Effects of stacking LSTM with different patterns and input schemes on streamflow and water quality simulation DOI Creative Commons

Yucong Hu,

Yan Jiang,

Huiting Yao

и другие.

Research Square (Research Square), Год журнала: 2024, Номер unknown

Опубликована: Янв. 23, 2024

Abstract Streamflow and water quality parameters (WQs) are commonly forecasted by mechanism models statistics models. However, these challenged due to computational complexity, redundant parameters, etc. Therefore, a stacking Long short-term memory networks (LSTM) model with two patterns different input schemes was applied simulate streamflow eight WQs in this study. The results showed that sliding windows detected as the more stable pattern for both forecasts. accuracy of predicting using only meteorological inputs limited especially low-volume flow. Whereas, prediction three variables (i.e., factors, streamflow, other influential WQs) reliable reaching an average relative error (RE) below 17%. When adding historical data into dataset, accuracies could be increased close benchmarks Delft 3D model. Our study documents LSTM is effective method

Язык: Английский

Study on optimization and combination strategy of multiple daily runoff prediction models coupled with physical mechanism and LSTM DOI
Jun Guo, Yi Liu, Qiang Zou

и другие.

Journal of Hydrology, Год журнала: 2023, Номер 624, С. 129969 - 129969

Опубликована: Июль 20, 2023

Язык: Английский

Процитировано

124
Deep transfer learning based on transformer for flood forecasting in data-sparse basins DOI

Yuanhao Xu,

Kairong Lin,

Caihong Hu

и другие.

Journal of Hydrology, Год журнала: 2023, Номер 625, С. 129956 - 129956

Опубликована: Июль 19, 2023

Язык: Английский

Процитировано

68
Temporal Forecasting of Distributed Temperature Sensing in a Thermal Hydraulic System with Machine Learning and Statistical Models DOI Creative Commons
Stella Pantopoulou, M. Weathered, Darius Lisowski

и другие.

IEEE Access, Год журнала: 2025, Номер 13, С. 10252 - 10264

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

2
Improved monthly runoff time series prediction using the CABES-LSTM mixture model based on CEEMDAN-VMD decomposition DOI Creative Commons

Dong-mei Xu,

An-dong Liao,

Wenchuan Wang

и другие.

Journal of Hydroinformatics, Год журнала: 2023, Номер 26(1), С. 255 - 283

Опубликована: Дек. 11, 2023

Abstract Accurate runoff prediction is vital in efficiently managing water resources. In this paper, a hybrid model combining complete ensemble empirical mode decomposition with adaptive noise, variational decomposition, CABES, and long short-term memory network (CEEMDAN-VMD-CABES-LSTM) proposed. Firstly, CEEMDAN used to decompose the original data, high-frequency component decomposed using VMD. Then, each input into LSTM optimized by CABES for prediction. Finally, results of individual predictions are combined reconstructed produce monthly predictions. The employed predict at Xiajiang hydrological station Yingluoxia station. A comprehensive comparison conducted other models including back propagation (BP), LSTM, etc. assessment model's performance uses four evaluation indexes. Results reveal that CEEMDAN-VMD-CABES-LSTM showcased highest forecast accuracy among all evaluated. Compared single root mean square error (RMSE) absolute percentage (MAPE) decreased 71.09 65.26%, respectively, RMSE MAPE 65.13 40.42%, respectively. R NSEC both sites near 1.

Язык: Английский

Процитировано

22
Machine learning method is an alternative for the hydrological model in an alpine catchment in the Tianshan region, Central Asia DOI Creative Commons
Wenting Liang, Yaning Chen, Gonghuan Fang

и другие.

Journal of Hydrology Regional Studies, Год журнала: 2023, Номер 49, С. 101492 - 101492

Опубликована: Авг. 1, 2023

Kaidu River catchment in the Tianshan Mountain, northwestern China. This paper compared applicability and accuracy of four machine learning models two hydrological ones to simulate daily streamflow extreme catchment. The are Support Vector Regression (SVR), eXtreme Gradient Boosting (XGBoost), Random Forests (RF), Long Short-Term Memory (LSTM), while Soil Water Assessment Tool (SWAT) extended SWAT with a glacier dynamic module (SWAT-Glacier). LSTM achieved better model performance simulating than SWAT-Glacier, Kling-Gupta efficiency 0.92, 0.82, 0.80, respectively. Meanwhile, SVR, XGBoost, RF showed satisfactory performance, KGE 0.67, 0.71, 0.70, LSTM, SWAT-Glacier could well annual peak flow (i.e., maximum 1-day 5-day average streamflow) but failed mimic minimum 7-day streamflow, PBIAS exceeding 28%. Furthermore, all reproduce dates extremes. Nevertheless, using quantile loss function resulted significantly improved low indices, that mean squared error as function. Overall, be good alternative for data-scarce catchments.

Язык: Английский

Процитировано

21
Improving Hydrological Modeling with Hybrid Models: A Comparative Study of Different Mechanisms for Coupling Deep Learning Models with Process-based Models DOI
Yiming Wei, Renchao Wang, Ping Feng

и другие.

Water Resources Management, Год журнала: 2024, Номер 38(7), С. 2471 - 2488

Опубликована: Фев. 10, 2024

Язык: Английский

Процитировано

7
Transferred Long Short-Term Memory Network for River Flow Forecasting in Data-Scarce Basins DOI
Zhenglei Xie, Wei Xu, Bing Zhu

и другие.

Water Resources Management, Год журнала: 2025, Номер unknown

Опубликована: Март 11, 2025

Язык: Английский

Процитировано

1
Monthly Runoff Prediction Via Mode Decomposition-Recombination Technique DOI
Xi Yang, Zhihe Chen, Min Qin

и другие.

Water Resources Management, Год журнала: 2023, Номер 38(1), С. 269 - 286

Опубликована: Дек. 1, 2023

Язык: Английский

Процитировано

17
Review of Machine Learning Methods for River Flood Routing DOI Open Access

Li Li,

Kyung Soo Jun

Water, Год журнала: 2024, Номер 16(2), С. 364 - 364

Опубликована: Янв. 22, 2024

River flood routing computes changes in the shape of a wave over time as it travels downstream along river. Conventional models, especially hydrodynamic require high quality and quantity input data, such measured hydrologic series, geometric hydraulic structures, hydrological parameters. Unlike physically based machine learning algorithms, which are data-driven do not much knowledge about underlying physical processes can identify complex nonlinearity between inputs outputs. Due to their higher performance, lower complexity, low computation cost, researchers introduced novel methods single application or hybrid achieve more accurate efficient routing. This paper reviews recent river

Язык: Английский

Процитировано

5
A novel smoothing-based long short-term memory framework for short-to medium-range flood forecasting DOI
Amina Khatun, Chandranath Chatterjee, Gaurav Sahu

и другие.

Hydrological Sciences Journal, Год журнала: 2023, Номер 68(3), С. 488 - 506

Опубликована: Янв. 27, 2023

ABSTRACTA novel smoothing-based long short-term memory (Smooth-LSTM) framework for flood forecasting up to five days ahead is proposed, and compared with the benchmark LSTM (LSTM) model, an Artificial Neural Network (ANN) conceptual Nedbør Afstrømnings Model (MIKE11 NAM)-Hydrodynamic (HD) (MIKE) hydrological model. This was tested in typical middle Mahanadi River basin (India), which has a tropical monsoon-type climate. Variation of training loss indicated network higher learning ability at smaller batch sizes. The Smooth-LSTM model could predict streamflow Nash-Sutcliffe efficiency 0.82–0.87 lead time better reproduction observed crucial high peak floods, whereas corresponding MIKE, ANN model-based forecasts were acceptable only four-, three- one-day times, respectively. Overall, found be robust operational forecasting, lower uncertainty least sensitivity redundant input information.KEYWORDS: ANNflood forecastinglong (LSTM)MIKE 11 NAM-HDsmoothing windows Editor A. Castellarin Associate O. KisiEditor KisiAcknowledgementsThe authors sincerely thank Hirakud Dam Circle (HDC), Department Water Resources (Prachi Division), Odisha, Central Commission (CWC) India Meteorological (IMD) providing necessary datasets carry out study.Disclosure statementNo potential conflict interest reported by authors.Data availability statementAll data, models, code generated or used during study appear submitted article.Supplementary materialSupplemental data this article can accessed online https://doi.org/10.1080/02626667.2023.2173012

Язык: Английский

Процитировано

11