基于EMA法的洪水频率分布参数估计方法研究

发布时间：2018-01-27 01:14

本文关键词： 洪水频率分析参数估计简单和非简单样本期望矩法部分和高阶线性矩法　出处：《西北农林科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：受全球气候变化和人类活动加剧的影响，洪涝灾害带来的损失越来越严重。洪水设计值的估计精度直接关系到工程的投资和安全，因而设计洪水的计算精度研究成为洪水计算的重要问题。洪水频率分析是推求给定设计频率（或重现期）洪水设计值的关键技术，合理的洪水设计值可为洪水管理、大坝溢洪道的设计和水利工程规划决策提供依据。本文综述了近年来国内外水文频率分析的研究进展，探索了常用和新型参数估计方法在简单和非简单样本中的应用。本文主要的研究内容如下。（1）选定P-III型分布线型，研究新型参数估计方法——期望矩法（ExpectedMoments Algorithm，简称EMA）和常用参估方法（矩法、权函数法、概率权重矩法和线性矩法）在非简单样本中的应用。将其运用到陕西省10个水文测站的年最大洪峰流量（加入历史洪水资料）中，并进行拟合结果分析和评价。（2）选定广义极值分布（Generalized Extreme Value Distribution，简称GEV）线型，研究新型参估方法（部分线性矩法和高阶线性矩法）在简单样本中的应用，并进行研究区年最大洪峰流量系列的实例应用和分析。（3）采用蒙特卡洛试验进行参数估计方法的统计性能研究。以定量和直观图形评判标准为评判依据，假设总体统计参数，构建简单和非简单样本（门限模型和个数模型），借助Matlab2009a计算各组设计方案的结果，最后分析和评判相应参数估计方法的优劣。通过上述研究，可得以下结论。（1）实例应用中，非简单样本中各参数估计方法的拟合效果优劣各异。其中EMA法对长历史洪水资料系列的拟合效果较佳，概率权重矩法和线性矩法的结果接近，且历史洪水资料的长度对其拟合效果影响不大，而权函数法优于矩法。简单样本中部分线性矩法和高阶线性矩法均能有效改善研究洪峰系列较大流量的拟合效果，可为外延的大重现期设计值提供有利依据。（2）蒙特卡洛试验中，在各删失水平下的部分线性矩法和各高阶线性矩法均具有良好的统计性能，且随着删失水平和阶数的增长，参数估计值和设计值的估计偏差随之降低。但部分线性矩法和高阶线性矩法的设计值估计偏差随着重现期的增大而相应变大。而非简单样本的模拟结果中，门限值模型和个数模型结果相差不大，实测资料和历史洪水资料越长，各参数估计方法的统计性能均有所改善，其中期望矩法的效果最为显著，概率权重矩法和线性矩法较为稳定，，权函数法和矩法次之。
[Abstract]:Due to the influence of global climate change and the intensification of human activities, the losses caused by flood and waterlogging are becoming more and more serious. The accuracy of flood design value estimation is directly related to the investment and safety of the project. Therefore, the calculation accuracy of design flood becomes an important problem in flood calculation, and flood frequency analysis is the key technology to calculate the design value of flood given design frequency (or recurrence period). Reasonable flood design value can provide basis for flood management, dam spillway design and water conservancy project planning decision. This paper reviews the recent advances in hydrologic frequency analysis at home and abroad, and explores the application of common and new parameter estimation methods in simple and non-simple samples. The main contents of this paper are as follows. 1) the P-III type distribution line is selected to study the new parameter estimation method, the expected moments method, and the expected moments Algorithm. EMA) and commonly used parameter estimation methods (moment method, weight function method). The application of probabilistic weighted moment method and linear moment method in non-simple samples is applied to the annual maximum Hong Feng discharge (including historical flood data) of 10 hydrological stations in Shaanxi Province. The fitting results were analyzed and evaluated. 2) selecting generalized Extreme Value distribution. The application of new parameter estimation methods (partial linear moment method and high order linear moment method) in simple samples is studied, and the application and analysis of the annual maximum Hong Feng flow series in the study area are carried out. The statistical performance of the parameter estimation method is studied by Monte Carlo test. Based on the quantitative and visual graph evaluation criteria, the statistical parameters of the population are assumed. The simple and non-simple samples (threshold model and number model) were constructed, and the results of each group design scheme were calculated by Matlab2009a. Finally, the advantages and disadvantages of the corresponding parameter estimation methods were analyzed and evaluated. Through the above research, we can draw a conclusion. In the application of the example, the fitting effect of each parameter estimation method in the non-simple sample is different, among which the EMA method is better for the long history flood data series. The results of probabilistic weighted moment method and linear moment method are close, and the length of historical flood data has little effect on the fitting effect. But the weight function method is superior to the moment method. The partial linear moment method and the high order linear moment method in the simple sample can effectively improve the fitting effect of studying Hong Feng series large flow rate, and can provide a favorable basis for the design value of the extension of large recurrence period. 2) in Monte Carlo test, the partial linear moment method and the higher order linear moment method have good statistical performance under each censored level, and with the increase of deletion level and order. The deviation between the parameter estimation and the design value decreases, but the partial linear moment method and the higher order linear moment method estimate the deviation of the design value with the increase of the recurrence period, but not in the simulation results of simple samples. The results of the threshold model and the number model are not different. The longer the measured data and the historical flood data, the better the statistical performance of each parameter estimation method, among which the expected moment method is the most effective. The probabilistic weight moment method and linear moment method are more stable than the weight function method and moment method.
【学位授予单位】：西北农林科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：P333.2

【相似文献】