基于深度学习的山丘区中小河流洪水预报误差校正方法  

Error Correction Method for Flood Forecasting of Medium and Small Catchments in Hilly Areas Based on Deep Learning

作　　者：白云鹏 郑爱民 王文川[2] 赵延伟 徐冬梅[2] 杜昀宸 BAI Yunpeng;ZHENG Aimin;WANG Wenchuan;ZHAO Yanwei;XU Dongmei;DU Yunchen(Hydrological Survey and Research Center of Hebei Province,Shijiazhuang 050023,China;School of Water Resources,North China University of Water Resources and Electric Power,Zhengzhou 450046,China)

机构地区：[1]河北省水文勘测研究中心,河北石家庄050023 [2]华北水利水电大学水资源学院,河南郑州450046

出　　处：《水文》2025年第1期37-44,共8页Journal of China Hydrology

摘　　要：中小河流产汇流情况复杂,洪水预报难度很大。为了提高中小河流洪水预报的精度,以河北省邢台市坡底流域为研究对象,分别基于蓄满产流和混合产流模式构建分布式模型进行降雨径流模拟。分别采用LSTM、Transformer、Transformer+LSTM叠加模型(TFLS)构建校正模型,采用差分进化算法对超参数进行优化。以实测降雨和分布式模型模拟结果为输入,对各时段的残差进行拟合,进而对径流模拟结果进行校正。研究结果表明,在17场洪水模拟结果中,混合产流模型表现优于蓄满产流。与混合产流模型相比,经TFLS校正后的模型洪峰误差不超过20%的场次从9场增加至12场,占全部场次的70.6%,确定性系数不低于0.8的场次从5场增加到9场,占比为52.9%。TFLS模型在流量不超过500 m^(3)/s时的校正效果优于LSTM和Transformer模型,LSTM模型对流量在500 m^(3)/s及以上的校正效果略优于其它模型。Due to the complex runoff and concentration situation,flood forecasting for small and medium-sized catchments is very difficult.In order to improve the accuracy of flood forecasting,this study constructed a distributed model for flood forecasting based on full storage runoff and mixed runoff model respectively,taking the PoDi Basin in Xingtai City,Hebei Province as a study case.LSTM,Transformer,Transformer and LSTM combining models(TFLS)were used to construct the correction model,and the differential evolution(DE)algorithm was used to optimize the hyperparameters.Taking the observed rainfall and distributed model simulation results as input,the residuals of observation flow are fitted,and then the simulation results are corrected.The re-sults show that in the 17 flood simulation results,the mixed runoff model performs better than the full-storage runoff model,and the correction effect of TFLS model is better than LSTM and Transformer models.Compared with the mixed runoff model,the number of floods with a maximum error of 20%has increased from 9 to 12,accounting for 70.6%of all floods.The number of floods with a Nash-Sutcliffe coefficient of not less than 0.8 has increased from 5 to 9,accounting for 52.9%of all floods.The performance of the TFLS model is better than other models when the observed flow does not exceed 500 m^(3)/s,and the perfor-mance of LSTM is slightly better than other models when the flow exceed 500 m^(3)/s.

关 键 词：分布式模型 洪水预报 误差校正 LSTM TRANSFORMER 

分 类 号：P338[天文地球—水文科学]