آمار

تعداد نشریات22
تعداد شماره‌ها329
تعداد مقالات3,440
تعداد مشاهده مقاله4,046,158
تعداد دریافت فایل اصل مقاله2,835,843
SAFAVI, MARYAM, Khashei-Siuki, Abbas, HASHEMI, REZA, ehteram, mohammad, PIRI, JAMSHID. (1404). Comparative Analysis of Hybrid Deep Learning Models for Dam Inflow Prediction: LSTM-GRU, CNN-LSTM, Attention-LSTM, and Transformer Approaches. , (), -. doi: 10.22077/jwhr.2025.9938.1185
MARYAM SAFAVI; Abbas Khashei-Siuki; REZA HASHEMI; mohammad ehteram; JAMSHID PIRI. "Comparative Analysis of Hybrid Deep Learning Models for Dam Inflow Prediction: LSTM-GRU, CNN-LSTM, Attention-LSTM, and Transformer Approaches". , , , 1404, -. doi: 10.22077/jwhr.2025.9938.1185
SAFAVI, MARYAM, Khashei-Siuki, Abbas, HASHEMI, REZA, ehteram, mohammad, PIRI, JAMSHID. (1404). 'Comparative Analysis of Hybrid Deep Learning Models for Dam Inflow Prediction: LSTM-GRU, CNN-LSTM, Attention-LSTM, and Transformer Approaches', , (), pp. -. doi: 10.22077/jwhr.2025.9938.1185
SAFAVI, MARYAM, Khashei-Siuki, Abbas, HASHEMI, REZA, ehteram, mohammad, PIRI, JAMSHID. Comparative Analysis of Hybrid Deep Learning Models for Dam Inflow Prediction: LSTM-GRU, CNN-LSTM, Attention-LSTM, and Transformer Approaches. , 1404; (): -. doi: 10.22077/jwhr.2025.9938.1185

Comparative Analysis of Hybrid Deep Learning Models for Dam Inflow Prediction: LSTM-GRU, CNN-LSTM, Attention-LSTM, and Transformer Approaches

Water Harvesting Research
مقالات آماده انتشار، پذیرفته شده، انتشار آنلاین از تاریخ 28 شهریور 1404
نوع مقاله: Research Paper
شناسه دیجیتال (DOI): 10.22077/jwhr.2025.9938.1185
نویسندگان
MARYAM SAFAVI1؛ Abbas Khashei-Siuki* 2؛ REZA HASHEMI3؛ mohammad ehteram4؛ JAMSHID PIRI5
1aPh.D Student of Water Resources Management,university of Birjand, , Birjand, Iran
2Professor of Water Engineering Dpt.
3Department of Water Engineering, University of Birjand, Birjand, Iran
4department of water engineering semnan university. iran
5Department of Water Engineering, University of Zabol, Zabol, Iran
چکیده
A comprehensive framework is presented for the prediction of inflows to the Jiroft Dam using four state-of-the-art hybrid machine learning approaches: LSTM-GRU, CNN-LSTM, Attention-based LSTM, and Transformer models. Data from the study span a period of 14 years (2010-2023), encompassing 168 data points based on 12 monthly time series that incorporate hydrological and operational features, such as dam volume, precipitation, agricultural discharge, turbine operation, evaporation, and leakage. In addition to traditional performance measures, Taylor diagrams and violin plots can be used to enhance the assessment of models and provide additional insight. A leakage analysis showed evolution from baseline values of 766.3 L/s to the optimized scenarios, with the three-PAT setup yielding a significant 45% decrease to a final leakage rate of 424.57 L/s. This significant result verifies the effectiveness of PAT-based pressure management approaches in real networks. Seasonal analysis revealed winter peaks of 391.5 m³/s and autumn troughs of 56.2 m³/s, resulting in an average seven-fold range of extreme variability, thus posing challenges from a management viewpoint. The models' robustness was confirmed through the use of the Kolmogorov-Smirnov, Ljung-Box, and Breusch-Pagan tests. This work lays the groundwork for the development of reproducible frameworks in hydrological modeling using artificial intelligence and hence enables operational models for real-time regulation of water management under changing environmental conditions.
کلیدواژه‌ها
Machine learning؛ Hybrid models؛ LSTM-GRU؛ Dam inflow prediction؛ Taylor diagrams
آمار
تعداد مشاهده مقاله: 3


نماد الکترونیک 

 

سامانه مدیریت نشریات علمی. قدرت گرفته از سیناوب