中文摘要：

以黄河上游兰州站月径流序列为研究对象,在介绍相空间重构原理的基础上,探讨了混沌分析的主要定量指标：饱和关联维数d,最大Lyapunov指数λ和Kolmogorov熵k。得到该时间序列的最佳嵌入滞时τ=3个月,饱和关联维数d=3.230,最小嵌入维数m=12和最大Lyapunov指数λ=0.241,Kolmogorov熵k=0.14,指出该序列的预测时限为4个月。在此基础上建立了基于混沌特性的支持向量机径流预测模型,用1995-2004年的月径流数据进行仿真试验后,用2005年1-12月的径流数据作为预测检验,结果表明,该模型可用于混沌时间序列的月径流预测,并验证了由最大Lyapunov指数所确定的可预报时限为4个月的结论。

英文摘要：

Based on introducing the phase space reconstruction theory of chaotic time series, the main quantitative indexes of saturation correlation dimension d, maximal Lyapunov exponent 3, and Kolmogorov entropy k are discussed for chaotic analysis. The saturation correlation dimensional, minimum embedding dimensionm, optimal built in delay time r, maximal Lyapunov exponent 3, and Kolmogorov entropy k are calculated, that is d = 3. 230, m = 12, r = 3 and 3, = 0.241, k = 0.14. The support vector machine model of mid-long prediction of streamflow is established based on chaos characteristics. After the leaning and training simulation experiments are completed with the statistic data of runoff from 1995 to 2004, the model is verified by use of runoff data of 2005. The result shows that the C-SVM model is effective for monthly runoff prediction, and verifies the conclusion that the forecasting time for this runoff should not exceed 4 months.