| 陈小泽,王忠静,刘丹,石羽佳,KANGBoosik.径流序列两段动态分解-预测-重构中长期预报模型[J].水利学报,2025,56(7):933-944 |
| 径流序列两段动态分解-预测-重构中长期预报模型 |
| A two-stage dynamic decomposition-prediction-reconstruction model for medium-long term runoff forecasting |
| 投稿时间：2024-05-27 修订日期：2025-05-07 |
| DOI：10.13243/j.cnki.slxb.20240319 |
| 中文关键词: 径流预报 周期趋势分解 变分模态分解 Informer 永定河 |
| 英文关键词: runoff forecast STL VMD Informer Yongding River |
| 基金项目:国家重点研发计划项目(2022YFE0101100); 国家自然科学基金项目(52300246,42307558) |
|
| 摘要点击次数: 72 |
| 全文下载次数: 51 |
| 中文摘要: |
| 中长期径流预报是水资源优化调度的基础支撑。随着全球气候变化加剧，径流时间序列变异性呈增加趋势，基于时间序列分析方法的径流预报难度进一步增大。为克服这一困难、增强预报能力，本文通过局部加权回归周期趋势分解(STL)、变分模态分解(VMD)与带有知识蒸馏的深度学习组合，构建了基于STL-VMD-Informer的两段动态分解-预测-重构的中长期径流组合预报模型。该模型在永定河流域石匣里水文站断面应用表明，在预见期1、3、6个月时，其纳什效率系数NSE分别可达到0.897、0.843和0.796,验证了所提出的方法可较好地分离时间序列的变异性，延长预见期，提高预报精度。本方法对改进数据驱动的径流序列中长期预报方法和策略有积极的意义。 |
| 英文摘要: |
| Medium-long term runoff forecasting is a critical foundation for optimizing water resource management.With the intensification of climate change, the variability of runoff has increased, making the forecasting more challenging.To improve the accuracy of runoff forecasts and extend the forecast period, in this paper, by integrated Seasonal and Trend decomposition using Loess（STL）,Variational Mode Decomposition（VMD） and the deep learning model Informer, a two-stage dynamic decomposition-prediction-reconstruction model（STL-VMD-Informer） for medium-long term runoff forecast was development.The application at the Shixiali hydrological station of Yongding River Basin shows that the forecast Nash-Sutcliffe efficiency（NSE） reaches 0.897,0.843,and 0.796 for prediction periods of 1,3,and 6 months, respectively, indicating the proposed method with the potential in separating the time series, extending the forecasting horizon, and improving forecast accuracy.The approach will benefit to the medium-long term runoff forecasting. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|