http://repositorio.ufc.br/handle/riufc/461 2026-07-26T18:56:01Z http://repositorio.ufc.br/handle/riufc/86948 Título: Previsão da produção de hidrogênio usando modelos de aprendizagem de máquinas: estudo comparativo de cenários de conversão de energia Autor(es): Sacramento, Diosael Ferreira Abstract: Green hydrogen emerges as a strategic alternative for the global energy transition, especially in countries with high wind energy potential, such as Brazil. This dissertation aims to forecast and analyze green hydrogen production from wind energy in three Brazilian states, Ceará, Rio Grande do Sul, and Bahia, using machine learning-based models. The methodology involved the processing and modeling of wind speed time series data collected between 2019 and 2021. Wind speed forecasts were used as input for five distinct wind-to-hydrogen conversion models (Models 1 to 5), ranging in complexity from fixed-efficiency scenarios to simulations incorporating real efficiency curves of wind turbines and electrolyzers. Five forecasting algorithms were evaluated: XGBoost, Random Forest (RF), Support Vector Regression (SVR), K-Nearest Neighbors (KNN), and Long Short-Term Memory (LSTM). In the Ceará case, Model 4, which considered the real efficiency curve of wind turbines, achieved the best absolute performance, with an RMSE of 45.19 kg_H2/day, MAE of 31.72 kg_H2/day, and a coefficient of determination (R2) of 0.88, indicating high accuracy. Model 2, which used variable electrolyzer efficiency, stood out for its lowest training time (649 s) and lowest percentage error (MAPE = 24.02%), making it suitable for applications with computational constraints. Models 1, 3, and 5 showed inferior performance, with RMSE above 70 kg_H2/day and R2 below 0.75. Among the machine learning algorithms, Random Forest (RF) delivered the best overall performance, with the lowest RMSE (42.45 kg_H2), lowest MAE (29.45 kg_H2), highest R2 (0.5406), and highest Skill Score (0.8415), demonstrating robustness and consistency. SVR and LSTM performed competitively, with similar errors and moderate training times. XGBoost, while efficient in many scenarios in the literature, was the least accurate in this study, with the highest MAPE (30.36%) and lowest R2 (0.4928), although it achieved the fastest training time (692 s), potentially benefiting applications requiring rapid inference. The KNN model was the least effective in both absolute and percentage error metrics. Keywords: Green hydrogen; wind energy; XGBoost; machine learning; time series forecasting; renewable energy; predictive modeling. Tipo: Dissertação 2026-01-01T00:00:00Z http://repositorio.ufc.br/handle/riufc/86382 Título: Modelos de aprendizado de máquina para estimativa de temperatura de operação de módulos fotovoltaicos em clima semiárido Autor(es): Costa, Antônio Wellington Dantas da Abstract: The growing demand for clean and renewable energy sources drives the expansion of photovoltaic (PV) systems in semi-arid regions, where high solar irradiance is coupled with high temperatures that reduce energy conversion efficiency. The operating temperature of PV modules (T mod ) is a critical variable for sizing, monitoring, and performance optimization. This work implements and evaluates Machine Learning (ML) models for estimating T mod under tropical semi-arid climate conditions. Using experimental data collected at the Alternative Energy Laboratory of the Federal University of Ceará (LEA-UFC), in Fortaleza, during two years of monitoring (May 2023 to May 2025), models organized into three categories were implemented and compared: ML algorithms (Linear Regression - LR, K-Nearest Neighbors - KNN, Random Forest - RF, and eXtreme Gradient Boosting - XGBoost), hybrid semi-empirical correlations (linear, non-linear, and rational), and consolidated physical-empirical models (Ross, Faiman, and PVsyst). The final dataset comprised 273,310 observations over two years, using five meteorological predictor variables: solar irradiance (G), ambient temperature (T a ), wind speed (V w ), and solar angles. The results demonstrated the superiority of ML models, particularly KNN, which achieved a Mean Absolute Error (MAE) of 0.83°C, Root Mean Square Error (RMSE) of 1.44°C, and a coefficient of determination R 2 = 0.97, representing a reduction of approximately 77% in MAE compared to physical models. The hybrid semi-empirical correlations showed intermediate performance, with a 22% reduction in RMSE and a 68% reduction in systematic bias compared to conventional physical models. Feature importance analysis revealed that G and T a account for 70–80% of the predictive capacity, while V w contributed less than 2%, a consequence of the urban microclimate with high roughness and low average speeds. The superior performance of ML is explained by the stability of the local semi-arid climate, which facilitated pattern learning, and by the algorithms’ ability to automatically adjust to local data, capturing complex thermal variations under high irradiation that generic physical models fail to predict. This work deepens the understanding of PV module heating in the Brazilian semi-arid region, providing a foundation for optimization, cooling, and intelligent control strategies. Tipo: Dissertação 2026-01-01T00:00:00Z http://repositorio.ufc.br/handle/riufc/85728 Título: Estudo integrado da macaúba (Acrocomia aculeata): caracterização físico-química, pirólise termogravimétrica lenta e de determinação da energia de ativação por métodos isoconversionais Autor(es): Coimbra, Kennedy Romualdo Guedes Abstract: The continuous use of fossil fuels, the growing concern with global warming, and environmental pollution have driven the search for renewable energy sources and alternative technologies. In this context, the use of lignocellulosic biomass, such as macaúba (Acrocomia aculeata), has stood out as a promising renewable raw material to produce bioenergy and biofuels through thermochemical conversion. This study investigated the fractions of macaúba (epicarp, mesocarp, endocarp, and almond cake) through physicochemical analyses and, by means of thermogravimetric pyrolysis, the reaction kinetics. Proximate and ultimate analyses indicated that the mesocarp has the highest volatile content (87%) and a higher heating value (HHV) of 22 MJ/kg, being the most promising fraction for slow pyrolysis. FTIR analyses confirmed the predominant presence of cellulose, hemicellulose, and lignin. Thermogravimetric analyses were carried out in an inert atmosphere with heating rates of 5, 10, 15, and 20 °C/min, showing that mesocarp and epicarp decompose rapidly between 250 and 400 °C, while endocarp and almond press cake present slower degradation, up to 600 °C. With the pyrolysis kinetics, the activation energy (Ea) was calculated using the isoconversional methods of Friedman (FR), Kissinger-Akahira-Sunose (KAS), and Flynn-Wall-Ozawa (FWO), presenting average values of 186 to 202 kJ/mol for the mesocarp and 232 to 235 kJ/mol for the endocarp. Under oxidative atmosphere, the almond press cake reached Ea values up to 342 kJ/mol. The results validated the potential of macaúba for energy applications with integrated residue utilization, with epicarp and mesocarp indicated to produce bio-oil and endocarp and almond press cake to produce biochar. Tipo: Dissertação 2025-01-01T00:00:00Z http://repositorio.ufc.br/handle/riufc/84215 Título: Análise de desempenho e emissões de simulações de caminhões híbridos elétricos com autonomia extendida (REEV) Autor(es): Uchoa Filho, Francisco Elmo Lima Abstract: The worsening global climate crisis demands urgent strategies to mitigate environmental impacts in a wide range of human activities. The transport sector is responsible for around 23% of all global CO 2 emissions, and of this total, 29.4% comes from trucks. On the other hand, little research has been conducted on this transport mode. Given the global trend toward the adoption of electric trucks, the strategies adopted in many countries are evaluated. Brazil’s history with biofuels, especially ethanol, indicates that range extended electric vehicles (REEV) are the ideal solution for the country. This work proposes a study of the performance of different truck configurations with power trains (the set of components responsible for propelling the vehicle) in the REEV configuration. Four truck models are proposed, two with a maximum load of 23t and two with 57t, three power levels for the ethanol-powered combustion engine, combined with up to 17 different groups of battery banks in parallel. Thus, one-dimensional powertrain models are developed on the Realis Ignite platform based on commercial components and vehicles. Aided by routines developed in Python, the various powertrain configurations are simulated in successive standardized driving cycles. The combinations with the lowest specific energy consumption are selected, being the minimum 0.633MJ/t.km. The emissions of CO 2 , NO x , CO, and HC were evaluated according to national standards, equivalent to Euro VI, with all simulations being in compliance. Tipo: Dissertação 2025-01-01T00:00:00Z