小样本观测资料条件下的耿贝尔极值水文频率分析模型
Gumbel Extreme Hydrological Frequency Analysis Model in Situations of Insufficient Data
作者:钱龙霞(南京邮电大学 理学院, 江苏 南京 210023;北京师范大学 水科学研究院, 北京 100875);王红瑞(北京师范大学 水科学研究院, 北京 100875);张韧(南京信息工程大学 气象灾害预报预警与评估协同创新中心, 江苏 南京 210044);焦志倩(惠济区环保局, 河南 郑州 450000)
Author:QIAN Longxia(School of Sci., Nanjing Univ. of Posts and Telecommunications, Nanjing 210023, China;College of Water Sciences, Beijing Normal Univ., Beijing 100875, China);WANG Hongrui(College of Water Sciences, Beijing Normal Univ., Beijing 100875, China);ZHANG Ren(Collaborative Innovation Center on Forecast Meteorological Disaster Warning and Assessment, Nanjing Univ. of Info. Sci. & Technol., Nanjing 210044, China);JIAO Zhiqian(Environmental Protection Agency of Huiji District, Zhengzhou 450000, China)
收稿日期:2018-03-22 年卷(期)页码:2019,51(5):41-48
期刊名称:工程科学与技术
Journal Name:Advanced Engineering Sciences
关键字:小样本;分布熵;最大熵估计;取值范围;最大日降水量
Key words:insufficient data;distribution entropy;maximum entropy estimation;data range;maximum daily precipitation
基金项目:国家自然科学基金项目(51609254;51479003;51879010)
中文摘要
气象水文要素极值预测是预防自然灾害、控制和降低灾害损失的重要基础性工作,然而传统极值水文频率分析模型需要大量样本资料,在资料稀少地区无法进行水文频率分析研究。本文构建一种小样本条件下的耿贝尔水文频率分析模型,提出最大熵估计方法,只需要水文变量的最小值和最大值这两个数据。耿贝尔水文频率分析模型建模步骤如下:1)首先定义耿贝尔分布熵;2)基于最大熵原理建立优化模型估计耿贝尔分布的未知参数;3)对耿贝尔分布模型进行K-S拟合检验。以黄河流域4个站点的最大日降水量的水文频率分析为例,验证最大熵估计的效果,结果表明:最大熵估计的拟合效果与传统参数估计方法几乎一样,而传统参数估计方法需要大量数据。为验证最大熵估计在小样本条件下的拟合效果,共进行了33次模拟实验。结果表明最大熵估计具有如下潜力:1)当样本长度大于25时,3种参数估计方法的拟合效果几乎一致;当样本长度小于15时,最大熵估计表现出非常大的优越性,极大似然估计的拟合效果最差。2)最大熵估计对最小值准确性的敏感性小,对最大值准确性较敏感。
英文摘要
The prediction of hydrometeorological extreme elements is valuable for preventing natural disasters, controlling and reducing losses. However, the methods for analyzing hydrological frequency are based on a large amount of data, and the studies on hydrological frequency analysis cannot be made in data scarcity regions. This paper builds a Gumbel hydrological frequency analysis model in situations when insufficient data are available, and proposes a new method of parameter estimation (maximum entropy estimation, MEE) for Gumbel distribution. The maximum entropy estimation requires only the lower and upper bounds of hydrology variables. The main steps of the Gumbel hydrological frequency analysis model are as follows: Firstly, the entropy of Gumbel distribution is defined. Secondly, an optimization model is built to estimate the parameters of Gumbel distribution based on the maximum entropy principle. Finally, a K-S goodness of fit test is made for Gumbel distribution model. The experiments of the hydrological frequency analysis of the maximum daily precipitation at the Yellow River were performed to validate the effect of MEE. All the experiments showed that the performance of the maximum entropy estimation was nearly identical to those of the conventional methods of parameter estimation which requires a large amount of data. Furthermore, thirty-three simulations were performed to validate the reliability of MEE in situations when insufficient data were available. The simulations show that the MEE has the following potentials: The performance of three parameter estimation methods are almost identical when the sample size is greater than 25. However, the MEE performs the best and the maximum likelihood estimation performs worst where the sample size is less than 15. Moreover, the MEE is not sensitive to the accuracy of the minimum value but sensitive to that of the maximum value.
【关闭】