%0 Journal Article %T 水文时间序列概率预报方法的通用架构 %T Universal framework for hydrological time series probabilistic forecasting %A 桑燕芳 %A 李鑫鑫 %A 谢平 %A 刘勇 %A SANG,Yanfang %A LI,Xinxin %A XIE,Ping %A LIU,Yong %J 湖泊科学 %J Journal of Lake Sciences %@ 1003-5427 %V 30 %N 3 %D 2018 %P 611-618 %K 水文过程;中长期水文预报;不确定性;概率预报;分解合成 %K Hydrological process;medium-long-term hydrological forecasting;uncertainty;probabilistic forecasting;decomposition and combination %X 在准确揭示水文过程变化特性的基础上开展中长期(月尺度及以上)水文预报,是掌握未来水文情势和演变规律,以及研究解决实际水文水资源问题的重要基础.水文时间序列预报方法是揭示未来水文情势和演变规律的重要技术手段.本文首先梳理了目前常用的各类水文序列预报方法,分析讨论了各方法的基本原理和主要缺陷.然后,通过综合分析相关研究成果,总结得到关于水文序列预报方法的4点重要认识:序列预报前应进行序列分解;序列中确定成分和随机成分应分别建模预报;序列预报结果需要估计不确定性;模型集成效果常常优于单个模型效果.最后,提出一个水文时间序列概率预报方法的通用架构.利用该通用架构能够克服常规模型或方法的缺陷,进行物理成因分析的基础上,针对水文序列中不同特性的确定成分和随机成分别进行分析,既可得到准确的确定性预报结果,又可对预报结果的不确定性进行定量评估,并可提高最终预报结果的合理性和可靠性. %X Accurately identifying the complex characteristics of hydrological processes, and then conducting hydrological simulation and forecasting at large time scales (longer than monthly scale) is an essential and important issue, because it is the basis of understanding the future hydrological regimes, and solving various water resources problems. Hydrological time series simulation and forecasting (HTSF) is an effective approach to revealing the future hydrological regimes. In this study, recent progresses on those methods used for HTSF are summarized, and the basic ideas and main shortcomings of these methods are discussed. We summed up four main understandings about the issue of HTSF:(1) accurate decomposition of series should precede the hydrological time series simulation and forecasting; (2) deterministic components and random component in hydrological time series should be forecasted respectively; (3) uncertainty evaluation should be carefully studied for HTSF; and (4) hybrid model generally performs better than single model for HTSF. Finally, we proposed a framework for hydrological time series simulation and forecasting. It is to first separate different deterministic components and to remove noise in original hydrological time series; then, forecast the former with considering physical causes of time series and quantitatively describe noise's random characters; finally, add them up and obtain the final hydrological time series forecasting result. Forecasting of deterministic components is to obtain the deterministic forecasting results, and noise analysis is to estimate uncertainty by considering statistical significance. The framework can overcome the shortcomings of conventional methods, and it can improve the accuracy and reasonability of the results of hydrological time series forecasting. %R 10.18307/2018.0303 %U http://www.jlakes.org/ch/reader/view_abstract.aspx %1 JIS Version 3.0.0