基于CEEMDAN-LightGBM模型的洪水预测研究  

Research on Flood Prediction Based on CEEMDAN⁃LightGBM Model

作　　者：王军 张宇航 崔云烨 李怡豪 吕鹏祥 WANG Jun;ZHANG Yuhang;CUI Yunye;LI Yihao;LYU Pengxiang(Institute of Big Data Science,Zhengzhou University of Aeronautics,Zhengzhou 450015,China;Henan Daily,Zhengzhou 450014,China)

机构地区：[1]郑州航空管理学院大数据科学研究院,河南郑州450015 [2]河南日报社,河南郑州450014

出　　处：《人民黄河》2024年第9期99-105,共7页Yellow River

基　　金：河南省科技攻关项目(222102210292);河南省科技智库调研项目(HNKJZK-2021-61C)。

摘　　要：为了应对暴雨可能引发的洪涝灾害风险,基于黄河利津水文站监测的水文等数据,以LightGBM为基准模型,运用经自适应噪声完备集合经验模态分解(CEEMDAN)算法优化后的CEEMDAN-LightGBM模型对水位进行预测,并将其与长短期记忆网络(LSTM)模型、LightGBM模型的预测效果进行对比。以2个气候条件不同的黄河水文站(利津、花园口)的水文数据为原始数据集输入CEEMDAN-LightGBM模型,验证模型的适应性和稳定性。结果表明:CEEMDAN-LightGBM模型在水位预测方面表现出优越的性能,相较于LSTM、LightGBM模型,该模型的E_(MA)分别减小了46.08%、9.95%,E_(RMS)分别减小了33.01%、43.01%,E_(MAP)分别减小了94.99%、3.82%,R^(2)分别增大了30.48%、7.58%。CEEMDAN-LightGBM模型还能预测流量这一重要水文特征,为模型预测洪水发生提供更有力的判断依据。对比CEEMDAN-LightGBM模型预测花园口水文站与利津水文站的水位和流量效果,除预测两站水位的E_(MAP)值相差23.64%外,E_(MA)值、E_(MAP)和E_(RMS)值相差均不超过10%,R^(2)相差不超过2%。In order to deal with the risk of flood disaster caused by rainstorm,based on hydrological data monitored by Lijin Hydrological Station on the Yellow River,the LightGBM model was taken as the benchmark model,the CEEMDAN⁃LightGBM model optimized by the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)algorithm was used to predict water levels,and compared its prediction performance with the Long Short⁃Term Memory(LSTM)model and LightGBM model.The hydrological data from two Yellow River hydrological stations(Lijin and Huayuankou)with different climatic conditions were input into the CEEMDAN⁃LightGBM model as original dataset to verify the adaptability and stability of the model.The results show that the CEEMDAN⁃LightGBM model exhibits superior performance in water level prediction.Comparing with the LSTM and LightGBM models,the model's E_(MA) decrease by 46.08%and 9.95%,E_(RMS) decrease by 33.01%and 43.01%,E_(MAP) decrease by 94.99%and 3.82%,and R^(2) increase by 30.48%and 7.58%,respectively.The CEEMDAN⁃LightGBM model can also predict the important hydrological feature of flow,providing stronger judgment basis for the model to predict flood occurrence.Comparing with the CEEMDAN⁃LightGBM model for predicting the water level and flow of Huayuankou hydrological station and Lijin hydrological station,except for the predicted difference of 23.64%in E_(MAP) values between the two stationswater levels,the difference between E_(MAP) values and E_(RMS) values does not exceed 10%,and the difference in R^(2) does not exceed 2%.

关 键 词：洪水预测 LightGBM模型 CEEMDAN算法 CEEMDAN-LightGBM模型 LSTM模型 利津水文站 花园口水文站 

分 类 号：P333[天文地球—水文科学] TP183[水利工程—水文学及水资源] TV882.1[天文地球—地球物理学]