基于WD-DO-HKELM模型的月径流时间序列预测  

Monthly Runoff Time Series Prediction Based on WD-DO-HKELM Model

作　　者：赵祥 崔东文 ZHAO Xiang;CUI Dongwen(Yunnan Provincial Hydrological and Water Resources Bureau,Honghe Branch,Honghe 661000,China;Water Affairs Bureau of Wenshan Prefecture,Wenshan 663000,China)

机构地区：[1]云南省水文水资源局红河分局,云南红河661000 [2]云南省文山州水务局,云南文山663000

出　　处：《云南水力发电》2025年第2期26-30,共5页Yunnan Water Power

基　　金：国家高分辨率对地观测系统重大科技专项(89-Y50-G31-9001-22)。

摘　　要：为提高月径流时间序列预测精度,提出小波分解(WD)-蒲公英优化(DO)算法-混合核极限学习机(HKELM)月径流时间序列预测模型。首先,利用WD将月径流时序数据分解为4个更具规律的子序列分量D1~D3、A3,划分各分量训练集和预测集;其次,简要介绍DO原理,基于训练集构建DO优化HKELM超参数的目标函数,利用获得的最佳超参数建立DO-DO-HKELM模型对预测集各分量进行预测和重构;最后,通过云南省滴水水文站月径流预测实例对WD-DO-HKELM模型进行检验,并与WD-DO-LSSVM、WD-DO-BP、DO-HKELM模型作对比分析。结果表明:(1) WD-DO-HKELM模型预测的平均绝对百分比误差(MAPE)、平均绝对误差(MAE)、均方根误差(RMSE)、决定系数R^(2)分别为1.118%、0.021 m^(3)/s、0.027 m^(3)/s、0.9998,预测误差小于WD-DO-LSSVM、WD-DO-BP模型,远小于DO-HKELM模型,具有更高的预测精度和更好的泛化性能;(2)在相同分解和优化的情形下,HKELM预测性能优于LSSVM、BP。To improve the accuracy of monthly runoff time series prediction,a WD-DO-HKELM monthly runoff time series prediction model is proposed.Firstly,using WD,the monthly runoff time series data is decomposed into four more regular sub sequence components D1~D3,A3,and the training and prediction sets for each component are divided;Secondly,it briefly introduces the principle of DO,construct an objective function for optimizing HKELM hyperparameters based on the training set,and use the obtained optimal hyperparameters to establish a DO-DO-HKELM model for predicting and reconstructing each component of the prediction set;Finally,the WD-DO-HKELM model was tested using a monthly runoff prediction example from the Yunnan Dishui Hydrological Station,and compared and analyzed with the WD-DO-LSSVM,WD-DO-BP,and DO-HKELM models.The results showed that:(1) The MAPE,MAE,RMSE,and R2 predicted by the WD-DO-HKELM model were 1.118%,0.021m^(3)/s,0.027m^(3)/s,and 0.9998,respectively,with prediction errors smaller than the WD-DO-LSSVM and WD-DO-BP models,and much smaller than the DO-HKELM model.It has higher prediction accuracy and better generalization performance.(2) Under the same decomposition and optimization conditions,HKELM performs better in prediction than LSSVM and BP.

关 键 词：月径流预测 小波分解 蒲公英优化算法 混合核极限学习机 超参数优化 

分 类 号：TV121[水利工程—水文学及水资源]