基于EMD组合模型的径流多尺度预测

受全球气候变化与人类活动影响,径流序列愈发呈现出非稳态与非线性特征,为降低由此而引发的预报误差,充分发挥不同模型对提高径流预测精度的优势,针对传统径流预报模型的单一性,以干旱区典型内陆河玛纳斯河为例,采用经验模态分解(EMD)提取径流序列中具有物理含义的信号,得到不同时间尺度的多个固有模态函数(IMF)及1个趋势项,利...

Multiscale runoff prediction based on the EMD combined model

Affected by global climate change and human activities, runoff sequences increasingly show unsteady and non-linear characteristics. In order to reduce the forecast errors caused by these runoff characteristics, we took full advantages of different models to improve the accuracy of runoff prediction traditionally done by the single model approach. Taking the Manas River, a typical inland river in the arid area as an example, we used empirical mode decomposition (EMD) to extract physically meaningful signals from the runoff sequence to obtain multiple intrinsic mode functions (IMF) at different time scales and a trend indicator. We then used the ARIMA and GRNN models to simulate the IMF components at different time scales and analyze the future runoff changing trends. Next, we used the multiple linear regression, Spearman correlation coefficient and average influence value methods to screen the atmospheric circulation factors and them used as inputs to the neural network model; we then constructed the combined model according to the local frequency characteristics of the sub-sequences. Finally, we reconstructed the prediction results of each IMF component to obtain the runoff prediction. However, a single evaluation index cannot fully evaluate the accuracy of the prediction model. In this paper, we constructed the TOPSIS evaluation model to quantitatively evaluate the runoff prediction model and objectively evaluate the model’s superiority. The results show that using EMD can effectively extract the multi-timescale signals hidden in the runoff sequence, and the trend indicator indicated that the Manas River runoff is on the rise. Using EMD could improve the passing rate of the ARIMA model by 25%; but the relative errors of high frequency components IMF1, IMF2 and IMF3 in the ARIMA model were more than 70%, indicating the prediction results are not ideal. In the GRNN model, selected predictors could effectively improve the model accuracy, and the predictors selected by the MIV method were shown to be most suitable for the Manas River. Overall, the GRNN-EMD-ARIMA combination model had the highest passing rate, and the TOPSIS model had the highest score. The prediction results can be used as a scientific basis for water resources planning and dispatching, and the modeling ideas can also provide new ways to optimize runoff prediction models.