Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
ECHO STATE NETWORKS APPLIED TO HYDROLOGICAL SERIES FORECASTING
Submission Author:
LUIZ EDUARDO THOMAZ , PR , Brazil
Co-Authors:
José Henrique Kleinübing Larcher, LUIZ EDUARDO THOMAZ, Leandro dos Santos Coelho, Viviana Mariani
Presenter: LUIZ EDUARDO THOMAZ
doi://10.26678/ABCM.COBEM2023.COB2023-1027
Abstract
Time series analysis and forecasting are essential for effective water resource management, especially in countries like Brazil that heavily rely on hydroelectric power. This study investigates the effectiveness of employing various decomposition methods in conjunction with Echo State Networks (ESNs) to forecast natural water flow in two Brazilian hydroelectric reservoirs: Itaipu and Furnas. The research aims to evaluate ESNs in hydrological series forecasting and enhance accuracy by integrating decomposition methods with ESNs. The tested forecasting horizons are 7, 14 and 21 days ahead. The study utilizes ESNs tuned with the Coyote Optimization Algorithm (COA) and integrates them with the decomposition methods Variational Mode Decomposition (VMD), Empirical Wavelet Transform, and Empirical Mode Decomposition. The performance of these models is compared using metrics mean absolute error (MAE), mean absolute percentage error, root mean square error, and root mean square logarithmic error. The results demonstrate that incorporating decomposition methods into ESNs enhances their accuracy in forecasting hydrological series. VMD consistently outperforms other methods across all forecast horizons, reducing MAE by 43.70% to 88.88% compared to other models. The statistical significance of the improvement achieved by employing VMD is confirmed by the Diebold-Mariano test at a 1% level.
Keywords
Time series forecasting, signal decomposition, echo state networks, hydroelectric reservoirs
