Stacking ensemble learning models for daily runoff prediction using 1D and 2D CNNs

被引:64
作者
Xie, Yutong [1 ,2 ]
Sun, Wei [1 ,2 ]
Ren, Miaomiao [1 ,2 ]
Chen, Shu [1 ,2 ]
Huang, Zexi [1 ,2 ]
Pan, Xingyou [1 ,2 ]
机构
[1] Sun Yat Sen Univ, Sch Geog & Planning, Guangzhou 510275, Guangdong, Peoples R China
[2] Southern Marine Sci & Engn Guangdong Lab Zhuhai, Zhuhai 519082, Peoples R China
基金
美国国家科学基金会;
关键词
Runoff prediction; CNN; Dimensions; Stacking ensemble learning; Time steps; ARTIFICIAL NEURAL-NETWORKS; MONTHLY STREAMFLOW; RIVER; DECOMPOSITION; PERFORMANCE; BOOTSTRAP; FRAMEWORK; MACHINES; OUTPUTS; SURFACE;
D O I
10.1016/j.eswa.2022.119469
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
In recent years, applications of convolutional neural networks (CNNs) to runoff prediction have received some attention due to their excellent feature extraction capabilities. However, existing studies are still limited since merely either 1D or 2D CNNs are developed to predict runoff. In this study, a stacking ensemble learning model for daily runoff prediction based on different types of 1D and 2D CNNs is proposed and applied to the Quinebaug River Basin, Connecticut, USA. The structure of the CNN models is developed with reference to the classic LeNet5 network. Especially, the predictors are reconstructed into 1D vectors and 2D matrices with 10-, 20-and 30-day time steps. Totally 18 member models are constructed through selecting 3 representative 1D and 2D CNN models with 3 time steps. The simple average method (SAM) is used to integrate different CNN member models. The results show that the performance of the same-type SAM based on either 1D or 2D CNNs improves because it can counteract the effects of both positive and negative predicted values by the member models to some extent. Furthermore, the mixed-type SAM models based on both 1D and 2D CNN member models can further improve the prediction accuracy. The optimal model SAM15 consists of two 1D CNNs and four 2D CNNs. Compared with the optimal CNN member models, SAM15 reduces the validation RMSE by about 13% and improves the vali-dation R and NSE by about 3% and 7%, respectively. This study highlights that the proposed stacking ensemble learning model can improve the daily runoff prediction accuracy through integration of the nonlinear fitting ability of 1D and 2D CNN member models.
引用
收藏
页数:15
相关论文
共 50 条
[31]   Retrieval of wave information using nautical radar images based on the 2D CWT and 1D FFT algorithm [J].
Ju, Feng-zhou ;
Chu, Xiao-liang ;
Wang, Jian ;
Gu, Yan-zhen .
PROCEEDINGS OF THE 2015 INTERNATIONAL CONFERENCE ON AUTOMATION, MECHANICAL CONTROL AND COMPUTATIONAL ENGINEERING, 2015, 124 :1379-1383
[32]   A comparative study of superhydrophobicity of 0D/1D/2D thermally functionalized carbon nanomaterials [J].
Lim, Chang Sheng ;
Von Lau, Ee ;
Kee, Kok Eng ;
Hung, Yew Mun .
CERAMICS INTERNATIONAL, 2021, 47 (21) :30331-30342
[33]   Application of 1D/2D carbon material supported metal nanoclusters for electrochemical conversion [J].
Song, Zhaoxin ;
Shi, Haiting ;
Liu, Liangsen ;
Zeng, Ming ;
Wang, Shuo ;
Wu, Xianyan ;
Xu, Zhiwei .
CATALYSIS SCIENCE & TECHNOLOGY, 2024, 14 (06) :1462-1479
[34]   Evaluation of 1D and 2D Deep Convolutional Neural Networks for Driving Event Recognition [J].
Escotta, Alvaro Teixeira ;
Beccaro, Wesley ;
Ramirez, Miguel Arjona .
SENSORS, 2022, 22 (11)
[35]   The Effect of Network Formation on the Mechanical Properties of 1D:2D Nano:Nano Composites [J].
Gabbett, Cian ;
Boland, Conor S. ;
Harvey, Andrew ;
Vega-Mayoral, Victor ;
Young, Robert J. ;
Coleman, Jonathan N. .
CHEMISTRY OF MATERIALS, 2018, 30 (15) :5245-5255
[36]   PRECONDITIONERS FOR SADDLE POINT SYSTEMS WITH TRACE CONSTRAINTS COUPLING 2D AND 1D DOMAINS [J].
Kuchta, Miroslav ;
Nordaas, Magne ;
Verschaeve, Joris C. G. ;
Mortensen, Mikael ;
Mardal, Kent-Andre .
SIAM JOURNAL ON SCIENTIFIC COMPUTING, 2016, 38 (06) :B962-B987
[37]   1D/2D heterostructure nanofiber flexible sensing device with efficient gas detectivity [J].
Wang, Kang ;
Wei, Wei ;
Lou, Zheng ;
Zhang, Hao ;
Wang, Lili .
APPLIED SURFACE SCIENCE, 2019, 479 :209-215
[38]   Dzyaloshinskii-Moriya interaction and magnetic ordering in 1D and 2D at nonzero T [J].
Torres, F. ;
Altbir, D. ;
Kiwi, M. .
EPL, 2014, 106 (04)
[39]   Interfacial engineering of 1D/2D heterostructured photoanode for efficient photoelectrochemical water splitting [J].
Wang, Ziyu ;
Qin, Yimo ;
Wu, Xin ;
He, Kui ;
Li, Xiaolong ;
Wang, Juan .
NANOTECHNOLOGY, 2022, 33 (49)
[40]   Low-complexity CNN with 1D and 2D filters for super-resolution [J].
Park, Jangsoo ;
Lee, Jongseok ;
Sim, Donggyu .
JOURNAL OF REAL-TIME IMAGE PROCESSING, 2020, 17 (06) :2065-2076