基于深度学习和信号分解的北方寒区河流开河日期预报  

Forecasting break-up date of river ice in northern China based on deep learning and signal decomposition technology

作　　者：丁红 王伟泽 杨泽凡[1] 刘欢[1] 胡鹏[1] DING Hong;WANG Weize;YANG Zefan;LIU Huan;HU Peng(State Key Laboratory of Simulation and Regulation of Water Cycle in River Basin,China Institute of Water Resources and Hydropower Research,Beijing 100038,China;School of Civil Engineering and Architecture,Xi’an University of Technology,Xi’an 710048,China)

机构地区：[1]中国水利水电科学研究院流域水循环模拟与调控国家重点实验室,北京100038 [2]西安理工大学土木建筑工程学院,陕西西安710048

出　　处：《水利学报》2024年第5期577-585,共9页Journal of Hydraulic Engineering

基　　金：国家重点研发计划课题(2022YFF1300902);国家自然科学基金项目(52122902,42001040);流域水循环模拟与调控国家重点实验室自主研究课题(SKL2022ZD01);中国水利水电科学研究院基本科研业务费项目(WR0145B022021)。

摘　　要：中国北方寒区河流春季开河时易产生冰凌现象,威胁涉河水工建筑物的安全。准确地预测寒区河流开河日期可为防凌指挥、调度决策提供重要参考依据。本文基于中国北方典型寒区-黑龙江省的5个代表水文站近60年的历史开河日期序列,采用完全自适应集合经验模态分解(CEEMDAN)技术和深度学习长短期记忆模型(LSTM)方法构建河流开河日期预报的耦合模型,以期提高河流开河日期预报的精度。结果表明:本研究构建的开河日期预报耦合模型(CEEMDAN-LSTM)预测精度明显优于单一深度学习方法(LSTM)计算结果;与LSTM相比,CEEMDAN-LSTM可将开河日期预报的平均绝对误差从2.51 d降低至1.20 d,合格率从91.59%提高至100%。验证期平均绝对误差从3.85 d降低至1.65 d,合格率从88%提高至96%。因此,所构建的开河日期预报耦合模型具有较高的预报精度,可为我国北方寒区春季防凌指挥和调度提供技术支持。Ice floods occasionally occur during river ice breaking up in northern China in spring,threatening the safety of hydraulic structures.Forecasting the break-up date of river ice(BUDRI)accurately is an important reference for anti-flooding command and dispatching decision-making during ice breaking period.For forecasting the BUDRI in northern China,the observed break-up date series of river ice of 5 representative hydrological stations in Heilongjiang province located in northern China was selected,and the Complementary Ensemble Empirical Mode Decomposition with Adaptive Noise technology and deep learning model Long Short Term Memory(CEEMDAN-LSTM)was used to forecast the BUDRI.The results show that the forecast accuracy of CEEMDAN-LSTM,compared with LSTM,had been significantly improved with the mean absolute error reduced from 2.51 d to 1.20 d,the qualification rate increased from 91.59%to 100%in the training period.and the mean absolute error reduced from 3.85 d to 1.65 d,the qualification rate increased from 88%to 96%in the validation period.The CEEMDAN-LSTM performed well in forecasting the BUDRI in northern China,which can provide important information for command,dispatch,and decision-making of ice flood control.

关 键 词：河流开河日期 信号分解技术 深度学习 预报方法 北方寒区 

分 类 号：P338.4[天文地球—水文科学]