三峡水库中长期径流预测及不确定性分析研究

Medium and Long term Runoff Forecasting and Uncertainty Analysis for the Three Gorges Reservoir

以三峡水库为例,采用轻量梯度提升树(LGB)对1965-2016年逐月流量过程开展了中长期预测研究,并在其基础上结合水文不确定性处理器(HUP)对预测结果的不确定性进行了定量分析,得到以下结论:(1)各精度指标结果表明,与GBDT模型相比,LGB模型在率定期及验证期均具有更高的精度,能较好地应用于三峡水库中长期径流预测中;(2)基于HUP模型提取的倾向值预报(Q50值)精度显著高于LGB模型结果,尤其对于汛期径流的提升效果更为显著;(3)90%不确定性区间在率定期及验证期均能以较窄带宽覆盖多数实测点据,且相对偏移程度较小,说明不确定性分析结果是合理可靠的。

In this paper,the Lightweight Gradient Boosting Tree Model(LGB)is employed to simulate and forecast monthly runoff sequences of the Three Gorges Reservoir,and the forecasting uncertainties are analyzed by using the Hydrological Uncertainty Processor(HUP).The findings of this paper are as follows:(1) the comparison between predictions of GBDT and LGB models indicates that the LGB model has higher accuracy than the GBDT model for both calibration and validation periods;(2) the preferred forecasts(Q50 values)generated by the HUP model present a higher accuracy than the deterministic prediction generated by the LGB model,especially for the flood season;(3) the90% uncertainty confidence interval with a narrow bandwidth for both calibration and validation periods can cover most observed points,which suggests that the result of uncertainty analysis is accurate and reasonable.