基于机器学习算法的万家寨水库排沙预测研究

Research on Sand Discharge Prediction of Wanjiazhai Reservoir Based on Machine Learning Algorithms

为克服水库排沙多因素、非线性复杂关系建立难题,实现水库排沙准确预测,利用万家寨水库2002~2020年水沙系列数据,基于XGBoost、KNN、SVR、RF四种机器学习算法分别预测分析水库出库含沙量。结果表明,利用机器学习算法可有效预测综合考虑不同影响因素的水库排沙;不同机器学习算法在水库排沙预测的适用性有所不同,对比之下,基于RF算法建立的水库排沙预测模型的决定系数R2最高为0.934 9,平均绝对误差及均方根误差均最小,分别为2.974、4.886,其预测效果更优于其他三种算法。研究成果可为水库排沙精确预测及调度方案优化提供参考。

In order to overcome the difficult problem of establishing multi-factor and non-linear complex relationship of reservoir sand discharge and achieve its accurate prediction, four machine learning algorithms including XGBoost, KNN, SVR and RF were used to predict and analyze the sand content of reservoir outflow based on the series data of Wanjiazhai reservoir from 2002 to 2020, respectively. The results show that the use of machine learning algorithms can effectively realize the reservoir discharge prediction considering different influencing factors. The applicability of different machine learning algorithms in reservoir discharge prediction varies. In comparison, the highest coefficient of determination R2 of the reservoir discharge prediction model based on RF algorithm is 0.9349, and the corresponding average absolute error and root mean square error are the smallest, which are 2.974 and 4.886, respectively. The prediction effect of the RF algorithm is better than the other three algorithms. The proposed method can provide a theoretical basis for accurate prediction of reservoir sand discharge and optimization of scheduling scheme.