黄河三门峡水文站非汛期流量预测研究  

Study on Non-Flood Season Flow Prediction of Sanmenxia Hydrologic Station of the Yellow River

作　　者：程桂芳 周芸 CHENG Guifang;ZHOU Yun(School of Mathematics and Statistics,Zhengzhou University,Zhengzhou 450001,China)

机构地区：[1]郑州大学数学与统计学院,河南郑州450001

出　　处：《人民黄河》2025年第4期38-43,57,共7页Yellow River

基　　金：河南省高等教育教学改革研究与实践项目(2021SJGLX060);河南省科技攻关项目(252102211117)。

摘　　要：黄河径流具有非稳态、非线性的特点,为了给河南省保障水安全等提供参考,对黄河三门峡水文站非汛期流量进行了研究。构建变分模态分解(VMD)与长短期记忆网络(LSTM)、支持向量回归机(SVR)相结合的非汛期径流预测模型,利用麻雀优化算法(SSA)调节模型参数以提高预测精度。采用VMD算法将非汛期流量数据分解为多个本征模函数(IMF),基于K-Means聚类法计算分量间的欧氏距离,将欧氏距离的倒数作为各分量的权重,最后将各分量结果输入LSTM/SVR进行模型预测,加权重构分量预测值得到流量预测结果,并与加权前后VMD-SSA-LSTM、VMD-SSA-SVR模型进行对比。结果显示,提出的K-Means加权VMD-SSA-LSTM模型预测三门峡水文站2003年1月—2023年5月(非汛期月份)每日平均流量,平均绝对误差为82.54 m^(3)/s、均方根误差为106.64m^(3)/s、拟合优度达0.92,能有效预测非汛期流量。The Yellow River runoff has unsteady and non-linear characteristics.In order to provide reference for ensuring water security in Henan Province,the non-flood season discharge of Sanmenxia Hydrology Station of the Yellow River was studied.The paper built non-flood season runoff prediction models by combining variational mode decomposition(VMD)with long short term memory(LSTM)and support vector regression(SVR).The sparrow search algorithm(SSA)was used to adjust the model parameters to improve the prediction accuracy.The runoff data was decomposed into multiple eigenmode functions(IMF)by the VMD algorithm,Euclidean distance between components was calculated based on K-Means clustering method and the reciprocal of Euclidean distance was used as the weight of each component.Finally,the results of each component were put into LSTM/SVR for model prediction,and the runoff results were obtained by weighted reconstruction of the predicted values of components.Comparing with before and after weighted VMD-SSA-LSTM and VMD-SSA-SVR model,the results show that the proposed K-Means weighted VMD-SSA-LSTM model predicts the average daily runoff of Sanmenxia Hydrology Station from January 2003 to May 2023(non-flood season month),with the mean absolute error being 82.54 m^(3)/s,the root-mean-square error being 106.64 m^(3)/s and the fitted coefficient being 0.92,the trend of runoff can be predicted more effectively.

关 键 词：径流预测 变分模态分解 LSTM SVR K-MEANS聚类 黄河流域 

分 类 号：TV734.1[水利工程—水利水电工程] TV882.1