https://repositorio.unifei.edu.br/jspui/handle/123456789/88 Tue, 29 Sep 2026 06:04:42 GMT 2026-09-29T06:04:42Z https://repositorio.unifei.edu.br/jspui/handle/123456789/4470 Framework metodológico de modelagem e simulação híbrida com aprendizagem por reforço aplicado à racionalização de pontos de ônibus The rationalization of bus stops represents a major challenge in urban public transportation planning, with direct implications for population accessibility, system operational efficiency, and travel quality. Despite advances in urban computing, there remains a lack of methodological frameworks capable of integrating Agent-Based Simulation (ABS), Discrete-Event Simulation (DES), and Reinforcement Learning (RL) in a structured and synergistic manner to support the rationalization of public transportation infrastructure. To address this conceptual and methodological gap, this dissertation aims to develop a Hybrid Modeling and Simulation (HMS) methodological framework to support the rationalization of urban public transportation bus stop locations by integrating ABS, DES, and RL based on measures of accessibility and system usage. The methodological approach comprised a Systematic Literature Review (SLR) conducted according to the PRISMA protocol, the theoretical formulation of the framework, and its computational implementation in the AnyLogic platform, using a georeferenced case study of the municipality of Passos, Minas Gerais, Brazil, for validation. The computational model integrated a spatially distributed synthetic population, origin–destination (OD) matrices, a georeferenced road network obtained from OpenStreetMap, urban Points of Interest (POIs), and the behavioral and operational processes of the public transportation system. Agent behavioral parameters were calibrated and statistically validated against real boarding and electronic ticketing data. During the rationalization stage, a tabular Q-Learning algorithm employing an ε-greedy policy was implemented, in which the reward function combined variations in the global accessibility indicator with variations in the number of public transportation users. In an experiment configured with 10 episodes and 2,000 training steps per episode, the model identified alternative spatial configurations for the bus stop network, producing an improvement of approximately 8% in the global accessibility indicator and an increase of more than 10% in the number of public transportation users compared with the calibrated and validated baseline scenario. These improvements were achieved exclusively by modifying bus stop locations while keeping routes, itineraries, and all other operational characteristics of the system unchanged. The resulting locations were interpreted as candidate solutions and subsequently subjected to physical and topological feasibility assessments. The results demonstrate the methodological feasibility and practical applicability of the proposed framework as an evidence-based decision-support tool for urban transportation planning. Scientifically, this dissertation contributes by proposing a systematic, replicable, and adaptable methodological framework that expands the possibilities for integrating hybrid simulation, reinforcement learning, and spatial analysis to support the rationalization of urban public transportation. Tese Wed, 26 Aug 2026 00:00:00 GMT https://repositorio.unifei.edu.br/jspui/handle/123456789/4470 2026-08-26T00:00:00Z https://repositorio.unifei.edu.br/jspui/handle/123456789/4467 In-context learning versus trained models: generative AI and machine learning for surface roughness prediction in dry Ti-6Al-4V turning Surface roughness Ra is a critical quality indicator in dry turning of Ti-6Al-4V titanium alloy, a material widely used in aerospace and biomedical applications. Empirical powerlaw models remain the industrial standard but cannot represent interaction effects or non-linear responses. Machine learning offers greater flexibility, yet its deployment is constrained by small experimental datasets and the need for rigorous validation protocols. Large language models represent an alternative predictive paradigm, potentially encoding implicit knowledge of machining physics that may support quantitative prediction through in-context learning alone. This work compares five predictive models, comprising two conventional mathematical approaches (Power Law and RSM) and three machine learning models (SVR, Gaussian Process Regression, and XGBoost), against three LLMs (Claude Sonnet 4.5, GPT-4o-mini, and Gemini 3 Flash Preview) for Ra prediction in Ti-6Al- 4V turning. A Central Composite Design with k = 3 factors and two carbide inserts of distinct nose radii yielded N = 38 observations. The conventional and ML models were evaluated under nested LOOCV with 5-fold grid search where applicable. LLMs were accessed via API at temperature zero under two prompting strategies, few-shot and chain-of-thought, across three independent runs, with results reported as mean ± standard deviation. Claude Sonnet 4.5 in few-shot mode achieved the best overall performance (RMSE = 0.288 ± 0.016 μm, R2 = 0.985), outperforming SVR, the best trained model (RMSE = 0.341 μm), by 16% without task-specific training or fine-tuning. The effect of prompting strategy was model dependent: few-shot prompting outperformed chain-ofthought for Claude and GPT, whereas Gemini CoT achieved a substantially lower mean RMSE than Gemini FewShot but exhibited markedly greater run-to-run variability. This variability was especially pronounced for Gemini CoT, whose RMSE ranged from 0.31 to 0.99 μm across identical executions. The results show that a capable LLM supplied with well-structured experimental demonstrations can surpass optimised trained models for Ra prediction without task-specific training, while also demonstrating that predictive performance and stability depend strongly on both model architecture and prompting strategy. Tese Thu, 06 Aug 2026 00:00:00 GMT https://repositorio.unifei.edu.br/jspui/handle/123456789/4467 2026-08-06T00:00:00Z https://repositorio.unifei.edu.br/jspui/handle/123456789/4464 Teletrabalho no contexto brasileiro: fatores socioeconômicos, ambientais e de transporte a partir de uma abordagem de aprendizado de máquina Telework emerges as a potential tool for reducing commuting, lowering external costs, and promoting quality of life. In this context, the main objective of this research is to investigate the adoption of telework in the Brazilian context from a machine learning perspective. Initially, a comprehensive conceptual model was developed to examine the relationships between telework and socioeconomic, transportation, and environmental variables, based on a systematic literature review of studies published up to 2025. The temporal analysis of the literature highlights the growing academic interest in the topic, emphasizing the profile of teleworkers, characterized by factors such as age, educational level, income, and occupation type. Additionally, interactions between telework and urban aspects are explored, including the adoption of sustainable transportation modes, such as walking and cycling, as well as its relationship with urban sprawl. The impacts on quality of life are also discussed, particularly regarding schedule flexibility and work–life balance. The thesis analyzes the evolution of telework in Brazil from 2022 to 2025 through the application of machine learning models to representative microdata from the Continuous National Household Sample Survey, covering approximately 210,000 households per period. A standardized processing workflow was implemented, including data preprocessing, handling of missing values, class balancing through random undersampling, variable encoding and normalization, as well as stratified data splitting with k-fold cross-validation. Nine classification algorithms were evaluated, including Multinomial Logistic Regression, Decision Trees, Random Forest, XGBoost, Support Vector Machines, MARS, and Neural Networks, with hyperparameter tuning performed using the ANOVA racing method. Model performance showed consistently high results (ROC AUC > 0.80) across all analyzed periods. The analysis of variable importance indicates that the main determinants of telework remain relatively stable over time, although their contributions vary, with a notable increase in the influence of gender in more recent periods. The results demonstrate that telework in Brazil is driven by a combination of sociodemographic and occupational factors, reinforcing its selective nature. Overall, this thesis contributes to a deeper understanding of the multiple dimensions of telework by integrating theoretical and empirical evidence. Furthermore, the findings provide relevant insights for policymakers, organizational managers, and urban planners, highlighting how telework shapes urban dynamics and can support efforts to address contemporary urban challenges. Tese Wed, 20 May 2026 00:00:00 GMT https://repositorio.unifei.edu.br/jspui/handle/123456789/4464 2026-05-20T00:00:00Z https://repositorio.unifei.edu.br/jspui/handle/123456789/4455 Análise da vulnerabilidade energética dos aproveitamentos hidrelétricos do sistema interligado nacional frente às mudanças climáticas e aos padrões de usos consuntivos da água Climate change (CC), resulting from natural processes, has been intensified by anthropogenic activities, mainly due to the increased emission of greenhouse gases (GHGs). These changes have affected several sectors of society, including the environment, water resources, and electricity generation, among others. The latest climate assessment reports published by the Intergovernmental Panel on Climate Change (IPCC) indicate future reductions in precipitation in some regions of Brazil, an increase in the frequency and intensity of extreme events such as droughts and floods—particularly in the North and Northeast regions—and a rise in global average temperatures. These factors, combined with changes in water consumption patterns, which are also influenced by climate conditions, may further impact the Brazilian Electric Sector (BES) due to its strong dependence on hydropower generation. This study assesses the impacts of climate change and water demands on streamflows in the main river basins and hydropower plants in Brazil, as well as their effects on electricity generation. For this purpose, the MGB-IPH hydrological model was applied to simulate the historical period (1970–2014) and future scenarios (2015–2100) under climate change conditions. The simulations were driven by projections from five climate models from the NASA Earth Exchange Global Daily Downscaled Projections/Coupled Model Intercomparison Project Phase 6 (NEX-GDDP/CMIP6) dataset, considering two greenhouse gas emission scenarios, namely the Shared Socioeconomic Pathways (SSP2-4.5 and SSP5-8.5), as well as projections of consumptive water uses provided by the Brazilian National Water and Basic Sanitation Agency (ANA). The energy assessment proposed in this study is based on Natural Inflow Energy (NIE), since this variable is directly related to streamflows at hydropower plants. The results show that all five climate models consistently project reductions in streamflows for the future scenarios at the major hydropower plants evaluated. However, the models differ regarding the magnitude of these reductions. The NIE projections obtained with the CanESM5 model indicate the largest decreases, reaching reductions of up to 100% in the Paraguaçu River Basin. In contrast, the projections generated by the INM model indicate positive variations, with increases approaching 10%. For the remaining climate models, reductions in NIE average around 20%. The most significant findings of this study highlight how climate change and increasing water consumption may affect hydropower generation and further increase the vulnerability of the sector. Therefore, the results presented here can provide valuable information to support planning and decision-making by responsible agencies, enabling the implementation of measures to mitigate future impacts on hydropower generation and enhance the resilience of the Brazilian electricity sector. Tese Fri, 13 Mar 2026 00:00:00 GMT https://repositorio.unifei.edu.br/jspui/handle/123456789/4455 2026-03-13T00:00:00Z