【作者】 吕静渭；

【导师】 马孝义；

【作者基本信息】 西北农林科技大学 ， 农业水土工程， 2010， 硕士

【摘要】 水资源是支撑人类生存和经济增长与可持续发展的基本要素之一,而其中径流要素的变化主导着水资源系统的变化,由于受气候、社会发展与人类活动的综合影响,径流变化规律非常复杂,具有不确定性、随机性、多时间尺度性、非线性等特性。论文依托国家科技支撑计划项目(2006BAD11B04)和国家自然科学基金项目(50879072),选用泾河流域为研究对象,利用传统分析方法、经验模态分解法和小波分析方法,系统研究了泾河流域水文气象序列的变化特性,在此基础上将经验模态分解法、小波分解与BP人工神经网络结合建立预测模型对径流变化进行预测分析,主要取得如下研究成果:(1)泾河流域径流年内分配极不均匀,差异较大,年内分配表现为单峰型,径流集中程度较高,年最大径流量出现在8月份;径流年际变化呈现丰枯交替的年际关系。(2)采用滑动平均法和Kendall秩次相关检验法对泾河流域的年径流序列进行了趋势分析,得出泾河流域年径流量呈现显著的下降趋势,平均下降幅度达1.13亿m3/10a;同时对引起径流量变化的因素—气象序列进行分析,年际之间流域降水量的变化影响着河流径流量的变化,径流量与降水量有一致的变化趋势,与气温的年际变化整体一致性不是很明显。(3)采用EMD方法对泾河流域的径流量及平凉、环县、西峰镇和长武四个水文站的历年实测降水序列进行了多时间尺度分析,泾河流域年径流量变化的多时间尺度表现在:具有准2~3a,6~8a,10~12a和21a的波动周期;泾河流域近45年年降水变化可能存在准2~3a,5~7a,10~13a和18~22a的波动周期。通过对降水EMD分量与径流EMD分量的相关性分析表明降水与径流的整体变化趋势基本一致。(4)采用Morlet复小波变化法来展现泾河流域年径流、年降水序列在时域、频域中具有的多层次时间结构,以便分析水文气象序列的多时间尺度变化和周期变化。对泾河流域68年年径流量变化进行多尺度小波分析,得出存在13a及32a左右的主要周期变化和3a及21a左右的次要周期变化决定着泾河流域年径流量的丰枯变化特征,2000年以后几年内泾河流域年径流量仍处于偏枯期。通过小波变换得到泾河流域年降水序列可能存在3a,6a,11a和18a左右的周期变化。(5)基于EMD的BP神经网络模型是利用EMD对年降水序列进行平稳化处理,得到一组本征模态分量IMF和一个剩余量RES,通过以分量为输入,以相应年径流序列为输出,用BP神经网络对年径流序列进行预测。小波人工神经网络预测模型是基于小波分析而构成的具有BP神经网络思想的模型,将小波分解与BP神经网络有机地结合起来,充分发挥了小波分析的多分辨率功能和人工神经网络的非线性逼近功能。将两种模型分别对泾河流域年径流量进行预测分析,预测相对误差<10%的合格率为66.7%和80%,相对误差<20%的合格率为83.3%和93.3%,结果表明这两种模型对年径流预测精度较高,可用于泾河流域年径流预测。更多还原

【Abstract】 Water Resources is one of the basic elements of supporting human life and economic growth and sustainable development, among them, the change of runoff dominates the whole system changes, due to climate, social development and human activities combined effects, the variation of runoff is complex and shows the unpredictable, random, multi-time scale and nonlinear characteristic. The paper depended on National Key Technology R&D Program(2006BAD11B04) and National Natural Science Foundation of China(50879072), studied the variation characteristics of the hydrological and meteorological series systematically in Jinghe River basin with traditional methods, empirical mode decomposition and wavelet analysis theory and method, in this basis, the forecast model was established to predict the runoff series with empirical mode decomposition, wavelet decomposition and BP artificial neural network. The main research results are as follows:(1) In Jinghe River basin, the annual runoff distribution is extremely uneven and variant, the annual runoff distribution shows a single peak type. The centralized degree of runoff is high, and the annual maximum runoff appears in August. The interannual variations of runoff present the relationship of alternating between wet and dry.(2) After analyzing the trend of annual runoff series in Jinghe River basin with moving average and Kendall rank correlation methods, the annual runoff has a significant downward trend and the average decrease has per decade 113 million stere. Meanwhile, analyzing meteorological series which are the factors causing the changes of runoff, it found that the changes of interannual precipitation affect runoff clearly, and the change tendency of runoff is similar to that of precipitation, but the overall consistency of the runoff variation and temperature variation is not obvious.(3) Using EMD, multiple-time-scale analysis of annual runoff and precipitation of 4 typical hydrologic stations, including Pingliang, Huanxian, Xifengzhen and Changwu stations in Jinghe River basin, The results of investigation showed that the runoff time series has periods that about 2~3, 6~8, 10~12 and 21 years. The precipitation series has periods that about 2~3, 5~7, 10~13 and 18~22 years, the change tendency of runoff is similar to that of precipitation by analyzing the correlation of precipitation EMD components and runoff EMD components.(4) After using complex Morlet wavelet transform of annual runoff and annual precipitation series in Jinghe River basin, and analyzing periodic variation and multi-time scales of hydrological and meteorological time series. The result shows that the series in time-frequency domain have multiple time scales structures, the major periods of runoff time series are about 13 years and 32 years, the minor periods are about 3 years and 21 years, and the annual runoff within the next few years is still in the dry period after 2000. The precipitation series has periods that about 3 years, 6years, 11 years and 18 years.(5) EMD-BP neural network model uses EMD method on the precipitation series in the smooth processing, and gets a set of intrinsic mode component(IMF) and a residual amount(RES), through components as the input, the corresponding annual runoff series for the output, with BP neural network to forecast annual runoff series. WANN prediction model is a neural network model based on wavelet analysis, the model integrates the wavelet transform and neural network organically and fully plays the advantages of them. The two models forecasted annual runoff of Jinghe River basin, the rate of relative error less than 10% is 66.7% and 80%,the rate of relative error less than 20% is 83.3% and 93.3%. The result shows that the two models predict higher accuracy and can be used to forecast annual runoff in Jinghe River basin.更多还原