Stock market dynamics show that assets within the same sector tend to exhibit similar behavior. In contrast, stocks from different sectors frequently display uncorrelated movements, making the adoption of a single trading model for the entire market inadequate. In this work, a multicriteria recommendation approach for trading models is proposed for the Brazilian stock market (B3), grounded in the TODIM (TOmada de Decisão Interativa Multicritério) method and Prospect Theory, with evaluation at two levels of granularity: sector-level and individual stock-level, the latter being the main contribution of this work. The dataset comprises 239 stocks organized into eight economic sectors. Technical indicators are extracted as input variables, and Mutual Information-based feature selection is applied to identify the most relevant features, thereby mitigating redundancy and irrelevance. Twelve machine learning algorithms are trained via walk-forward analysis to generate trading signals, and their performance is evaluated using eight backtesting indicators. The resulting recommendation algorithm is assessed in both its sector-level and individual stock-level versions. Experiments conducted during the high-volatility period of the COVID-19 pandemic (Jan.–Jun. 2020) demonstrate that both versions outperform the Ibovespa index in capital protection and risk-adjusted return metrics. Furthermore, generalization metrics confirm the robustness of the recommendation algorithm. Finally, long-short portfolios constructed using the recommendations demonstrated a significant capacity to protect against systemic risk.

**

**

**

**

Estimating the point of occurrence of a short-circuit in a transmission line is a crucial task, impacting the reliability of the power system and the costs associated with outages. In seriescompensated lines, where overvoltage protection of the capacitor is provided by a varistor—a nonlinear device—fault location becomes more complex and challenging. This thesis aims to present two proposals, with potential for practical application by power utilities, for locating faults in transmission lines equipped with series capacitors. The formulation of the problem uses data typically available to engineering teams in the field, such as transmission line parameters and short-circuit oscillography. The first approach, called the Proposed Network Methodology (M_RP), does not rely on Thevenin voltages and impedances to represent the power system connected to the transmission line terminals. Training and validation files for the artificial neural network (ANN) are generated using electromagnetic transient simulation
software, based on phasor estimates of the voltages at both terminals during the fault. These data are used to simulate current signals corresponding to different fault locations, types, and fault resistances. The ANN receives the current values, associates them with the fault location, and, once trained and validated, is capable of receiving current signals from an actual fault and estimating the fault distance. The second approach, referred to as the Optimization Methodology (M_OT), involves the development of objective functions tailored to each fault type. In this case, the varistor overvoltage protection curve provided by the manufacturer is not used. Instead, a polynomial with fixed coefficients—originally developed for stability studies— is adopted to describe a linearized model of the varistor/capacitor assembly’s resistance and reactance. In the proposed nonlinear optimization problem, the fault distance and the varistor conduction threshold current—which corresponds to the maximum current supported by the capacitor—are the minimization variables. Simulated files containing fault event currents and voltages, as well as field-recorded oscillographs, are used to test the methodologies. In both cases, M_RP and M_OT yielded promising results, enabling fault location with errors below 1%. Although the results obtained with M_RP are particularly relevant for compensated transmission lines, the methodology is also applicable to faults in uncompensated lines.

**

Financial time series help to understand market behavior and attract interest from fields such as economics, mathematics, and physics. In this work, we present a model for the characterization of price series inspired by differential equations that describe the dynamics of a physical system composed of linearly coupled harmonic oscillators. After discretizing the equations, a vector autoregressive (VAR) model was obtained. The model was evaluated using daily closing prices of 65 stocks that were part of the Ibovespa index from 12/17/2004 to 05/29/2020. We verified the stationarity of these series using the Dickey-Fuller test. The residuals resulting from applying the series to the model were found to be uncorrelated according to the Breusch-Godfrey test. The parameters of the coupled oscillators were determined from the VAR model coefficients, and from this, we defined quantities analogous to the energies of the system. The graphs of the energy time series showed large peaks that were associated with events that caused shocks in financial markets. We evaluated the empirical distribution of some model parameters, which mostly exhibited heavy tails that deviate from the normal distribution. We investigated the existence of linear correlation between the oscillator parameters, liquidity, and the average trading volume of the stocks. We calculated the Hurst exponent for quantities analogous to kinetic and potential energy, which showed a high level of anti-persistence. The analysis of the singularity spectrum width indicated the presence of nonlinear correlations in the series of such quantities. We applied the proposed model to forecast the price trend of financial assets, using 1-minute candle closing prices of stocks that composed the Ibovespa index in the first quarter of 2024. We rejected the null hypothesis that the forecast accuracies are statistically equal to those of a random predictor equivalent to a fair coin toss. Furthermore, we used the model to forecast the absolute return of asset prices. By comparing the results with those of a moving average-based model, we defined the accuracy of the oscillator-based model. In this way, we rejected the null hypothesis that the accuracies are equal to 0.5. Using a Euclidean distance metric, we found that the prices predicted by the harmonic oscillator-based model were closer to the actual prices than those predicted by the moving window model.

This thesis investigates the design of sampled-data controllers for Linear Parameter Varying (LPV) systems with actuators subject to magnitude and rate saturation. The objective is to guarantee stability while allowing larger admissible sampling intervals. The growing presence of autonomous and networked systems motivates the development of control methods that efficiently integrate continuous-time processes with digital controllers, even under practical constraints such as communication delays, packet losses, and network congestion. Sampled-data control provides a framework for handling the interplay between continuous-time dynamics and discrete-time actions, enabling both stability and performance guarantees. In LPV systems, additional challenges arise because the plant evolves continuously while the controller acts at discrete instants. This mismatch may cause asynchronous parameter measurements and potential conservatism. This thesis proposes methods to overcome practical limitations in LPV sampled-data control. The focus is on mitigating the adverse effects of actuator saturation and reducing design conservatism using Lyapunov theory and linear matrix inequality (LMI)-based conditions. First, a discrete-time model-based approach is developed to design parameter-dependent state-feedback controllers, ensuring regional stability under actuator saturation and amplitude-limited disturbances. Simulations and real-time experiments on nonlinear coupled tank systems validate this design approach. Second, new conditions for the negative definiteness of quadratic functions are introduced using Young’s inequality, Pólya’s theorem, and Bézier/de Casteljau representations. These conditions guarantee recursive monotonic convergence, yield less conservative results, and are employed to ensure the negativity of the time derivative of the Lyapunov candidate function. Finally, the parameter mismatch between continuous-time plant parameters and the control signal, which depends on sampled measurements, is addressed through a novel application of the mean value theorem. This approach enables larger sampling intervals. The framework is further extended to aperiodic sampled-data control of LPV systems with saturated actuators. This extension avoids the need for discretization of plant models and ensures local stability through parameter-dependent looped-functionals and generalized sector conditions. Numerical examples and real-time control experiments show feasibility and effectiveness, establishing new approaches for sampled-data control of LPV systems under realistic actuator and network constraints.

This study aims to contribute to the advancement and dissemination of studies in the field of Intellectual Property, especially from the perspective of Information Science. To this end, the primary objective is to propose a systematic process for the collection and processing of extensive sets of patent bibliographic data, with the purpose of circumventing the limitations inherent to open access repositories of patent documents. The basis of this study is supported by the development of a set of strategies that enable the construction of a protocol, that is, a set of rules and procedures that enable the collection and processing of patent bibliographic data. Based on the proposition of eleven axioms, the patent repositories that are used as a source of data in the collection process are selected. In sequence, strategies are outlined that enable the collection, treatment and aggregation of the collected data sets, in order to culminate in a local database, capable of supporting analyzes from different perspectives. Furthermore, to enrich the established database, complementary data sources are incorporated, such as records from the curriculum base of the Lattes Platform of the National Council for Scientific and Technological Development, which provide information about researchers who have patents reported in their CVs. And finally, to validate the effectiveness of the proposed protocol, it was applied in the process of collecting all patents filed in Brazil in the period between 1900 and 2022 and which are available in the Espacenet repository. Thus, as a result of the case study, the collected data is subjected to a detailed analysis from different perspectives, making use of advanced techniques such as bibliometrics, patentometrics, text mining and natural language processing. The results obtained are considered relevant and the strategies outlined in this study contribute and are pertinent to research involving the analysis of patent documents, thus offering an effective alternative to the challenges faced in collecting large sets of patent bibliographic data.

The present project will deal with the study of the robustness of the behavior of steel pipes of high quality and their respective collapse pressures that are found installed in oil platforms. The deeper the ocean waters and the thicker the layer submerged land under which oil and gas deposits are located, the greater the challenges technologies to extract them. One of the critical points of extraction is to have reliable methods of analysis and monitoring of pipes and systems that support high pressures and large differences in temperature. The focus is primarily on describing how data can be manipulated to understand the strengths, weaknesses and overall suitability of a product considering its application end, and about the outputs of a system that uses statistical analysis to predict collapse pressure of seamless steel tubes based on historical data and can be used as a design tube design virtually simulated by computer analyses. The high computing performance will be used to develop comparative views between the techniques statistics and computational intelligence in the analysis of the robustness of the tubes.

This dissertation addresses the challenge of ensuring transactional integrity and reducing costs in corporate governance through blockchain technology. We propose an on-chain methodology for certifying, registering, and querying institutional transactional status. Our decentralized governance approach utilizes consensus mechanisms and smart contracts to automate and enforce business rules. The framework aims to reduce the transaction costs associated with contractual measurement reports and enhance overall transactional integrity. We provide a detailed exploration of how blockchain technology can be effectively harnessed to offer a robust solution to these challenges, setting the stage for our proposed solution and its potential impact on corporate governance.

**

Nowadays, with the advent of climate change and human activity, the occurrence of large-scale forest fires causing irreversible damage to the ecosystem is becoming increasingly common. This motivates research into different ways to prevent and monitor these events from happening. One possible prevention method is monitoring the areas most susceptible to fires and utilizing computer vision techniques for early detection of these events while they are still small, thus accelerating the response of responsible authorities, reducing environmental damage, and restoration costs. Currently, systems using convolutional neural networks (CNNs) for this purpose rely on large and complex infrastructure due to high computational costs and memory consumption, which hinders scalability and accessibility. Therefore, a system will be proposed that utilizes a CNN-based fire detector, but is self-sustainable and low-power to increase scalability and accessibility in more remote areas, requiring less maintenance and infrastructure. The approach involves removing less important layers of the neural network during training to reduce computational costs while maintaining the model’s high detection capability.

In this work, we theoretically study a finite and undamped two-mode optomechanical model consisting of a high quality optical cavity containing a thin, elastic, and dielectric membrane. The main objective is to investigate the precursors of quantum phase transition in such a model by studying the behavior of some observables in the ground state. By controlling the coupling between membrane and modes, we find that the two lowest energy eigenstates become degenerate, as is indicated by the behavior of the mean value of some operators and by other quantifiers as a function of the coupling. Such degenerate states are characterized by a coherent superposition of eigenstates describing one of the two modes preferentially populated and the membrane dislocated from its equilibrium position due the radiation pressure (Schrödinger’s cat states). The delocalization of the compound system photons+membrane results in an increase in fluctuations as measured by Robertson-Schrödinger uncertainty relations

In this work, models with massive extensions of electrodynamics are studied, either through theories that violate Lorentz symmetry or not. First, we study the particular case of a model that, although massive, preserves a residual gauge invariance. The physical consistency of the model is evaluated through its classical field equations and the analysis of the propagator poles of the vector field. Then, in a more general way, the physical equivalence, at the classical and quantum levels, of dual models is studied. To obtain the dual model (gauge invariant) to the original massive theory, the Noether dualization method is used. At the classical level, physical equivalence is investigated by calculating the electron-electron scattering cross section (Moller scattering) at tree level. In the quantum case, we will present the on-shell equivalence for a single loop amplitude and then we will show that the equivalence holds for off-shell. We will start by considering an electron eigenenergy of an extended massive electrodynamics and its corresponding gauge invariant dual model.

Many significant chemical processes benefit from advancements in heterogeneous catalysts, particularly those catalyzed by metallic surfaces, which are often influenced by imperfections like kinks and dopants. In our study, we first investigated the impact of single-atom Rh dopants on the adsorption behavior of OH and CO on stepped Ag(211) surfaces using density functional theory calculations with van der Waals correction. The stability of Rh dopant on the surface is highest at highly-coordinated sites, but when both molecules are present, the least coordinated site is favored. This shift is attributed to dopant-induced charge redistribution and a distinct peak in the density of states. Molecules near the dopant can modify the local charge distribution and affect dopantsurface interactions. This may shift the dopant to favor less coordinated adsorption sites compared to the surface without molecules. OH strongly chemisorbs to silver surfaces, and its bond is stronger with Rh dopant due to their electronegativity difference. In contrast, CO physisorbs on non-doped silver but chemisorbs with single-atom Rh dopants. In the second phase of our study, we investigated the reduction process of CO2 using the computational hydrogen electrode (CHE) method, analyzing clean and Fe, Co, Ni, Cu, Rh and Ir single-atom doped Ag(211) surfaces. We also assessed whether the proposed materials are susceptible to OH poisoning or enhance the competing HER reaction (HER). Our calculations revealed weak electronic density mixing between the host and dopant, characterized by sharp peaks in the density of states near the Fermi energy, except in the case of copper-doped SAAs. The adsorption energy of intermediates varies across different substrates, leading to unique reaction profiles. Copper-doped SAAs tend to spontaneously desorb CO, unlike other fourth-period elements (Fe, Co, Ni) that bind to CO and may enhance the HER. In contrast, Ir and Rh dopants are promising electrocatalysts, favoring CO reduction and suppressing the HER at lower potentials than copper. Moreover, they are not prone to poisoning by OH radicals.

Muitos processos químicos importantes se beneficiam dos avanços em catalisadores heterogêneos, especialmente aqueles que são catalisados por superfícies metálicas, que muitas vezes são influenciados por imperfeições, como degraus e dopantes. Na primeira parte deste trabalho, foi investigado o impacto da dopagem com um único dopante Rh no comportamento de adsorção de OH e CO em superfícies de Ag(211) com degraus, usando cálculos de teoria do funcional da densidade (DFT) com correção de van der Waals. A estabilidade do dopante Rh na superfície é maior em sítios com alta coordenação, mas, quando as moléculas estão presentes, os sítios menos coordenados são favorecidos. Essa mudança é atribuída à redistribuição de carga induzida pelo dopante e a um pico distinto na densidade de estados. Moléculas próximas ao dopante podem modificar a distribuição de carga local e afetar as interações entre dopante e superfície. Isso pode deslocar o dopante para favorecer locais de adsorção menos coordenados em comparação com a superfície sem moléculas. O OH se adsorve fortemente em superfícies de prata, e sua ligação é mais forte com o dopante de Rh devido à diferença de eletronegatividade entre eles. Em contraste, o CO se adsorve fisicamente na superfície não dopada, mas se adsorve quimicamente em superfícies dopadas. Na segunda fase do estudo, foi investigado o processo de redução de CO2 usando o método computacional do eletrodo de hidrogênio (CHE), analisando superfícies de Ag(211) limpas e dopadas com um único átomo de Fe, Co, Ni, Cu, Rh ou Ir. Também avaliamos se os materiais propostos são suscetíveis a envenenamento por OH ou favorecem a reação de evolução de hidrogênio (HER), reação competidora. Nossos cálculos revelaram uma fraca mistura de densidade eletrônica entre o hospedeiro e o dopante, caracterizada por picos agudos na densidade de estados próximos à energia de Fermi, exceto no caso de dopantes de cobre. A energia de adsorção de intermediários varia entre diferentes substratos, levando a perfis de reação únicos. Superfícies com o dopante cobre tendem a desorver espontaneamente CO, ao contrário de outros elementos do quarto período (Fe, Co, Ni) que se ligam ao CO e podem favorecer aHER. Em contraste, os dopantes Ir e Rh são eletrocatalisadores promissores, favorecendoa redução de CO e suprimindo a HER em potenciais mais baixos que o cobre. Além disso,eles não são propensos a envenenamento por radicais OH. 

**

**

In almost every daily activity, the internet is always present. Many of these activities require the sharing of personal data (e.g. address, telephone, age, location, among others) and data considered sensitive (e.g. medical diagnosis, bank account, among others), generating a privacy concern. of this information to society. Sharing information requires the use of data protection tools, so the information in this data cannot be used to identify an individual. In order to comply with applicable laws and regulations, as well as informed data protection policies, a data protection mechanism called data anonymization has emerged. This mechanism basically consists of removing identifiers and hiding sensitive data, making it impossible to re-identify. As large amounts of data from individuals are disseminated, new challenges arise for the protection of their privacy. Various anonymization algorithms have been proposed, making publishing privacy-preserving data a very broad area of research. However, it is difficult to identify and select the most suitable algorithm, due to the large number of available algorithms and limited information about their performance. In this work, three anonymization algorithms called DataFly, Incognito and Mondrian are presented. Its efficiencies in data processing, its effectiveness in the amount of data used will be addressed, employing a wide set of different parameters, metrics and data sets. To guide the selection of an algorithm, a battery of experiments will be carried out among them to identify which factors can influence performance, presenting the conditions in which each algorithm outperforms each other in certain privacy requirements as well as their advantages and disadvantages

**

**

**

**

Computer diagnostic aid systems are used in the various health sectors to facilitate the work of the most used health professionals. These computer systems help healthcare professionals to detect the chance of error, correct the accuracy of the diagnosis, in addition to reducing the time and cost of making the diagnosis. More specifically in the measures traced from cephalometric treatment in linear and not crucial to determine an orthodontic treatment. For the elaboration of this exam, measurements are made in a lateral medicine of the head called teleradiography. The diagnosis of a tracing tracing was obtained through the evaluation of measurements performed on teleradiography in comparison with the defined ideal pattern. This can be a comprehensive and well-defined feature as a candidate for the application of computational assessment techniques, without the possibility of automatic diagnosis, an intervention manual of the specialist. Research is evaluated as an existing computational research technique and currently applied. For the application of computer vision techniques, we intend to extract a basis from one of cephalometric tracing training tests performed in the most computerized tracing data collection tool of Cfaznet, one of the tracing data software in Latin America. . This training base will be used for nutritional calculation as a computer vision technique that can best perform the task of anatomical points in a teleradiography. In order to find the techniques it is best to carry out the tracing exam, most likely the job of searching to find the best techniques comparatively. In this way, computer vision techniques were studied today capable of performing a projected cephalometric determination exam without a theoretical intervention.

**

**

**

Air transportation remains the best option to connect distant regions and support world social and economic development, despite environmental challenges. Brazil depends strongly in this mode of transportation due to its continental dimensions. In this work, we conducted a study using the multiplex approach of complex networks to analyze the air transport network of Brazilian passengers, BATMN. In this context, we create the multiplex efficiency index (MEI) and its versions for targeted and/or weighted networks, which allow to quantify the level of connection heterogeneity of a multiplex network on a scale of 0 to 1, considering the characteristics of direction and importance level of edges, being appropriate for research in transportation networks. We used the MEI to evaluate the diversification of routes of airlines in Brazil between 2010 and 2021, a period of occurrence of major events in the country. We analyzed the influence of some of these events on network efficiency, such as the FIFA World Cup in 2014 and the Rio Olympic Games in 2016, as well as other sector changes, including the merger between airlines, airport privatization, the closure of operations of an important air carrier and COVID-19 pandemic. In addition, we investigated whether the diversification of routes affected the prices of air tickets in Brazil. Additionally, through a recently introduced methodology in the literature, we examined the network’s diffusion capacity and observed the relative gain that occurred in the diffusion of the air transportation system when the flight connections between different airlines are considered, outlining a scenario possible partnerships between airlines. The evaluations carried out in this work provide new and significant information about the passenger air transportation network in Brazil, contributing to optimize the offer of this type of transportation in the country and, consequently, reduce the costs involved. In addition, the proposed methodology allows not only a differentiated analysis of air transportation networks, but also the quantification of the efficiency of multiplex networks in general.

**

According to data from the United Nations (UN), there are currently more than 3.8 billion people living in cities and in 2050, almost 70% of the world’s population will live in urbana reas. Cities are considered places of economic, political, cultural and social development, but disorderly urbanization, especially in developing countries, has deteriorated the quality of various systems and services offered to citizens. Among the systems that make up the urban infrastructure, the transportation system can be seen as an indicator of the residents’ quality of life. This is because, through the analysis of the transport system, from wich people gain access to medical, educational, cultural and leisure services, it is possible to identify social, economical and cultural differences. Thus, in order to ensure the construction of sustainable cities that offer a good quality of life to their inhabitants, the planning and efficient management of urban mobility is essential. In order to contribute to the construction of an efficient transport system, this work presents the development of a framework composed of low cost hardware and software, which can be used to generate and monitor real time indicators for the transport system. The proposedframeworkuses data from different sources such asprobe request, census and smart card and employs machine learning algorithms such as Kalman Filter and Artificial Neural Networks to generate the origin-destination matrix and other indicators of the transportation system. Preliminary results indicate that the strategy used manages to reduce errors in forecasting indicators to less than 5%.

**

The perturbative calculations in Quantum Field Theory can be represented, graphically, in terms of Feynman diagrams. The calculations of amplitudes represented by Feynman diagrams involve integrals, often divergent beyond tree level. These integrals require a treatment known as regularization, in order to extract the physical content of the amplitude. One of the most powerful techniques, Dimensional Regularization, involve dimensional extensions of space-time. However, there are models that involve mathematical objects whose the dimensional extension is ambiguous or ill-defined. A technique, called Implicit Regularization, proved to be appropriate for the treatment of models whose dimensional extension is not well defined. The method allows the calculations to be performed in the own theory dimension and prescribes simple rules for the preservation of the symmetries of the models. However, for a single Feynman integral, the expansion to separate divergences can generate a large set of finite integrals in the case of high degrees of divergence. In addition, expansions generate high powers of momenta in the numerator and denominator, which results in long and labor-intensive calculations. In this thesis, we developed a new procedure for the application of Constrained Implicit Regularization that simplifies the calculation of amplitudes, including finite parts.

The main objective of this work is to present an overview of Brazilian technical production based on the analysis of patents registered at the National Institute of Industrial Property (INPI), as well as international repositories such as Espacenet. Initially, using the curricular base of the Lattes Platform of the National Council for Scientific and Technological Development (CNPq), this study seeks to carry out an analysis of patents registered in the curricula of Brazilian researchers and their validation with the Espacenet repository, in order to verify the validity of the data, as well as obtaining a set of additional information about the recovered patents. For an analysis that extrapolates the data extracted from the résumés, complementary information will be extracted from the patents identified in the résumés through the collection of information in the INPI repository and also in the Espacenet repository, the latter being an international repository of patent registrations of recognized relevance . In order to identify innovation networks and the formation of invisible schools resulting from collaboration between researchers throughout the national territory, this analysis will be carried out using data extraction tools, text mining, graph construction, as well as algorithms of natural language processing. Subsequently, patterns will be explored and data quantification will be explored, as well as an evaluation of the formed collaboration networks, with an emphasis on individuals and institutions.

Aedes aegypti and Aedes albopictus are the main vectors of arbovirus infections such as dengue (DENV), zika (ZIKV), chikungunya fever (CHIKV) and yellow fever (YFV) in the world, being responsible by the infection of millions of people annually. The range of the highest global incidence of these vectors is concentrated in tropical and subtropical regions due to the climate of these regions favoring the development of mosquitoes. Rainfall has a considerable influence on the development of Aedes, with increasing rainfall, the number of breeding sites available for egg laying increases, causing a positive correlation with the incidence of arboviruses. Mathematical models related to meteorological variables have been widely developed in order to assist in the control of the vector. The oviposition of Aedes spp. it can occur in wet and dry environments and this factor is able to modulate the population influencing its seasonal pattern by the female and the seasonal pattern of the vector population. In this work a theoretical-numerical study of the model will be carried out in order to reproduce the dependence of the entomological parameters of the mosquito’s life cycle with the accumulated weekly rainfall of certain Brazilian regions. An entomological compartmental mathematical model composed of three nonlinear differential equations representing the dynamics of egg populations in dry and wet environments, and the female population under the effect of accumulated weekly rainfall variables, was proposed and analyzed. The model was validated by comparison with experimental catch data obtained by entomological monitoring in the city of Caratinga, Lavras and Sete Lagoas (Minas Gerais, Brazil). It was found that the model describes the mosquito’s life cycle and is able to assist in the monitoring and control of Aedes.

Private transport has become a viable and increasingly popular option as an alternative to urban transport. With this growth, an old and recurrent problem emerges, that is the relationship between passenger demand and taxi supply. While taxi drivers spend considerable time looking for new passengers, on the other hand, some passengers cannot find taxi drivers. With this imbalance, taxi drivers pay more on fuel and depreciate the vehicle due to the excessive demand for passengers. Passengers end up not picking up taxis, causing disbelief in the availability of the service provided. This problem suggests the creation of techniques that make it possible to reduce the gap between the demand of passengers for taxis and the effective contingent of vehicles needed to meet this demand. In this context, this work presents a new approach to taxi demand prediction. The proposed approach uses historical data from taxi rides and weather data. The Kruskal-Wallis method identifies the most relevant variables, and an evolving fuzzy system is used for demand prediction. In this work, to perform the prediction, three evolving systems are evaluated: Autonomous Learning Multi-Model (ALMMo); Multivariable Gaussian Evolving Fuzzy Modeling System (eMG), and; evolving Neo-Fuzzy Neuron (eNFN). Computational experiments were conducted to compare the performance of the proposed approach. The results showed that the proposed approach achieved better or comparable results related to other published works, suggesting, therefore, promising applicability of this alternative as a solution to the problem of passenger supply and demand.

Scientific production has shown significant growth in recent times. Currently, there are numerous ways to propagate scientific knowledge. Therefore, scientific events emerge as one of the main means of disseminating research results in several areas of knowledge and aim to make information and scientific knowledge available to a group of people. And yet, publish the results of research in the format of annals of the events. This is a compilation of all the articles presented at a specific event. However, aiming at the integration of proceedings of different events, this task is not trivial, considering that each event has its publications in specific repositories and in different formats. In this context, integrating and analyzing the articles generated from the events becomes a complex task. The work in question aims to trace characteristics of scientific events and their published articles, using data extracted from the curricula registered on the Lattes Platform. After the collection of curricular data from the Lattes Platform in January 2021, steps of selection, treatment, curation of the data were carried out; featuring a set of 360,888 curricula of individuals with completed doctorates and 11,416,655 articles published in annals of events. In order to make the presented results possible, several analyzes were carried out in order to portray and detail the characteristics of scientific events, published articles and authors. The results presented will serve as a basis for several other studies that aim to understand the development of Brazilian science based on scientific events

**

**

**

 The field of study of complex systems is intertwined with many areas of science, being a promising approach to the study of biological problems. In this work, we applied some complex systems tools to study three biological problems of great current concern: the COVID-19 pandemic, the cyclicality of dengue outbreaks, and the impact of climate change on an invasive species. In the COVID-19 study, we developed an agent-based model to simulate the spread of the disease and the effect of superspreaders under different mobility scenarios. A network analysis was applied to investigate the role of superspreaders. The results showed that some individuals are crucial to the infection process. Furthermore, we performed simulations for some Brazilian cities using networks, mobility data extracted by cell phones, and different infection scenarios. In the case of the dengue study, we initially applied the visibility graph technique and entropy analysis to interpret two dengue time series data. Also, a tool called impact-frequency histogram was applied. This tool is generated from the visibility graph and showed better results for dengue cycling than the approach using entropy. Then, an agent-based model was adapted to investigate the dynamics of different scenarios of dengue. The result made it possible to identify humans and mosquitoes that play an essential role in the networks. Finally, we adapted an agent-based model to simulate the seed bank dynamics of an invasive species under different climate change scenarios. The results of the three models can help in decision-making in different areas. The COVID-19 and dengue models can be adapted for other scenarios and diseases. The invasive species model can also be adapted to assist in managing these species. Thus, the complex systems approach allowed us to achieve unprecedented results in treating and analyzing biologically motivated problems, such as identifying superspreaders in the spread of COVID-19, cyclicality in dengue epidemic outbreaks, and the influence of climate change on the seed bank of invasive species. All studies discussed in this work are likely to be continued, whether for the same problems or similar applications.

**

**

**

The theme of women in science has been the focus of several studies that seek to understand women’s role and their varied relations with science, bringing approaches that analyze scientific and technological participation and seek to understand aspects of their academic trajectory. In this context, this study aimed to analyze the participation of women in Brazilian science using as a data source the set of PhDs who have curricula registered in the Lattes Platform and whose gender registered in their curriculum is female. After the collection of curricular data from the Lattes Platform, the selection and treatment of the data was carried out to characterize the set to be analyzed, obtaining a total of 125,515 curriculum with informed female gender and completed doctorate degree. The data of the PhDs female were analyzed under several aspects such as orientations, bibliographical and technical productions, areas of expertise and academic training. In sequence, the PhDs female who have research productivity grants at CNPq were also analyzed, making it possible to verify how the scientific collaboration of this group occurs, as well as, in the group of all PhDs female. Therefore, it is highlighted that studying the various aspects of women in general and particularly in science, in addition to being relevant, makes it possible to understand how their research has evolved, and may contribute to a better understanding of the participation of women in the scope of Brazilian science, as well as as for formulating policies and strategies that enhance the performance of women in academic and scientific environments.

**

Angiogenesis is the process of growing new blood vessels from existing vessels. Under normal conditions, the new vessels are robust and organized and in non-normal conditions, as with the appearance of a tumor, the vessels formed are stunted and entangled, as they are stimulated to grow quickly to feed the tumor that lacks oxygen and nutrients. The attempt to inhibit angiogenesis can cause several side effects such as hypertension, thrombosis, fatigue, among others. Therefore, studies that focus on the characterization of the signaling pathways that work in this process contribute to research directed to the use of new drugs for various diseases, including cancer. The main objective of this study is to present a new mathematical model for angiogenesis and apply it to Alves (2018) experimental research. The model is described by ordinary differential equations translated into equations of differences in the MatLab software and for validation, followed by the steps outlined by Boareto et al. (2015a) and Boareto et al. (2015b). Moreover, the model was applied to Alves (2018) experimental research, where it was noticed that the balance of the system is disturbed when external VEGF (Vascular Endothelial Growth Factor) is altered, causing great changes in the angiogenesis process.

This study presents an algorithm to solve the problem of optimizing the supply of the power demand in the medium term operation of a hydroelectric plant. Instead of using generic nonlinear algorithms, this work explores the particularities of the mentioned problem in order to produce efficient solutions. In the first phase, an analytical formula that maximizes the turbined for a given target profile, is demonstrated. Then, this formulation in applied to the following phase of the algorithm, which considers the nonlinear model hat converts the discharged flow into power using the efficiency, deriving a nonlinear problem to maximize the power following a given profile. The method second was successfully applied with real data of existing hydropower plants.

**

**

**

**

.The classical economic theory argues that markets are efficient, that prices move towards a long-term equilibrium and that there is no possibility of arbitrage. To investigate this Efficient Market Hypothesis (EMH), we defined a model for forecasting trends in financial asset returns, based on a discretized coupled linear differential equation system. In this model, we consider that the relationship between stock price deviations from a price considered fair provides information on the dynamics of stocks prices. Such deviations are calculated using the logarithmic difference between actual and fair prices, modeled by an exponential moving average. To compare an emerging and a developed market, we applied the model to the Brazilian and North American stock exchanges. The consistency of the model was investigated by comparing results in different periods of four months. The relationship between model accuracy and Hurst exponent was studied. The execution parameters of the model, such as the fitting window, number of assets, and frequency of data sampling, are explored in order to find a combination that presents relevant accuracy rates. Hypothesis tests were applied to the forecast results to investigate whether the model has a success rate different than a completely random process. Given the parameters explored, we chose a set that presented satisfactory results and applied the model in a realistic simulation of the market using an algotrading platform. In the simulation, we defined two operating models for the trading strategies: MO1 and MO2. In MO1, the operations are opened according to the signal sent by the predictor and closed at the closing of the same candle. For MO2, operations are closed using stops. The results were compared taking into account performance indicators, such as: risk of exposure to the market, financial return, operating rates and allocated capital.

**

Percolation and Ising models have applications in several areas of science and are among the most studied models in Statistical Mechanics. They exhibit a phase transition and determining the critical value at which it occurs is one of the main issues related to these models. The aim of this work is to study three models, two percolation and one Ising, in which the critical value varies as a function of some parameter. The first model presented is the Multirange Percolation Model (MPM), in which we consider the classic model of percolation in the hypercubic lattice L^d and add bonds of sizes _m1,m2,...,mn, parallel to each coordinate axis, so that the sizes of the larger bonds are multiples of the smaller ones, and _{mi >} 1 for all _i. There is an analytical result showing that the critical point of this model converges to that of the classical model in the hypercubic lattice L^d(n+1) when _{mi → ∞} for all _i. We find numerical evidence that, if _d = 2, this convergence is monotonous and follows a power law when _n = 1 and _n = 2. In order to verify whether the convergence related to the critical point in the MPM also occurs in other models that exhibit phase transition, we studied the Multirange Ising Model, which is defined similarly to the MPM. We verified that, if we analyze the critical temperature instead of the critical point, convergence will also occur for the same situations studied in the MPM. Finally, we also study the Constrained-Degree Percolation Model (CDPM). In it, we try to open the bonds of the graph in a random order that is previously defined. Each bond is opened if it satisfies a certain restriction that depends on a _k parameter. This model has applications in the study of dimers and polymers and we simulate it in the hypercubic lattice L^d for d _{∈ {}2_,3_,4_}. Through a numerical analysis, we show that the critical value of the model decreases as function of the constraint _k. Furthermore, we found evidence that the CDPM is in the same universality class as the classical Ising model.

Combining complex mathematical models that optimize investment portfolios with efficient and practical computational techniques is possible with the advent of technology. The aim is to build a computational framework capable of dealing with the actual peculiarities of the financial world while using available theoretical knowledge. Therefore, this work proposes a computational framework composed of different models that use Model Predictive Control (MPC) to optimize investment portfolios. An innovative strategy that combines MPC and multi-objective optimization and realistic financial market constraints, such as investment limits, automatic finance, cardinality, and transaction cost, is proposed. Performance criteria (objective functions) are the expected values of wealth, variance, and CVaR. We proposed multiperiod formulations for the objective functions and the genetic algorithm that performs the optimization since the multiperiod perspective is essential for the MPC strategy. With the experiments performed, it was possible to obtain various findings involving the prediction horizon, cardinality, risk, and return of the portfolio. Among those that can be highlighted, the prediction horizon benefits the portfolio optimization problem. It offers wealth value not obtained by the myopic strategy; portfolios with smaller cardinalities have greater risk because they have capital allocation in the risk-free asset. The five cardinality portfolios have the highest wealth values. In the out-of-sample analysis, the accumulated wealth of the proposed strategy surpassed Ibovespa and prestigious investment funds in 2020.

Nanoparticles are small particle agglomerates (clusters) of few nanometers ranging from a dozen up to a million atoms. The cluster behavior is strongly linked to its geometry, number of particles and composition. The cluster configuration cannot be determined by experimental data solely and further hypothesis are necessary for giving meaning to the experimental results. The improvement of the current computer programs makes possible the advance of ab initio approaches which are able to determine the optimal geometries without the need for any experimental data, but only for small clusters and at expenses of high computational cost. Quantum effects involved in clusters contribute to the unpredictability of the output geometries, specially when high accuracy is required. The higher the expected accuracy is, the more of those quantum aspects should be taken under consideration, increasing the complexity of the required algorithms. Search for solutions methods, especially using Genetic Algorithms (GA), have been used in a variety of cluster studies bringing light to good solutions in a highly parallel fashion accessing marginally the geometry domain. GA methods benefit from elegant metaphors and have been constantlty improved throughout the years. In the present work, a mechanistic method for random generation of clusters is studied to improve the GA learning in cluster geometry optimization methods.

**

**

Breast cancer is the most commonly occurring one in the female population. Early diagnosisof this disease, through mammography screening, can increase the chances of cure to 95%.Studies show that Brazil has a relatively satisfactory number of mammography units, but thisequipment is poorly geographically distributed. This research focuses on the MammographyUnit Location Problem (MULP), which aims an efficient distribution of mammography units,in order to increase the covered demand. Focusing on the State of Minas Gerais, Brazil,an analysis is made considering that, int the real world, there is a difficulty in relocatingequipment already installed. Therefore, it would be interesting to optimize the location ofnew equipment purchases. Since MULP is NP-hard, an algorithm based on the SimulatedAnnealing meta-heuristic is also developed to handle large instances of the problem.

The mineral extraction sector is an activity of great source of income, contributing financially and economically, and in Brazil, this contribution is very relevant due to the global scales of mining assets existing on the ground. Large mining companies in Brazil have multiple productive and logistical assets, which need to have integrated planning to optimize theuse of these assets in order to meet customer demands and reduce operating costs. In this context, the problem of integrated short-term programming in mines is defined, in whicht actical-level decisions are defined jointly in integrated operations centers (IOC). This problemis solved using goal programming and four different strategies from the Relax−and−Fixheuristic. In addition, a testing and evaluation methodology, so much for the verification ofthe algorithms and for the validation of the mathematical model, is proposed and the results obtained confirm these evaluations.

Recommender systems suggest items that best fit the users’ interests and tastes. They are complex systems, which are often designed as black-box models. However, it is well- known that understanding recommendations is essential to help building users’ trust and engagement in these systems, so explanations for recommendations are a way to achieve this goal. An explanation is a piece of information displayed to users, explaining why a particular item is recommended. The case-based explanation is one of the most commonly used explanation styles in recommender systems. This approach is focused on providing explanations in the form of previously liked items that are used to make the recommendation, such as: “Because you liked A, we recommend B”. To measure the quality of methods that generate explanations, we can conduct online and/or offline evaluations. Online evaluations require direct interaction with users, and offline evaluations require just previously historic data. Online evaluation can handle more aspects of the explanations and produces more complete results, so it is encouraged, but tests involving users are not always accessible. In many scenarios, offline evaluation is preferred because it will not affect user’s experience through bad explanations, and it is also easier to implement than the online counterpart. Few research projects approach offline metrics to measure the quality of case-based ex- planation methods, so it is an ill-defined problem. Thus to cover this gap in the literature, we propose a new offline metric – ECT (Ease-of-interpretation Coverage Triviality) – to evaluate more complex aspects of case-based explanation methods. It is based on three sub metrics, one of them has already been proposed: (Coverage), and two new metrics: (Ease-of-interpretation and Triviality). Experiments were conducted in offline and online scenarios to validate the discriminative power of the proposed metric ECT, and show that it presents a 0.67 correlation with what users consider good explanations to their recommendations.

**

**

The topological analysis of networks is an important field of study in Network Theory, with applications in various fields of science. In this study, we modified the topology of a square network, through the reconnections of its edges and obtained different types of networks: conservative random, nonconservative random and scale-free. Under certain conditions, the random and free-scale networks showed small world properties. We applied First Passage analysis tools to investigate the properties and characteristics of random walks in these networks. In the topologies investigated, we analyzed the First Passage Time (FPT) of a significant number of non-interacting random walkers, varying the departure and arrival sites. To characterize these processes, we applied the concept of first passage simultaneity, through the so-called Uniformity Index (UI), which is a measure of the probability that two independent walkers will arrive at the target site together. The UI allows to avaliate whether the average first passage time (TMPP) is a good measure for the process, and allows to identify networks with small world characteristics. The analysis of the sites occupation during a random walk allowed us to differentiate the different types of networks, in particular identify the small world properties, a topic that is still controversial in the literature.

**

.

**

**

**

**

**

..

**

.

.

.

**

.

 **

.In several segments that require a prior budget and / or a revenue forecast, estimates are common. This procedure is based on statistical models. The Binomial distribution is a discrete statistical model that determines the chances of success or failure of the occurrence of a certain number of events in a limited sample space. The normal distribution is the standard statistical model that follows the law of large numbers, that is, random occurrences that tend to infinity converge to an average and present a variance. Gathering the characteristics of the two distributions in a single tool is of great value to make estimates and, for that, the convolution is used. In order to perform the convolution between the two distributions, it is necessary that both are discrete or continuous. The Binomial distribution was approximated to a continuous function and subsequently convolved with the normal distribution, while the normal distribution was approximated from a discrete series and was subsequently convolved with the Binomial distribution. It is made the forecast of the quantity of defective pieces that must be collected by Magneti Marelli throughout Brazil through the proportionality relation of the discrete convolution. The estimated values include the actual values disclosed by the company. At the UFMG Veterinary Hospital, estimates were made of the number of animals that were able to perform castration surgery in 2016, through discrete convolution and using data from the year 2015. All monthly values collected at the veterinary hospital in the year of 2016 are within the forecast made using the model. The National Broad Consumer Price Index (IPCA) for the years 2011 to 2016 was estimated based on data from 1996 to 2010 released by IBGE, using continuous convolution. The observed values are within the range predicted by the method with proximity to the mean. Due to the application in the several sectors mentioned with satisfactory results, the proposed method proved effective and multidisciplinary, that is, it can be applied in several areas.

**

*

***

**

***

**

***

.

.

.

.

.

The computational neuroscience attempts to understand the nervous system functions through the use of realistic models and large scale simulations. This work is based on the Theory of neural group selection (TNGS) that states that the neural connection topology is not established genetically, but through neural competitive activities in the embryonic phase. These neural competitive activities generate groups of neurons strongly connected, called neuronal groups. These groups, according to the TNGS, are the smallest functional unit of the nervous system, responsible for its most basic processing. Therefore, this work has the objective of designing a self-organized spiking neuron network (a neuronal group) that works as frequency generator. Among the several spiking neural models, the Izhikevich model can be highlighted because it can associate low computational cost and good biologic plausibility. Keeping that in mind, this model was chosen to represent the neurons in this work. The dynamics of the neuronal groups was first studied using a 2k Factorial Design with the objective of finding a relationship between the neuronal group parameters and its oscillation frequency. The neuronal group frequency was more sensitive to the parameters a and b of the Izhikevich equation and the synaptic weight s. The statistic relevance of these parameters was measured using the Friedman test and all of them presented statistical relevance. The parameters were then incremented to form groups with oscillation frequency of 5, 20 and 40 Hz. In this experiment, the behavior foreseen by the Factorial Design could be confirmed and the oscillation frequency could be controlled by the a, b and s parameters. A formulation able to directly relate the value of the parameters and the oscillation frequency of neuronal group was presented contributing to the construction of frequency generators with biologically plausible neural models. It is expected that the results can contribute to better understand the self-organized spiking neurons networks and the nervous system.

**

Nos últimos anos, a Dengue tem sido tema relevante nas pesquisas cientí ficas, na busca de novas técnicas de monitoramento e controle do Aedes aegypti, no desenvolvimento de vacinas e outros tratamentos da doença. Como ainda não temos no mercado uma vacina contra a Dengue, a política de prevenção da doença no Brasil se dá por meio de ações educativas monitoramento e controle do vetor transmissor. Estas técnicas se mostram trabalhosas, poucos e e ficazes e apresentam inviabilidade econômica na atual gestão pública. Uma vez que os recursos dos órgãos públicos não é ilimitado e o controle do vetor da dengue tem impacto social e ambiental, a otimização se torna fundamental. Visando minimizar o custo com controle do vetor da dengue, essa dissertação propõe modelos de otimização mono e multiobjetivo baseados em trabalhos anteriores, a serem resolvidos por algoritmos genéticos. No modelo, consideramos parâmetros diferentes para cada uma das estações do ano, a fi m de encontrar a quantidade e o tempo de controle ideal para cada fase do vetor: imatura e adulta. Resultados computacionais mostram que o controle desta forma é mais e ciente quando feito de maneira integrada em mais de uma estação do ano, não apenas no verão. E a versão multiobjetivo complementa o estudo dando mais liberdade ao decisor, quando fornece como opções um conjunto de pontos ótimos.

Hoje, o câncer é uma preocupação global, responsável por 12% das mortes no mundo. O tumor de mama é o segundo mais comum mundialmente, e predominante entre as mulheres. Essa neoplasia é responsável por mais de 50 mil novos casos anuais no Brasil. A primeira fase do tratamento consiste na quimioterapia pré-operatória, cujo objetivo é a redução ou a completa extinção do tumor. Entretanto, uma pequena parcela dos pacientes não possui sensibilidade a esse tipo de tratamento, sofrendo os efeitos colaterais da quimioterapia sem o resultado esperado. O objetivo principal deste trabalho é a criação de um preditor de sensibilidade quimioterápica baseado em dados de expressão gênica estruturado em duas etapas. Estão sendo utilizadas bases de dados extraídas do MD Anderson Câncer Center, cada uma com 22.283 valores de expressão gênica e rotulado como sensível ou não à quimioterapia pré-operatória. A primeira parte do trabalho consiste em selecionar as características mais relevantes para o problema de sensibilidade quimioterápica. O método Volcanoplot se mostrou mais eficiente para classificação, por isso, o utilizamos com diferentes configurações. Em seguida foram utilizados métodos supervisionados e não-supervisionadados para a construção do classificador em duas etapas. Os resultados da nova metodologia de classificação de pacientes para sensibilidade quimioterápica obtiveram uma melhora em comparação com classificadores da literatura e houve uma redução no número de sondas. Um dos experimentos executados, com o algoritmo de aprendizagem Extreme Machine Learning atingiu uma acurácia de 99% na classificação para um dos grupos, que possui uma pequena parcela de pacientes com o subtipo molecular basal.

The BIMASCO architecture project is an autonomous multi-agent system, inspired in biological systems, with the objective of to solve combinatorial and non-combinatorial optimization problems. Each agent contains an encapsulated heuristic or metaheuristic and interacts with others agents without requiring of languages or communications protocols. The previous version this architecture was designed with metaheuristics linked in the only benchmarking problem implemented , the Vehicle Routing Problems with Time Windows (VRPTW). In this project was implemented the generalization of the metaheuristics, unlinking them of the problems by standardizing of the descriptors data and their representations forms. Were generalized and implemented: Greedy and Random metaheuristics (constructive); Best Improvement Method, First Improvement Method and Random metaheuristics (improvement); Genetic Algorithm metaheuristics; GRASP; Iterated LocalSearch; Simulated Annealing and Variable Neighborhood Search. Were implemented binary, integer e real representations for some classical problems of combinatorial optimization (e.g., Maximum Diversity Problem; Number Partitioning Problem) and non-combinatorial optimization (e.g., Beale Function; Rastrigin Function). These heuristics and metaheuristics
were validated by computational experiments designed to select the appropriate representation of the problems. The results of experiments demonstrate a flexibility of the BIMASCO architecture for the implementation of the new heuristics/metaheuristics and implementation the new problems with minimal modification of architecture. The results demonstrate too that the architecture works as autonomous multi-agent system, interacting of indirect form, not requiring communications protocols, not establishing causality relation and working rightly. The experiments were not designed with intention of optimize the parameters of work of the heuristics/metaheuristics for generate results similar or better that the described in the technical and scientific literature.

 **

.

**

**

The present work is focused on demonstrating the functioning of the computational method known as heuristics, as used on the search of a solution for structural analysis. The problem of structural optimization was approached by optimizing plane trussed structures to obtain their minimal cross sectional area, restricted to the limitations imposed by the project. Computational Heuristics was used with the goal of finding alternative shapes to minimize the weight - and therefore the costs – of such structures. The advantages and drawbacks of each method are demonstrated, including the analysis of the results obtained in three different structural problems, as compared to the results of the methods applied on the different tests. This work and its results are aimed to contribute to the use of auxiliary methods in the field of Engineering as a mean to improve the structural projects, and to simultaneously compile a knowledge base on heuristics and its applications.

The generation of trustworthy and high quality computational systems, requires, basically, the adoption of an adequate development process. Within this subject, a question that it has deserved the attention of the researchers is the study of the process of development of scientific software of academic nature.One of the motivations for such concern is the existence of different types of processes for the construction of software, however, dedicated software conventional. Other kind of software, as those created by scientists to be used in their research, might have particularitities and possibly different concerns from the conventional and commercial systems. Hence, the existing processes for the development of scientific software might become inadequate. This work has the objective of defining a development process for scientific software of academic nature, here called PESC (Process of Specific Development for Scientific Software), with the intention to confer quality to these software. Such process was conceived on the basis of the directions indicated in the earlier phase of the work and with the support of the literature. It is expected, with this research, to define the PESC initial process.

In recent years, there has been an increasing interest in the development of applications of autonomous robots, capable of adapting themselves to an unpredicted and dynamic environment. However, the difficulties found in developing such applications using Classical Robotics ideas, are driving research efforts towards a new area: the Situated Robotics, which focuses on the embodied mind paradigm. So, it would be of great value, mainly for areas like Cognitive Science, Robotics and Artificial Intelligence, the development of an architecture for the creation of autonomous robots, even more if such architecture is flexible, extensible, and easy to use and test. This work presents an extension of the “Artífice” architecture for the development of emotional and cognitive autonomous robots. This architecture was first conceived for the creation of bio-inspired intelligent software agents. Changes were made in such a way that the core of the architecture (its nervous system) was encapsulated, while its peripheral system (everything else but the nervous system) became more flexible, in order to provide the framework to be usable for the creation of both software agents and hardware ones (i.e., robots).To ensure that the proposed architecture works adequately, computational experiments were done in a simulated world. Thus, a Pioneer simulated robot was living and navigating in a 3D simulated world, surrounded by a wall, in which some pillars were placed. The robot was expected to regulate its bodily needs: sleep (regulated whenever the robot momentarily stops), as well as its basic emotions of curiosity (regulation occurs as the robot gets closer to a pillar) and stress (regulation occurs as the robot experiences a pillar or interacts with it, staying near it for a while, seeing it, since different colors of the pillars contribute differently to the emotional regulation or deregulation of its stress). It was observed that while interacting with its environment, the robot tends to try pillars of those colors that lower its stress more rapidly and tends to avoid those colored pillars that increase the stress level. This same kind of behavior could be observed in all of the simulations carried out, showing that emotional-cognitive robots built with the proposed architecture can present a behavior that could be described as intelligent, from an observer point of view.