Forecasting reservoir levels using data-driven methods and typical scenarios

• Ano de defesa: 2024
• Autor(a) principal: Oliveira Júnior, Jordão Natal de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: aprendizado de máquina; bayesian methods; bayesian networks; convolução; convolution; dynamic bayesian networks; el niño and la niña; el niño e la niña; estatística; forecasting time series; machine learning; métodos bayesianos; nível de reservatório; previsão de séries temporais; redes bayesianas; redes bayesianas dinâmicas; reservoir level; statistics; teleconexões; teleconnections
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/18/18153/tde-30052025-100430/

Precise forecasting of reservoir levels is essential in hydroelectric power generation to maximize energy output and guarantee the sustainability of water resources. Forecasting water variables is intricate because of multiple uncertainties and intricate teleconnection couplings. Predicting the outcome and fully comprehending their relationship is challenging, thereby rendering it unfeasible to consolidate them into a single equation. A different strategy involves employing data-driven techniques for making predictions. This thesis presents a novel probabilistic model for forecasting reservoir levels. The proposed methodology combines traditional forecasting techniques with advanced probabilistic modeling methods. This integration addresses the uncertainties and complexities associated with hydrological and meteorological variables impacting reservoir levels. The model was extensively tested using historical data from the Furnas reservoir and energy equivalent reservoir, which included various seasonal cycles and hydrological fluctuations. The results demonstrate that it surpasses conventional deterministic-only forecasting models regarding predictive accuracy, providing strong evidence of its capability to enhance operational decision-making in hydroelectric power management. Additionally, it produces credibility intervals and assesses scenario-based forecasts for convenient evaluation and efficient computation.