基于Copula熵-随机森林的中长期径流预报研究

Research on medium and long-term runoff forecast based on Copula entropy-random forest model

预测因子作为中长期预报模型的输入项,是影响预报结果精度的关键要素。为进一步提高预报精度,提出了一种Copula熵与随机森林模型相结合的中长期径流预报方法。该方法首先采用Copula熵指标对预测因子进行筛选,然后将选取的预测因子作为输入项,导入随机森林模型中对月径流进行相应预测。将该方法应用于汉江流域丹江口水库的逐月入库径流预报中,并与相关系数筛选法进行对比。结果表明:基于Copula熵指标筛选出的预测因子对应的模拟结果具有更高的精度,尤其对于汛期而言,其模拟值与实测值的拟合优度显著优于比选方法,说明其筛选出的预测因子具有更好的合理性。

As the key input of hydrological model for medium and long-term runoff forecast,forecast factors play an important role in improving the forecast accuracy.In order to further improve the accuracy of forecast results,we proposed a medium and long-term runoff forecast method combined with the Copula entropy and random forest model.For this method,the forecast factors were first selected based on the Copula entropy index,and the selected factors were used as input items and imported into the random forest model to simulate and forecast monthly runoff series.Finally,this method was applied to predict the monthly runoff series of the Danjiangkou Reservoir in Hanjiang River Basin,and compared with the correlation coefficient screening selection method.The results showed that the forecast results corresponding to the Copula entropy theory had a higher accuracy in forecasting monthly runoff series.Especially for flood season,the fitting effect of the simulated value and the measured value of this method was significantly better than that of the correlation coefficient method,indicating that the screened forecast factors are more reasonable.