机器学习模型在岩溶地区水文预报中的适用性分析  

Applicability Analysis of Machine Learning Model in Hydrological Forecasting in Karst Areas

作　　者：赵泽锦 孙伟 周斌 张轩 王高旭[3] 吴巍[3] 李文杰 姚业 ZHAO Zejin;SUN Wei;ZHOU Bin;ZHANG Xuan;WANG Gaoxu;WU Wei;LI Wenjie;YAO Ye(Honghe Nanyuan Water Supply Co.,Ltd,Mengzi 651400,China;Nanjing R&D Hydro-Information Technology Co.,Ltd.,Nanjing 210029,China;State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering,Nanjing Hydraulic Research Institute,Nanjing 210029,China)

机构地区：[1]红河州南源供水有限公司,云南蒙自651400 [2]南京瑞迪水利信息科技有限公司,江苏南京210029 [3]南京水利科学研究院水文水资源与水利工程科学国家重点实验室,江苏南京210029

出　　处：《人民珠江》2024年第3期59-68,共10页Pearl River

基　　金：国家自然科学基金(91847301、92047203、42075191、52009080、42175177);蒙开个地区河库连通工程梯级泵站运行调度研究科技项目(MKG〔2021〕-KYKT-01);中央级公益性科研院所基本科研业务费专项资金(Y521011、Y521013)。

摘　　要：现有岩溶地区的水文预报主要采用基于物理机制的水文模型,而机器学习模型的应用较为罕见。为了探索机器学习模型在岩溶地区水文预报的适用性,以云南省沙甸河流域为研究区,采用了LSTM模型与随机森林模型对倘甸水文站的逐日径流量与场次洪水进行了模拟,并以针对岩溶地区的改进型新安江模型做参照。研究表明:机器学习模型与改进型新安江模型在日径流过程模拟方面都取得较好的效果,LSTM模型模拟效果更优;场次洪水模拟方面,改进型新安江模型达到了甲级预报精度,机器学习模型6 h预见期预报结果整体优于改进型新安江模型,但24 h预见期的预报结果不能满足预报业务的精度需求。对2种机器学习模型和水文模型的特点及预报精度进行的研究,为岩溶地区水文预报工作提供了参考。For hydrological forecasting in karst areas,existing research mainly uses hydrological models based on physical mechanisms,while rare research focuses on machine learning models.To explore the applicability of machine learning models in karst areas,this paper utilizes the LSTM model and random forest model to simulate the daily runoff and field floods at Tangdian hydrological station,using the Shadian River basin in Yunnan Province as the study area.The modified Xin anjiang model for karst areas is taken as a reference.The results show that both the machine learning model and the modified Xin anjiang model have achieved good results in simulating the daily runoff process,with the LSTM model showing better simulation results.In the simulation of floods,the modified Xin anjiang model achieves Class A forecast accuracy.The machine learning models have better forecast results for the 6-hour forecasting period than the modified Xin anjiang model,while the forecast results for the 24-hour forecasting period do not meet the accuracy requirements of the forecast operation.The study provides a reference for hydrological forecasting in karst areas by studying the characteristics and forecasting accuracy of two machine learning models and a hydrologic model.

关 键 词：岩溶地区 洪水预报 机器学习 改进型新安江模型 

分 类 号：P338[天文地球—水文科学]