摘要
可靠的径流模拟对流域水资源规划与管理意义重大。以岷江镇江关水文站实测径流为研究对象,通过与BP神经网络和Elman循环神经网络的对比,验证长短期记忆神经网络在受季节性融雪影响流域日尺度降水-径流模拟中的适用性,并进一步分析长短期记忆神经网络的关键参数——时间步长对日径流模拟精度的影响。结果表明:①采用BP神经网络进行日径流过程模拟时会丢失流域状态信息,模拟效果最差;②Elman循环神经网络相比BP神经网络,具有相对有限的记忆能力,在积雪时段较长的岷江镇江关水文站控制流域上的模拟效果一般;③长短期记忆神经网络以其特殊的CEC单元和“门”结构,实现了流域状态的长期储存与更新,在日降水-径流模拟中的效果最佳;④通过多次试验发现,当长短期记忆神经网络的时间步长设置为60 d时,模拟精度最高,结合春末夏初的降水、径流和气温变化过程,认为60 d时间步长符合岷江流域实际情况。
The reliable runoff simulation is of great significance to the planning and management of watershed water resources.Taking the measured runoff at Zhenjiangguan Hydrological Station located in the upper reach of the Minjiang River as the research object,comparing with the back propagation(BP)neural network and Elman recurrent neural network(ERNN),the applicability of long short-term memory(LSTM)neural network in the rainfall-runoff simulation at daily time scale affected by seasonal snowmelt is verified.It further analyzes the key parameter of the LSTM neural network,which is the influence of time step on the accuracy of daily runoff simulation.The results are as follows.Firstly,the BP neural network will lose the basin state information when simulating the daily runoff process,and the simulation effect performed poorly.Secondly,compared with BP neural network,Elman recurrent neural network has relatively limited memory ability.The simulation effect in the control basin of the Zhenjiangguan Hydrological Station of the Minjiang River with a long snow period is average.Thirdly,with its special CEC unit and″gate″structure,the LSTM neural network realizes the long-term storage and update of the watershed state,and performs best effect in daily rainfall-runoff simulation.Fourthly,after many experiments,it is found that the simulation accuracy is the highest when the time step of the LST neural network is set to 60 d.Based on the precipitation,runoff and temperature changes in late spring and early summer,it is concluded that the 60 d time step is in line with the actual situation of the Minjiang River basin.
作者
党池恒
张洪波
陈克宇
支童
卫星辰
DANG Chiheng;ZHANG Hongbo;CHEN Keyu;ZHI Tong;WEI Xingchen(School of Water and Environment,Chang′an University,Xi′an 710054,China;Xi′an Technological University,Xi′an 710021,China)
出处
《华北水利水电大学学报(自然科学版)》
2020年第5期10-18,33,共10页
Journal of North China University of Water Resources and Electric Power:Natural Science Edition
基金
国家自然科学基金项目(51979005)
陕西省水利厅科技计划项目(2018slkj-11)。