Environmental Monitoring and Assessment, Год журнала: 2024, Номер 197(1)
Опубликована: Дек. 26, 2024
Язык: Английский
Environmental Monitoring and Assessment, Год журнала: 2024, Номер 197(1)
Опубликована: Дек. 26, 2024
Язык: Английский
Journal of Hydrology, Год журнала: 2024, Номер unknown, С. 132223 - 132223
Опубликована: Окт. 1, 2024
Язык: Английский
Процитировано
11Water, Год журнала: 2024, Номер 16(10), С. 1407 - 1407
Опубликована: Май 15, 2024
Artificial intelligence has undergone rapid development in the last thirty years and been widely used fields of materials, new energy, medicine, engineering. Similarly, a growing area research is use deep learning (DL) methods connection with hydrological time series to better comprehend expose changing rules these series. Consequently, we provide review latest advancements employing DL techniques for forecasting. First, examine application convolutional neural networks (CNNs) recurrent (RNNs) forecasting, along comparison between them. Second, made basic enhanced long short-term memory (LSTM) analyzing their improvements, prediction accuracies, computational costs. Third, performance GRUs, other models including generative adversarial (GANs), residual (ResNets), graph (GNNs), estimated Finally, this paper discusses benefits challenges associated forecasting using techniques, CNN, RNN, LSTM, GAN, ResNet, GNN models. Additionally, it outlines key issues that need be addressed future.
Язык: Английский
Процитировано
10Engineering Applications of Artificial Intelligence, Год журнала: 2024, Номер 135, С. 108744 - 108744
Опубликована: Июнь 3, 2024
Язык: Английский
Процитировано
10Expert Systems with Applications, Год журнала: 2024, Номер 256, С. 124900 - 124900
Опубликована: Июль 30, 2024
Язык: Английский
Процитировано
7Опубликована: Янв. 1, 2025
Язык: Английский
Процитировано
1Earth Science Informatics, Год журнала: 2025, Номер 18(3)
Опубликована: Фев. 25, 2025
Язык: Английский
Процитировано
1Smart Agricultural Technology, Год журнала: 2025, Номер unknown, С. 100896 - 100896
Опубликована: Март 1, 2025
Язык: Английский
Процитировано
1Smart Agricultural Technology, Год журнала: 2024, Номер unknown, С. 100619 - 100619
Опубликована: Окт. 1, 2024
Язык: Английский
Процитировано
3Journal of Water and Climate Change, Год журнала: 2024, Номер 16(1), С. 230 - 247
Опубликована: Дек. 18, 2024
ABSTRACT Accurate forecasting of increasingly unpredictable river runoff is essential for effective water resource management in the face climate change and human activities. This study uses four machine learning models long short-term memory neural networks (LSTM), support vector (SVM), random forest, artificial network to improve accuracy explore combined models’ effectiveness. develops three advanced (empirical mode decomposition (EMD)–LSTM, VMD–LSTM, wavelet analysis (WA)–LSTM) by combining preprocessing techniques EMD, variational (VMD), WA with LSTM modeling method. These use signal analyze 41 years data from Huanren station (1980–2020). The findings reveal that model outperforms other individual when days high runoff. Among decomposed models, VMD–LSTM demonstrates best overall performance during validation period, achieving root mean square error, Nash–Sutcliffe efficiency coefficient, bias values 52.14 m3/s, 0.96, −0.002, respectively. combination shows promising potential enhancing prediction accuracy, practical implications flood control strategies.
Язык: Английский
Процитировано
3Опубликована: Янв. 1, 2025
In arid areas, estimation of crop water demand through potential evapotranspiration (PET) forecast has a guiding effect on water-saving irrigation, to cope with the crisis shortage. Neural network-based PET prediction methods is considered have huge application because its small error. However, physical conditions and data quality in different regions make choice neural network different, making it difficult provide general method. So an adaptive hybrid model based automatic machine learning for short-term proposed coupling formula. Process divided into two stages: forecasting. Learning stage includes three modules: meteorological reconstructing, set generation (PET calculation formula + network) selecting. Forecast rolling prediction. 105 standard weather stations Xinjiang were used as sets (43 them had missing data) test model. According modules, networks formulas process, corresponding labels generated each dataset result. Ratio training was 8:2. Grid search optimize best hyperparameter combination. set, average absolute error (MAE) squared (MSE) 0.338mm 0.270, achieving high accuracy. The mean smaller any single mixed We demonstrate that applicability varies among sources, Gate Recurrent Unit (GRU) 1 Dimension convolutional (1DCNN) are more suitable selected datasets, while Long Short Term Memory (LSTM) Multilayer Perceptron (MLP) not applicable. Combined analysis labels, find evidences independent geographic region degree drought. 2023, method 1-15 days verified, verification results show significantly than useing calculate PET. addition, by comparison,we determined input length can effectively reduce error, MAE 27.52% fixed length, MSE 45.76% length. realized forecast, predict accurately, be further expanded adding improve generalization ability.
Язык: Английский
Процитировано
0