基于多嵌入维数的径流组合预测

Combined Prediction of Runoff in Multi-dimension Embedding Phase Spaces

针对理论混沌参数计算方法很难得到实际径流时间序列的最佳嵌入维数,采用多嵌入维数组合预测对径流时间序列预测方法进行改进。利用Lyapunov指数预测模型分别计算了一系列不同嵌入维数情况下的预测值,根据预测效果对结果进行选择,并利用组合原理对其进行集成得到最后的预测值。实例分析表明,相对于单嵌入维数法,多嵌入维数组合预测方法可以综合利用不同相空间中的有用信息,提高径流时间序列预测的精度。

It is often difficult to calculate the best embedding dimension for the real runoff time series when the theoretical methods are used,so the authors used the combined prediction in multi-dimension embedding phase spaces to improve the chaotic prediction of runoff time series. The proposed procedure for this is, first, predict the same chaotic time series of runoff in different dimension embedding phase spaces using Lyapunov exponents; and then, choose the relative good prediction results and calculate the combined prediction value. The results of actual runoff prediction show that the proposed method could use information synthetically in multi-dimension embedding phase spaces, and effectively improve the prediction accuracy.