A Hybrid Residual-CNN-BiLSTM-Attention Model for Lake Water Level Forecasting
8th International Congress on Human-Computer Interaction, Optimization and Robotic Applications, ICHORA 2026, Ankara, Türkiye, 21 - 23 Mayıs 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/ichora69329.2026.11537056
- Basıldığı Şehir: Ankara
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: CNN, CNN-BiLSTM-Attention, Deep learning, Hybrid modeling, Lake water level forecasting, LSTM
- Erciyes Üniversitesi Adresli: Evet
Özet
Lake water level forecasting is crucial for water resource planning and management. However, hydrological time series, including lake levels, often exhibit nonlinear behavior and can have complex temporal dependencies, making reliable prediction challenging. In recent years, deep learning approaches have been increasingly applied to model such complex dynamics in hydrological systems. In this study, the lake water level forecasting problem for Eǧirdir Lake in Türkiye is studied by proposing a Residual-CNN-BiLSTM-Attention model that combines convolutional feature extraction, residual learning, bidirectional temporal modeling, and attention mechanisms. The performance of the proposed model was compared with other deep learning models. The experimental results show that lake water levels exhibit strong temporal dependencies and that lagged water-level variables significantly improve forecasting accuracy. The proposed hybrid architecture achieved the best prediction performance among the evaluated models.