| Evaluation of LSTM and GRU Deep Learning Models for Multi-Horizon Monthly Inflow Forecasting at Karaj Dam: A Decision-Support Approach for Tehran’s Drinking Water Supply |
| کد مقاله : 1488-IWWA |
| نویسندگان |
|
Masoumeh Zeinali *1، علی نشاط پور2، نیر صادقی3، عفت سادات کمالی4 1کارشناس منابع آب دفتر مطالعات و بررسی های فنی شرکت مهندسی آب و فاضلاب کشور 2مدیر کل مطالعات آب دفتر مطالعات و بررسی‌های فنی، شرکت مهندسی آب و فاضلاب کشور، 3رئیس گروه مطالعات آب دفتر مطالعات و بررسی‌های فنی، شرکت مهندسی آب و فاضلاب کشور، 4کارشناس دفتر مطالعات و بررسی‌های فنی، شرکت مهندسی آب و فاضلاب کشور |
| چکیده مقاله |
| Accurate forecasting of monthly inflows to dam reservoirs under non-stationary conditions induced by climate change and drought is a prerequisite for sustainable water resources management. This study compared the performance of two deep learning models, Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), for forecasting monthly inflows to Karaj Dam at one-, two-, and three-month forecast horizons. Monthly flow, precipitation, and temperature data spanning 288 months were standardized and used to derive seasonal and lagged features before being divided into training, validation, and testing datasets. Both models were trained using an identical architecture comprising 40 hidden units and a five-member ensemble. The testing results showed that GRU slightly outperformed LSTM in the aggregated case, with an RMSE of 3.84 m³/s and an NSE of 0.695, compared with an RMSE of 3.95 m³/s and an NSE of 0.678 for LSTM. The advantage of GRU was more evident at the shorter forecast horizons, whereas LSTM showed a slight advantage at the three-month horizon. Both models reproduced the seasonal pattern of inflows well; however, a systematic overestimation of approximately 14.6% was observed during the testing period, indicating the sensitivity of the models to changes in the hydrological regime. The findings suggest that these models can serve as supportive tools for managing Tehran's drinking water supply, provided that bias correction and periodic model updating are implemented. |
| کلیدواژه ها |
| inflow forecasting, LSTM, GRU, Karaj Dam, hydrological non-stationarity, water resources management. |
| وضعیت: پذیرفته شده برای ارسال فایل های ارائه پوستر |
