| 刘昱辰,刘佳,刘录三,李传哲,王瑜.基于LSTM实时校正的WRF/WRF-Hydro耦合径流预报[J].水利学报,2023,54(11):1334-1346 |
| 基于LSTM实时校正的WRF/WRF-Hydro耦合径流预报 |
| WRF/WRF-Hydro coupled streamflow forecasting based on real-time updating using LSTM |
| 投稿时间：2023-02-22 |
| DOI：10.13243/j.cnki.slxb.20230099 |
| 中文关键词: LSTM 实时校正 WRF/WRF-Hydro耦合系统 径流预报 数据同化 |
| 英文关键词: LSTM real-time updating WRF/WRF-Hydro runoff forecast data assimilation |
| 基金项目:国家自然科学基金项目（51822906）；长江生态环境保护修复联合研究（第二期）（2022-LHYJ-02-0601） |
|
| 摘要点击次数: 1865 |
| 全文下载次数: 845 |
| 中文摘要: |
| 为改进WRF/WRF-Hydro陆气耦合系统的径流预报效果，减小耦合系统在峰现时间、洪峰流量预报上的误差，本文在使用变分数据同化技术充分降低预报降雨误差水平的基础上，采用长短期记忆人工神经网络LSTM对WRF/WRF-Hydro耦合系统的径流预报过程开展了实时校正研究，并与自回归滑动平均模型ARMA实时校正结果进行对比。研究结果表明，通过数据同化技术可有效提升WRF模式降雨预报精度，降低WRF-Hydro模式的输入误差，但径流预报准确性仍有待提升。对比LSTM和ARMA两种实时校正模型对耦合径流预报结果的实时校正：在前3 h预见期，两种模型在中国北方半湿润、半干旱地区山区小流域6场典型洪水预报中的表现基本接近，除场次4外，LSTM和ARMA两种模型在3 h预见期的衰减速率分别为2.04~23.08和9.18~36.47，随着预见期的延长，LSTM径流预报精度的衰减速度在整体上慢于ARMA模型，预报效果优于ARMA模型。 |
| 英文摘要: |
| In order to improve the runoff prediction performance of the WRF/WRF-Hydro coupled atmospheric-hydrologic systems and reduce the errors in peak time and flood peak flow prediction, this study uses variational data assimilation technology to reduce the rainfall prediction error, at the same time, a real-time correction study on the runoff prediction process of the WRF/WRF-Hydro system is conducted using the long short-term memory(LSTM), and compare the real-time correction results with the autoregressive moving average model(ARMA).The research results indicate that data assimilation technology can effectively improve the accuracy of WRF model rainfall prediction and reduce the input error of WRF-Hydro model, but the accuracy of runoff prediction still needs to be improved.Comparing the real-time correction results of LSTM and ARMA models for runoff forecasting, it was found that during the first three hours of the foresight period, the performance of the two models is basically similar of small watersheds in semi humid and semi-arid mountainous areas in northern China.Except for Event 4, the attenuation rates of LSTM and ARMA models in the three hours of the foresight period are 2.04~23.08 and 9.18~36.47, respectively.As the foresight period extends, the decay rate of LSTM runoff prediction accuracy is generally slower than the ARMA model, and the prediction effect is better than the ARMA model. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|