基于机器学习的洪水预报实时校正  

Real-time Correction of Flood Forecasting Based on Machine Learning

作　　者：衣学军 汤岭 李致家[2] 盛奕华 姚成[2] 杜若愚 YI Xue-jun;TANG Ling;LI Zhi-jia;SHENG Yi-hua;YAO Cheng;DU Ruo-yu(Shandong Provincial Hydrological Center,Jinan 250000,China;College of Hydrology and Water Resources,Hohai University,Nanjing 210098,China)

机构地区：[1]山东省水文中心,山东济南250000 [2]河海大学水文水资源学院,江苏南京210098

出　　处：《水电能源科学》2023年第12期78-81,67,共5页Water Resources and Power

基　　金：国家自然科学基金项目(52079035)。

摘　　要：为了提高临沂流域水文模型的实时洪水预报精度,基于临沂流域的下垫面特征,建立了临沂流域的TOPKAPI网格模型,采用BP神经网络和LSTM模型对TOPKAPI模型模拟结果在不同预见期内进行了校正,在此基础上使用了堆叠方法并选用Transformer模型作为二级学习器,对BP和LSTM的校正结果进行了二次学习。结果表明,经过BP和LSTM模型的实时校正,TOPKAPI模型模拟精度得到了明显提高,预见期越短,校正效果越好;在经过堆叠方法进行二次学习后,校正效果最佳,可有效提升临沂流域洪水预报精度。In order to enhance the real-time flood forecasting accuracy in the Linyi River Basin,a TOPKAPI grid model was developed based on the underlying surface characteristics of the Linyi River Basin.The TOPKAPI model simulation results were corrected at different lead times using BP neural networks and LSTM models.Furthermore,a stacking approach was applied,employing the Transformer model as a secondary learning tool to refine the corrections made by BP and LSTM.The results indicate that after real-time correction with the BP and LSTM models,the improvement of the simulation accuracy of the TOPKAPI model is obvious,with better correction results for shorter lead times.Following the stacking method for secondary learning,the correction results is the best,effectively enhancing the flood forecasting accuracy in the Linyi River Basin.

关 键 词：TOPKAPI模型 实时校正 BP神经网络 LSTM模型 洪水预报 临沂流域 

分 类 号：TV122[水利工程—水文学及水资源] P338[天文地球—水文科学]