摘要
为探索提高径流预测精度,基于兰州水文站2001年8月~2019年12月的逐日径流数据,在控制变量法的基础上,应用LSTM、ARIMA、SVR、XGBoost四种模型,建立了单一模型、EMD分解重构、剔除噪声模态分量后的EMD分解重构等三类处理方式共12种模型方案,并对12种方案的评价指标进行对比。结果表明,EMD序列分解重构技术和基于Hurst指数的噪声模态分量剔除有助于提升预测精度,与单一模型相比,前者构建的模型的均方根误差(R_(RMSE))平均下降了15.16%,后者平均下降了28.49%;12种方案中,预测效果较好的方案是剔除噪声模态分量后的“EMD-SVR-ARIMA”模型。
In order to improve the accuracy of runoff prediction,based on the control variable method and the daily runoff data of Lanzhou hydrometric station from August 2001 to December 2019,the models of the LSTM,ARIMA,SVR and XGBoost were used to establish 12 model schemes,including single model,EMD decomposition and reconstruction,EMD decomposition and reconstruction after removing noise components,and evaluation indicators of the 12 schemes were compared.The results show that the EMD sequence decomposition and reconstruction technology and noise component elimination based on Hurst exponent are helpful to improve the prediction accuracy.Compared with the single model,the R_(RMSE) of the model constructed by the former decreased by 15.16%on average,and that of the latter decreased by 28.49%on average.Among the 12 schemes,EMD-SVR-ARIMA with noise components removed is the best model.
作者
路炜
魏霖静
LU Wei;WEI Lin-jing(College of Science,Gansu Agricultural University,Lanzhou 730030,China;College of Information Science and Technology,Gansu Agricultural University,Lanzhou 730030,China)
出处
《水电能源科学》
北大核心
2023年第8期19-22,9,共5页
Water Resources and Power
基金
2020年甘肃农业大学研究生教育研究项目(2020-19)
2021年度兰州市人才创新创业项目(2021-RC-47)
2021年教育部产学研合作协同育人项目(202102326036)。
关键词
日径流预测
EMD分解
兰州水文站
机器学习模型
daily runoff forecasting
EMD decomposition
Lanzhou hydrological station
machine learning model
