黑河出山径流的非线性特征分析

Nonlinear Features of the Runoff from Mountain Areas of the Heihe River, Qilian Mountains

应用非线性动力学的理论和方法 ,对黑河出山径流的非线性特征进行了分析 .结果表明 ,黑河月出山径流的年内分布、年平均流量的一次峰、谷变化符合单重或双重威布尔分布 ,并具有自相似性质 .黑河出山径流多年变化在相空间中的运动轨迹收缩到一个约为 4 .32维的吸引子上 ,而描述流量的动力方程需要 8个独立变量 .黑河出山径流的非线性特征还表现在对内部结构为非线性函数的输入输出模型的良好应用上 ,如GRNN神经网络模型、非线性回归模型等 .

Using several methods of nonlinear dynamics theory, the nonlinear features of the runoff from mountainous watersheds of the main Heihe River have been described, and several hydrologic models of main Heihe River have also been developed. The variation of monthly mean runoff and the variation of annual runoff in the Heihe River agree with Weibull Distribution. Using inserting space methods, it is found that the correlation dimensionality of annual runoff is 4.32, and the least insertion dimensionality is 8. Wavelet transformation method can be used to analyze the periodic feature of annual runoff. Gray system model can be used to analyze the linear tendency of relative long series, such as the runoff in the Heihe River. The analyses show that annual runoff and monthly runoff are all increasing, especially from May to August. Recently the artificial neural network has been used for runoff forecasting extensively, of which the generalized regression neural network (GRNN) has perfect result. In this paper, some new points of this method are introduced. The input and output series may be the primordial measured one, or the one transformed to standard forms, or the one decomposed to wavelet approximation coefficient and detail coefficient one. For using GRNN model to predict runoff, several climate scenarios are supposed according to the global warming theory, or the measured series in i k time is used as input datum. These methods have been used in the runoff model of the Heihe River in this paper and their results are perfect. Time series decomposition model has also been talked of in this paper. Firstly the tendency series is separated from the ordinal runoff series by using Gray System method, and then the periodic parts of the residual series is obtained according to wavelet transformation method, and at last the stochastic part is fitted. The sum of the elongation series of each separated series should be the forecasting results of runoff. The runoff forecasting results are also perfect. The monthly mean runoff at the pass and the monthly precipitation in the Heihe River mountainous watershed has perfectly linear relationship, and the relationship between the monthly mean runoff and the monthly mean air temperature is exponential. From the relationship between the monthly mean runoff and the monthly mean air temperature, base flow can be separated, and then the effect of meltwater of glacier and snow and the liquid precipitation in the runoff formation can be worked out. Using nonlinear dynamics method, a model including monthly mean runoff, monthly precipitation and monthly air temperature of the Heihe River mountainous watershed is put forward. Using the model, the monthly runoff also can be predicted.