摘要
岩溶地区因其复杂的产汇流机制,洪水预报精度普遍不高。以具有典型岩溶地貌特征的贵州省六硐河流域为例,分别构建基于机理分析的新安江岩溶模型和基于深度学习的长短期记忆网络(LSTM)模型,对比两类模型在岩溶地区洪水过程模拟中的适应性。结果表明,新安江岩溶模型在率定期(验证期)的平均纳什效率系数为0.66(0.70),洪峰合格率为61.3%(62.5%),峰现时间合格率为80.6%(75%);预见期4 h内LSTM模型在率定期(验证期)的平均纳什效率系数在0.89(0.89)以上,洪峰合格率在93.5%(75%)以上,峰现时间合格率在61.3%(50%)以上。总体上,机理分析和深度学习模型均能较好地模拟岩溶地区洪水的涨落过程,但预见期3 h内的LSTM模型精度要高于新安江岩溶模型。研究结果可为产汇流机制较为复杂的岩溶地区的洪水过程模拟提供新途径。
Due to the complex runoff generation and concentration mechanisms in Karst area,the accuracy of flood forecasting is generally low.Taking the Liudong river watershed with typical karst landforms as the study area,this paper investigates the adaptability of mechanism analysis Karst-Xin'anjiang model and LSTM model in floods simulation in Karst area.The results show that the Karst-Xin'anjiang model had an average Nash-Sutcliffe efficiency coefficient of 0.66(0.70),a peak flow passing rate of 61.3%(62.5%),and a peak occurrence time passing rate of 80.6%(75%)during the calibration(validation)period.Within a forecast period of 4 hours,the LSTM model achieved an average Nash-Sutcliffe efficiency coefficient of over 0.89(0.89),a peak flow passing rate of over 93.5%(75%),and a peak occurrence time passing rate of 61.3%(50%)during the calibrationc(validation)period.Overall,both mechanistic analysis and deep learning models are capable of adequately simulating the rising and falling processes of floods in Karst area.However,within a forecast period of 3 hours,the LSTM model demonstrates higher accuracy compared to the Karst-Xin'anjiang model.This provides a new approach for flood process simulation in Karst areas with complex runoff generation and concentration mechanisms.
作者
陶驷骥
李彬权
陈云瑶
夏奕洁
赵建飞
李匡
TAO Si-ji;LI Bin-quan;CHEN Yun-yao;XIA Yi-jie;ZHAO Jian-fei;LI Kuang(Guizhou Water&Power Survey-DesignInstitute Co.,Ltd.,Guiyang 550002,China;College of Hydrology and Water Resources,Hohai University,Nanjing 210098,China;State Key Laboratory of Water Cycle and Water Security,Hohai University,Nanjing 210098,China;ChinaInstitute of Water Resources and Hydropower Research,Beijing 100038,China)
出处
《水电能源科学》
北大核心
2025年第10期6-9,55,共5页
Water Resources and Power
基金
国家自然科学基金项目(42471049)。
关键词
新安江岩溶模型
长短期记忆网络模型
岩溶地区
洪水模拟
六硐河流域
Karst-Xin'anjiang model
long short-term memory network model
Karst area
flood simulation
Liudong river watershed
