∙/s,and theα-index of the VCBTF model is higher than 0.89 and better than the corresponding result of the GPR model,indicating that the probabilistic forecast results based on the VCBTF model have higher reliabili⁃ty;③The average value of the Percentage of measured points contained in the Unit Confidence Interval of the VCBTF model is 2.2,which is better than that of the GPR model with 1.87,indicating that the probabilistic prediction results of the VCBTF model have a higher concentra⁃tion.Therefore,the VCBTF model proposed in this paper can more effectively reduce the forecast uncertainty

关 键 词：中长期径流预报 预报不确定性 Vine Copula 概率预报 两河口水库 

分 类 号：P338.2[天文地球—水文科学]