当前位置: X-MOL 学术Front. Earth Sci. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Probabilistic forecasting based on ensemble forecasts and EMOS method for TGR inflow
Frontiers of Earth Science ( IF 1.8 ) Pub Date : 2019-12-28 , DOI: 10.1007/s11707-019-0773-9
Yixuan Zhong , Shenglian Guo , Feng Xiong , Dedi Liu , Huanhuan Ba , Xushu Wu

Probabilistic inflow forecasts can quantify the uncertainty involved in the forecasting process and provide useful risk information for reservoir management. This study proposed a probabilistic inflow forecasting scheme for the Three Gorges Reservoir (TGR) at 1–3 d lead times. The post-processing method Ensemble Model Output Statistics (EMOS) is used to derive probabilistic inflow forecasts from ensemble inflow forecasts. Considering the inherent skew feature of the inflow series, lognormal and gamma distributions are used as EMOS predictive distributions in addition to conventional normal distribution. Results show that TGR’s ensemble inflow forecasts at 1–3 d lead times perform well with high model efficiency and small mean absolute error. Underestimation of forecasting uncertainty is observed for the raw ensemble inflow forecasts with biased probability integral transform (PIT) histograms. The three EMOS probabilistic forecasts outperform the raw ensemble forecasts in terms of both deterministic and probabilistic performance at 1–3 d lead times. The EMOS results are more reliable with much flatter PIT histograms, coverage rates approximate to the nominal coverage 89.47% and satisfactory sharpness. Results also show that EMOS with gamma distribution is superior to normal and lognormal distributions. This research can provide reliable probabilistic inflow forecasts without much variation of TGR’s operational inflow forecasting procedure.

中文翻译:

基于集合预测和EMOS方法的TGR流入概率预测

概率流入预测可以量化预测过程中涉及的不确定性,并为水库管理提供有用的风险信息。这项研究提出了三峡水库(TGR)在1-3 d提前期的概率流量预测方案。后处理方法“集成模型输出统计信息(EMOS)”用于从集合流入预测中得出概率流入预测。考虑到流入序列的固有偏斜特征,除常规正态分布外,对数正态分布和伽马分布还用作EMOS预测分布。结果表明,TGR在1到3 d提前期的总体流入预测效果良好,模型效率高,平均绝对误差小。对于带有偏倚概率积分变换(PIT)直方图的原始集合流入预测,观察到预测不确定性的低估。在1到3 d的交付周期内,三个EMOS概率预测在确定性和概率性能方面都优于原始整体预测。通过更平坦的PIT直方图,EMOS结果更加可靠,覆盖率接近89.47%的标称覆盖率,并且清晰度令人满意。结果还表明,具有伽马分布的EMOS优于正态分布和对数正态分布。这项研究可以提供可靠的概率流入预测,而TGR的业务流入预测程序不会有太大变化。在1到3 d的交付周期内,三个EMOS概率预测在确定性和概率性能方面都优于原始整体预测。通过更平坦的PIT直方图,EMOS结果更加可靠,覆盖率接近89.47%的标称覆盖率,并且清晰度令人满意。结果还表明,具有伽马分布的EMOS优于正态分布和对数正态分布。这项研究可以提供可靠的概率流入预测,而TGR的业务流入预测程序不会有太大变化。在1到3 d的交付周期内,三个EMOS概率预测在确定性和概率性能方面都优于原始整体预测。通过更平坦的PIT直方图,EMOS结果更加可靠,覆盖率接近89.47%的标称覆盖率,并且清晰度令人满意。结果还表明,具有伽马分布的EMOS优于正态分布和对数正态分布。这项研究可以提供可靠的概率流入预测,而TGR的业务流入预测程序不会有太大变化。结果还表明,具有伽马分布的EMOS优于正态分布和对数正态分布。这项研究可以提供可靠的概率流入预测,而TGR的业务流入预测程序不会有太大变化。结果还表明,具有伽马分布的EMOS优于正态分布和对数正态分布。这项研究可以提供可靠的概率流入预测,而TGR的业务流入预测程序不会有太大变化。
更新日期:2019-12-28
down
wechat
bug