基于物理成因的中长期径流预测模型研究

可靠的中长期径流预测对水资源开发等具有重要意义。为此,筛选了影响径流的主要物理因子,引入极端梯度提升(XGBoost)算法构建中长期径流预测模型,通过纳什效率系数评价模型精度,并与多元线性回归模型(LR)、梯度提升决策树模型(GBDT)进行比较。实例应用结果表明,该模型对月径流过程的预测精度较高,训练期和验证期的纳什效率系数均值分别达到了0.9和0.7,且泛化能力优于GBDT模型和LR模型,用于中长期径流预测具有一定的可靠性和稳定性。

Research on Medium and Long-Term Runoff Forecasting Based on Physical Process

Medium and long term runoff forecasting is important for water resources development and utilization. This study established a medium and long-term runoff forecasting model based on extreme gradient boosting algorithm (XGBoost) after identification of controlling physical factors of runoff. Nash Sutcliffe efficiency coefficient (NSE) was adopted as the evaluation criteria. Consequently, comparative study was conducted between the XGBoost model and the multiple linear regression (LR) model as well as the gradient boosting decision tree model. The case results show that the model has a good performance for monthly runoff forecasting. The mean NSE during the training and verification period is 0.9 and 0.7, respectively. Furthermore, the model generalization capability is better than that of the GBDT and the LR models. In summary, this model has certain reliability and robustness in medium and long-term runoff forecasting.