Reservoir level control in hydroelectric power plants has importance for the stability of the electric power supply over time and can be used for flood control. In this sense, this paper proposes a sequence-to-sequence (Seq2Seq) long short-term memory (LSTM) neural network model with an attention mechanism and wavelet transform for noise reduction to predict reservoir levels for a one-hour horizon in advance. Accurate reservoir level predictions are essential components for effective water management. The proposed model was evaluated on real-world data and compared with different setups of LSTM, besides ensemble learning models such as bagging, boosting, and stacking. Results demonstrate that the proposed approach outperforms other LSTM models with a mean squared error of 0.0020 and a mean absolute error of 0.0347. Moreover, the wavelet transform with a BayesShrink thresholding method and six levels was the most effective for denoising the input signal. The proposed Wavelet-Seq2Seq-LSTM model provides reservoir managers with accurate and timely predictions of water levels, allowing for better decision-making in dam management under emergency conditions. Advanced techniques such as Seq2Seq, attention mechanism, and wavelet transform to make the proposed approach a promising reservoir-level prediction solution.
Wavelet-Seq2Seq-LSTM with attention for time series forecasting of level of dams in hydroelectric power plants
Stefenon, Stefano Frizzo;
2023-01-01
Abstract
Reservoir level control in hydroelectric power plants has importance for the stability of the electric power supply over time and can be used for flood control. In this sense, this paper proposes a sequence-to-sequence (Seq2Seq) long short-term memory (LSTM) neural network model with an attention mechanism and wavelet transform for noise reduction to predict reservoir levels for a one-hour horizon in advance. Accurate reservoir level predictions are essential components for effective water management. The proposed model was evaluated on real-world data and compared with different setups of LSTM, besides ensemble learning models such as bagging, boosting, and stacking. Results demonstrate that the proposed approach outperforms other LSTM models with a mean squared error of 0.0020 and a mean absolute error of 0.0347. Moreover, the wavelet transform with a BayesShrink thresholding method and six levels was the most effective for denoising the input signal. The proposed Wavelet-Seq2Seq-LSTM model provides reservoir managers with accurate and timely predictions of water levels, allowing for better decision-making in dam management under emergency conditions. Advanced techniques such as Seq2Seq, attention mechanism, and wavelet transform to make the proposed approach a promising reservoir-level prediction solution.File | Dimensione | Formato | |
---|---|---|---|
1-s2.0-S0360544223007442-main.pdf
non disponibili
Descrizione: paper
Tipologia:
Documento in Post-print
Licenza:
NON PUBBLICO - Accesso privato/ristretto
Dimensione
2.42 MB
Formato
Adobe PDF
|
2.42 MB | Adobe PDF | Visualizza/Apri Richiedi una copia |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.