Navegando por Assunto "Machine learning"

Agora exibindo 1 - 20 de 24

Acesso aberto (Open Access)
5G MIMO and LIDAR data for machine learning: mmWave beam-selection using deep learning
(Universidade Federal do Pará, 2019-08-29) DIAS, Marcus Vinicius de Oliveira; KLAUTAU JÚNIOR, Aldebaro Barreto da Rocha; http://lattes.cnpq.br/1596629769697284
Modern communication systems can exploit the increasing number of sensor data currently used in advanced equipment and reduce the overhead associated with link configuration. Also, the increasing complexity of networks suggests that machine learning (ML), such as deep neural networks, can effectively improve 5G technologies. The lack of large datasets make harder to investigate the application of deep learning in wireless communication. This work presents a simulation methodology (RayMobTime) that combines a vehicle traffic simulation (SUMO) with a ray-tracing simulator (Remcom’s Wireless InSite), to generate channels that represents realistic 5G scenarios, as well as the creation of LIDAR sensor data (via Blensor). The created dataset is utilized to investigate beam-selection techniques on vehicle-to-infrastructure using millimeter waves on different architectures, such as distributed architecture (usage of the information of only a selected vehicle, and processing of data on the vehicle) and centralized architectures (usage of all present information provided by the sensors in a given moment, processing at the base station). The results indicate that deep convolutional neural networks can be utilized to select beams under a top-M classification framework. It also shows that a distributed LIDAR-based architecture provides robust performance irrespective of car penetration rate, outperforming other architectures, as well as can be used to detect line-of-sight (LOS) with reasonable accuracy.
Acesso aberto (Open Access)
Abordagem Inteligente com Combinação de Características Estruturais para Detecção de Novas Famílias de Ransomware
(Universidade Federal do Pará, 2024-03-22) MOREIRA, Caio Carvalho; SALES JUNIOR, Claudomiro de Souza de; País de Nacionalidade Brasi
Ransomware is a malicious software that aims to encrypt user files and demand a ransom to unlock them. It is a cyber threat that can cause significant financial damage, as well as compromise privacy and data integrity. Although signature-based detection scanners commonly combat this threat, they fail to identify unknown ransomware families (variants). One method to detect new threats without the need to execute them is static analysis, which inspects the code and structure of the software, along with classification through intelligent approaches. The Detection of New Ransomware Families (DNFR) can be evaluated in a realistic and challenging scenario by categorizing and isolating families for training and testing. Hence, this thesis aims to develop an effective static analysis model for DNFR, which can be applied in Windows systems as an additional security layer to check executable files upon receipt or before execution. Early ransomware detection is essential to reduce the likelihood of a successful attack. The proposed approach comprehensively analyzes executable binaries, extracting and combining various structural features, and distinguishes them between ransomware or benign software employing a soft voting model comprising three machine learning techniques: Logistic Regression (LR), Random Forest (RF), and eXtreme Gradient Boosting (XGB). Results for DNFR demonstrated an average accuracy of 97.53%, precision of 96.36%, recall of 97.52%, and F-measure of 96.41%. Additionally, scanning and predicting individual samples took an average of 0.37 seconds. This performance indicates success in quickly identifying unknown ransomware variants and adapting the model to the constantly evolving landscape, suggesting its applicability in antivirus protection systems, even on resource-limited devices. Therefore, the method offers significant advantages and can assist developers of ransomware detection systems in creating more resilient, reliable, and fast-response solutions.
Acesso aberto (Open Access)
Aplicação de redes neurais artificiais para predição de RSSI e SNR em ambiente de bosque amazônico
(Universidade Federal do Pará, 2024-06-11) BARBOSA, Brenda Silvana de Souza; ARAÚJO, Jasmine Priscyla Leite de; http://lattes.cnpq.br/4001747699670004; https://orcid.org/0000-0003-3514-0401; BARROS, Fabrício José Brito; http://lattes.cnpq.br/9758585938727609
The presence of green areas in urbanized cities is crucial to reduce the negative impacts of urbanization. However, these areas can influence the signal quality of IoT devices that use wireless communication, such as LoRa technology. Vegetation attenuates electromagnetic waves, interfering with data transmission between IoT devices, resulting in the need for signal propagation modeling that considers the effect of vegetation on its propagation. In this context, this research was conducted at the Federal University of Pará, using measurements in a wooded environment composed of the Pau-Mulato species, typical of the Amazon. Two propagation models based on machine learning, GRNN and MLPNN, were developed to consider the effect of Amazonian trees on propagation, analyzing different factors such as the height of the transmitter relative to the trunk, the beginning of the foliage, and the middle of the tree canopy, as well as the LoRa spreading factor (SF) 12 and the copolarization of the transmitter and receiver antennas. The best models were the machine learning ones, GRNN and MLPNN, which demonstrated greater accuracy, achieving root mean square error (RMSE) values of 3.86 dB and 3.8614 dB, and standard deviation (SD) of 3.8558 dB and 3.8564 dB, respectively. On the other hand, compared to classical models in the literature, the best-performing model was the Floating Intercept (FI) model, with RMSE and SD errors around 7.74 dB and 7.77 dB, respectively, while the FITU-R model had the highest RMSE and SD errors, around 26.40 dB and 9.65 dB, respectively, for all heights and polarizations. Furthermore, the importance of this study lies in its potential to boost wireless communications in wooded environments, as it was observed that even at short distances at heights of 12 m and 18 m, the SNR (Signal-to-Noise Ratio) had lower values due to the influence of the foliage, but it was still possible to send and receive data. Finally, it was shown that vertical polarization achieved the best results for the Amazon forest environment.
Acesso aberto (Open Access)
Aplicação e comparação de técnicas de classificação automática de documentos: um estudo de caso com o dataset do domínio jurídico “Victor”
(Universidade Federal do Pará, 2024-02-01) MARTINS, Victor Simões; SILVA, Cleison Daniel; http://lattes.cnpq.br/1445401605385329; https://orcid.org/0000-0001-8280-2928
The application of Natural Language Processing (NLP) and Artificial Intelligence (AI) in the Brazilian legal context is a rapidly growing area that can alter the way legal professionals work, given the volume of generated text. Among the possible applications of NLP and AI is the automatic classification of documents, which, among other things, can be employed in the automation of the digitization process of Judicial Proceedings that are still in physical form. Therefore, this work applies and compares AI algorithms for the classification of legal documents. The algorithms are divided into two different approaches. The first approach (I) separates the computational representation process of the text from the classifier training itself and applies SVM and Logistic Regression in conjunction with computational representations based on TF-IDF, Word2Vec, FastText, and BERT. The second approach (II) simultaneously performs the computational representation of documents and the training of the classifier, applying Deep Learning algorithms based on recurrent neural networks, specifically ULMFiT (Universal Language Model Fine-tuning), and HAN (Hierarchical Attention Networks). The studied dataset is named VICTOR, composed of documents from the Supreme Federal Court (STF) of Brazil. The research concludes that both approaches can be applied to the classification of legal documents from the employed dataset. Additionally, despite being less computationally expensive, the classification pipelines of Approach I, which use the computational representation of the document with TF-IDF, yield results equivalent to pipelines employing Deep Learning. Furthermore, embedding documents specialization with data from the dataset under study, improves the performance of pipelines that employ Word2Vec, FastText and ULMFiT, compared to pipelines that apply the generic representations of these, i.e., models pre-trained with data from the general context.
Acesso aberto (Open Access)
Arquitetura de modelos híbridos, machine learning e otimizadores para análise de consumo de energia elétrica e produtividade em pintura automotiva
(Universidade Federal do Pará, 2024-03-27) OLIVEIRA, Rafael Barbosa de; OLIVEIRA, Roberto Célio Limão de; http://lattes.cnpq.br/4497607460894318
Strategies for optimizing energy consumption in the painting stages are emerging as key factors in promoting more sustainable and competitive production in the automotive sector. This dissertation seeks to predict energy consumption and maximize productivity in automotive painting, using an approach that combines variable selection, hybrid models, hyperparameters of these models and meta-heuristic optimization in a 3-stage architecture. Automotive painting processes have variables in the form of time series that describe the history of energy consumption. In stage 1, the best machine learning model is chosen (Random Forest, Long-Short Term Memory, XGBoost and GRU-LSTM) to predict energy consumption time series at t+1. In step 2, the RF, XGBoost and Dense Artificial Neural Network (ANN) models are evaluated to select the best predictor of the number of vehicles produced (cycles). In step 3, the best metaheuristic between Genetic Algorithm (GA), Differential Evolution (DE) and Particle Swarm Optimization (PSO) is selected to optimize the energy consumption predicted by the best model from step 1, using the best model from step 2 as a fitness measure. The final architecture reduced the energy consumed by up to 16% and increased the cycle by 127%, using the GRU-LSTM models in step 1, Dense ANN in step 2 and DE in step 3. The results highlight the opportunity to use the proposed approach to optimize energy consumption and productivity in automotive painting.
Acesso aberto (Open Access)
Beam-selection otimizado por aprendizado de máquina : uma abordagem multimodal
(Universidade Federal do Pará, 2023-12-30) FERREIRA, Jamelly Freitas; GOMES, Diego de Azevedo; http://lattes.cnpq.br/5116561408505726; KLAUTAU JÚNIOR, Aldebaro Barreto da Rocha; http://lattes.cnpq.br/1596629769697284
This dissertation aims to investigate the use of machine learning models using multimodal data as input to optimize the Beam-Selection process in millimeter-wave based networks. The use of Deep Learning has intensified in different areas, and it is possible to obtaing performance equal or superior to human performance, so its use is also promising in wireless communication scenarios. This work used data from different sources, which proved to be convenient since it is possible to adjust the model according to the quality/availability of this data. After executing the experiments and obtaining the results, it was observed that it is possible to obtain significant performance in different metrics even with simpler data such as image and coordinate.
Acesso aberto (Open Access)
Classificação de eletroencefalogramas epiléticos em estado de repouso com aplicação de classificadores lineares e um atributo derivado da densidade espectral de potência
(Universidade Federal do Pará, 2019-12-04) FIEL, José de Santana; PEREIRA JÚNIOR, Antonio; http://lattes.cnpq.br/3239362677711162
Millions of Brazilians are affected with epilepsy and the access to early diagnosis is crucial for their adequate treatment. However, epilepsy diagnosis depends on the evaluation of longduration electroencephalographic (EEG) recordings performed by trained professionals, turning it in a time-consuming process which is not readily available for many patients. Thus, the present work proposes a methodology for automatic EEG classification of epileptic subjects which uses short-duration EEG recordings obtained with the patient at rest. The system is based on machine learning algorithms that use an attribute extracted from the power spectral density of EEG signals. This attribute is an estimate of functional connectivity between EEG channel pairs and is called debiased weighted phase-lag index. The classification algorithms were linear discriminant analysis (LDA) and support vector machines (SVM). EEG signs were acquired during the interictal state, i.e., between seizures and had no epileptiform activity. Recordings of 11 epileptic patients and 7 healthy subjects were used to evaluate the method’s performance. Both algorithms reached their maximum classification performances, 100 % accuracy and area under the receiver operating characteristic (AUROC) curve, when a feature vector with 190 attributes was used as input. The results show the efficacy of the proposed system, given its high classification performance.
Acesso aberto (Open Access)
Clusterização de padrões espaço-temporais de precipitação na Amazônia via deep convolutional autoencoder
(Universidade Federal do Pará, 2023-07-07) SILVA, Vander Augusto Oliveira da; TEIXEIRA, Raphael Barros; http://lattes.cnpq.br/4902824086591521; https://orcid.org/0000-0003-2993-802X
Studies using different machine learning methods for knowledge discovery and pattern recognition in precipitation time series are increasingly frequent in the literature. Identify and analyze patterns in precipitation time series in a particular region is fundamental for its socioeconomic development. Therefore, it can be stated that knowledge and understanding of the rainfall characteristics of the regions are important to enable the planning of the use, management and conservation of water resources. The natural phenomenon of precipitation is a fundamental process with a direct impact on watersheds and on human and environmental development. The variability of this phenomenon has important implications for the navigability of rivers, individual abundance and species richness. In recent years, many studies with this approach have been carried out in Brazil, mainly in the Amazon region. This research aimed to develop a computational method for analyzing time series of precipitation using machine learning techniques with unsupervised learning, in order to propose an method capable of extracting complex features from the data, obtaining a map of attributes at low dimensionality for pattern recognition, discovery of homogeneous regions with respect to precipitation and approximate reconstruction of precipitation time series in the Legal Amazon. The proposed deep learning neural network model is trained to learn the main and most complex features of the original data and present them in low dimensionality in latent space. After the training, the results are promising, the observations of the reconstructed data showed a good performance as evaluated by the RMSE and NRMSE metric with resulting values equal to 0.06610 and 0.3355 respectively. The analysis of the representation of the data in low dimension was applied and analyzed by a clustering structure using hierarchical agglomerative with Ward’s method. This methodology also showed good results, as it carried out consistent groupings characterizing ho- mogeneous regions in relation to precipitation data. Thus, demonstrating that the representation in low dimensionality carried the main characteristics of the time series of the analyzed data. It is noteworthy that the method developed in this study can be applied not only in the Amazon region, but also in other areas with similar challenges related to time series analysis.
Acesso aberto (Open Access)
Conversão grafema-fone para um sistema de reconhecimento de voz com suporte a grandes vocabulários para o português brasileiro
(Universidade Federal do Pará, 2006-06-12) HOSN, Chadia Nadim Aboul; KLAUTAU JÚNIOR, Aldebaro Barreto da Rocha; http://lattes.cnpq.br/1596629769697284
Speech processing has become a data-driven technology. Hence, the success of research in this area is linked to the existence of public corpora and associated resources, as a phonetic dictionary. In contrast to other languages such as English, one cannot find, in public domain, a Large Vocabulary Continuos Speech Recognition (LVCSR) System for Brazilian Portuguese. This work discusses some efforts within the FalaBrasil initiative [1], developed by researchers, teachers and students of the Signal Processing Laboratory (LaPS) at UFPA, providing an overview of the research and softwares related to Automatic Speech Recognition (ASR) for Brazilian Portuguese. More specifically, the present work discusses the implementation of a large vocabulary ASR for Brazilian Portuguese using the HTK software, which is based on hidden Markov models (HMM). Besides, the work discusses the implementation of a grapheme-phoneme conversion module using machine learning techniques.
Acesso aberto (Open Access)
Custo de oportunidade (trade-off) para diferentes estratégias de manutenção de trilhos ferroviários na Amazônia
(Universidade Federal do Pará, 2022-12-22) CURCINO, Gabrielle dos Anjos; BRAGA, Eduardo de Magalhães; http://lattes.cnpq.br/4783553888547500
The emergency maintenance of railway assets in the Brazilian Amazon has generated revenue losses and opportunity costs. The general objective of this study was to identify the importance of opportunity cost in decision-making for corrective and preventive maintenance strategies. The methodology proposed the modeling of the variables referring to the economic and operational data of railway maintenance in the last ten years, by non-parametric Gradient Boosting Regression Tree machine learning, and hybridizing it with the analysis of the opportunity cost for the trade-off decision making of an ore railroad in the Brazilian Amazon. The results showed that the GBDT was efficient in fitting the training data with r2 equal to one. Similarly, the test data presented satisfactory r2 values, close to one, where the degree of importance of the independent variables in the prediction of the dependent variables was obtained. Pearson's method was used to construct the correlation matrix for each pair of variables. From the generated model, eight forecast groups were created for the year 2022. Then, conflict levels were established, suggested by the economic literature, between the forecast scenarios, where the opportunity cost was identified among the alternatives with the best benefit to maintenance strategies. In this way, the opportunity cost combined with machine learning serves as an instrument to help companies in the search for better maintenance decisions, which contributes to the improvement of rail asset management.
Acesso aberto (Open Access)
Desenvolvimento de sistema de diagnóstico de falhas em roletes de transportadores de correia
(Universidade Federal do Pará, 2024-03-28) SOARES, João Lucas Lobato; MESQUITA, Alexandre Luiz Amarante; http://lattes.cnpq.br/3605920981600245; https://orcid.org/0000-0001-5605-8381
Belt conveyors are essential equipment in mining industry and require constant monitoring to maintain good reliability. In order to support the belt and the material being conveyed, rollers are components that constantly fail during operation, in which they present faults in bearings and surface wear in the shell as the most common failure modes. Thus, monitoring based on predictive maintenance is essential, and machine learning techniques can be used as an alternative for detecting equipment failures. In diagnostics using machine learning, the feature selection step is important to avoid loss of accuracy in the classification of the equipment's condition. The present study analyzes the performance of the decision tree algorithm and Analysis of Variance (ANOVA) as alternative methods for dimensionality reduction. Initially, the vibration signals were collected on the rollers of a belt conveyor bench and the Wavelet Packet Decomposition (WPD) was applied to the signals to obtain the energy ranges, which were used as features for classification. After the determination of the best features, two approaches were analyzed for the selection of features: one with the application of the method without dimensionality reduction and the other with the application of the decision tree. In addition, different classification algorithms were used: Support Vector Machine (SVM), k-Nearest Neighbor (kNN) and Artificial Neural Network (ANN). As a result, it was found a superior performance of diagnostic accuracy in all techniques with a reduction in the dimensionality of the characteristics selected by the decision tree. In addition, SVM, kNN and ANN showed increases in accuracy ranging among the fault diagnosis models approached.
Acesso aberto (Open Access)
Detecção e rastreamento de componentes de vagões ferroviários utilizando redes neurais convolucionais e restricões geométricas
(Universidade Federal do Pará, 2020-04-27) GONÇALVES, Camilo Lélis Assis; BARROS, Fabrício José Brito; http://lattes.cnpq.br/9758585938727609
A inspeção de componentes de trem que podem causar descarrilamento possui um papel importante na manutenção ferroviária. A fim de aumentar a produtividade e a segurança, empresas prestadoras de serviços procuram por soluções de inspeção automáticas e confiáveis. Apesar da inspeção automática baseada em visão computacional ser um conceito consolidado, tais aplicações desafiam a comunidade de desenvolvimento em razão de fatores ambientais e logísticos a serem considerados. Este trabalho propõe uma técnica de detecção e estimativa das posições das regiões de dreno presentes em vagões de trem. Nosso detector/rastreador consiste em uma rede neural convolucional e um conjunto de restrições geométricas, que levam em conta a trajetória ideal dos componentes de interesse dos vagões e as distâncias entre eles. Detalhamos os procedimentos de treinamento e validação, juntamente com as métricas utilizadas para aferir a performance do sistema proposto. Os resultados apresentados são comparados com outras duas técnicas, e exibem um bom custo‑benefício entre confiança e complexidade computacional para a detecção dos componentes de interesse.
Acesso aberto (Open Access)
Development of machine learning-based frameworks to predict permeability of peptides through cell membrane and blood-brain barrier
(Universidade Federal do Pará, 2024-03-27) OLIVEIRA, Ewerton Cristhian Lima de; LIMA, Anderson Henrique Lima e; http://lattes.cnpq.br/2589872959709848; https://orcid.org/0000-0002-8451-9912; SALES JUNIOR, Claudomiro de Souza de; http://lattes.cnpq.br/4742268936279649
Peptides comprise a versatile class of biomolecules with diverse physicochemical and structural properties, in addition to numerous pharmacological and biotechnological applications. Some groups of peptides can cross biological membranes, such as the cell membrane and the human blood-brain barrier. Researchers have explored this property over the years as an alternative to developing more powerful drugs, given that some peptides can also be drug carriers. Although some machine learning-based tools have been developed to predict cell-penetrating peptides (CPPs) and blood-brain barrier penetrating peptides (B3PPs), some points have not yet been explored within this theme. These points encompass the use of dimensionality reduction (DR) techniques in the preprocessing stage, molecular descriptors related to drug bioavailability, and data structures that encode peptides with chemical modifications. Therefore, the primary purpose of this thesis is to develop and test two frameworks based on DR, the first one to predict CPPs and the second to predict B3PPs, also evaluating the molecular descriptors and data structure of interest. The results of this thesis show that for the prediction of penetration in the cell membrane, the proposed framework reached 92% accuracy in the best performance in an independent test, outperforming other tools created for the same purpose, besides evidencing the contribution between the junction of molecular descriptors based on amino acid sequence and those related to bioavailability and cited in Lipinski’s rule of five. Furthermore, the prediction of B3PPs by the proposed framework reveals that the best model using structural, electric, and bioavailability-associated molecular descriptors achieved average accuracy values exceeding 93% in the 10-fold cross-validation and between 75% and 90% accuracy in the independent test for all simulations, outperforming other machine learning (ML) tools developed to predict B3PPs. These results show that the proposed frameworks can be used as an additional tool in predicting the penetration of peptides in these two biomembranes and are available as free-touse web servers.
Acesso aberto (Open Access)
Identificação de sistemas multiforças a partir de dados de vibração e técnicas de aprendizado de máquinas
(Universidade Federal do Pará, 2024-11-07) PINHEIRO, Giovanni de Souza; NUNES, Marcus Vinícius Alves; http://lattes.cnpq.br/9533143193581447
The emergence of defects in dynamic components tends to produce changes in the forces generated, which can be detected through alterations in the vibration response spectrum of the equipment. Understanding the forces acting on a structure is extremely important, especially in cases where measurement points are limited or inaccessible, as it allows for assessing, among other things, whether the component's lifespan is compromised by the current condition of the machine. In such cases, an inverse problem needs to be solved. Machine Learning techniques have been standing out as a powerful tool for prediction among the solutions developed for this type of problem, being increasingly applied to engineering problems. Therefore, this work aims to evaluate different machine learning models for the identification of a system, composed of a suspended plate with one or more applied forces, based on measured vibration data. In this regard, a computational model was generated and calibrated using vibration responses measured in the laboratory. A robust database was created using Response Surface Methodology together with the Design of Experiment (DOE) and then used to assess the ability of machine learning models to predict the location, excitation frequency, magnitude, and number of forces acting on the structure. Among the six machine learning models evaluated, k-NN was able to predict with an error of 0.013%, and random forests showed a maximum error of 0.2%. Finally, a database, containing a line of experimental data, was used to evaluate the k-NN and Random Forest models, obtaining a score of 0.96 and 0.93, respectively. The innovation of the study lies in the application of the proposed method for parameter identification in multiforce systems.
Acesso aberto (Open Access)
Implementação de modelos computacionais na predição temporal e espaço-temporal de parâmetros de qualidade de água
(Universidade Federal do Pará, 2021-12-14) ALMEIDA, Anderson Francisco de Sousa; MERLIN, Bruno; http://lattes.cnpq.br/7336467549495208; HTTPS://ORCID.ORG/0000-0001-7327-9960; GONZÁLEZ, Marcos Tulio Amaris; http://lattes.cnpq.br/9970287865377659
The quality of water is directy related to is level of pollution caused by anthropic and industrial actions, with a consequent reduction in the availability of quality water. Therefore, limological monitorig of the basic parameters os water quality is carried out, as away of obtaining data that guide the decision-making of water resouces management bodies. In this context, the present study has the implementation of machine learning algorithms to predict temporal and spatiotemporal water quality parameter data. The ML techniques used were linear regression, ramdom forest, MLP and LSTM neural networks. Two collection points from a Water Resources Management Unit in São Paulo, Brazil were used. Models are evaluated using MAPE( mean absolute percentage eror) and RMSE( root mean squared erro) metrics. Therefore, in temporal prediction, the LSTM technique presented the best performace in relation to the other techniques and the data used, as it has the lowest average RMSE result, with 2.47. However, in spatiotemporal prediction, MLP has the best performace both in relation to the other techniques and the data used , as it has the lowest averagee results of MAPE and RMSE, respectively, 5.94% and 1.34. Thus, these performaces of neural networks can be justified by the non-linearity of the parameter data. Other than that, the results of the experiments aim to contribute to the water quality monitorng process and assist in the planning of water management, so that it meets current legislation and enales the indication of public policies, through machine learning models in prediction of water quality parametes.
Acesso aberto (Open Access)
Modelo de inteligência artificial para estimativa do desmatamento considerando a rede de transporte rodoviário do estado do Pará
(Universidade Federal do Pará, 2022-01-10) NEVES, Patrícia Bittencourt Tavares das; BLANCO, Claudio José Cavalcante; http://lattes.cnpq.br/8319326553139808; https://orcid.org/0000-0001-8022-2647; DUARTE, André Augusto Azevedo Montenegro; http://lattes.cnpq.br/1135221873341973; https://orcid.org/0000-0003-4586-1587
Since the decade of 1950s the Amazonian and Brazilian transportation complex prioritized the model of road transport. Past studies point that the regular roadway system that is integrated to a clandestine roadway complex is strongly related to the Amazon forest deforestation. Thus, in this work we performed a quantitative analysis of the variables related to the process of deforestation of the Amazon forest, a natural resource of great environment and economic significance, and the socioeconomic development of the region in the period between 1988 and 2018. The geographical study area is the state of Pará, located in the Oriental Amazon, the second largest state of Brazil in territorial extension and the most devastated. We used machine learning in the modeling of the quantitative variables related to the transportation infrastructure, social variables and economic variables, e.g., the devastated area. The random forest model presented the best performance with the generated function (using least squares method). It was estimated the devastated area for the years of 2020, 2030, 2040 and 2050. Sensitivity analysis was used to evaluate the devastated area after the implementation of the roads BR-163 and BR-210 in the north of Pará. The results show that given the current scenario the devastation tends to continue intensively in the next three decades, with a 25.77% increase over the current region albeit with decreasing ten-year rates of forestation loss, and the estimation of the deforested area caused by the implementation of federal roadway networks goes from 4,703.43 km2 to 6,567.48 km2 .
Acesso aberto (Open Access)
Output-only methods for damage identification in structural health monitoring
(Universidade Federal do Pará, 2017-04-27) SANTOS, Adam Dreyton Ferreira dos; FIGUEIREDO, Elói João Faria; http://lattes.cnpq.br/2315380423001185; COSTA, João Crisóstomo Weyl Albuquerque; http://lattes.cnpq.br/9622051867672434
In the structural health monitoring (SHM) field, vibration-based damage identification has become a crucial research area due to its potential to be applied in real-world engineering structures. Assuming that the vibration signals can be measured by employing different types of monitoring systems, when one applies appropriate data treatment, damage-sensitive features can be extracted and used to assess early and progressive structural damage. However, real-world structures are subjected to regular changes in operational and environmental conditions (e.g., temperature, relative humidity, traffic loading and so on) which impose difficulties to identify structural damage as these changes influence different features in a distinguish manner. In this thesis by papers, to overcome this drawback, novel output-only methods are proposed for detecting and quantifying damage on structures under unmeasured operational and environmental influences. The methods are based on the machine learning and artificial intelligence fields and can be classified as kernel- and cluster-based techniques. When the novel methods are compared to the state-of-the-art ones, the results demonstrated that the former ones have better damage detection performance in terms of false-positive (ranging between 3.65.4%) and false-negative (ranging between 0-2.6%) indications of damage, suggesting their applicability for real-world SHM solutions. If the proposed methods are compared to each other, the cluster-based ones, namely the global expectation-maximization approaches based on memetic algorithms, proved to be the best techniques to learn the normal structural condition, without loss of information or sensitivity to the initial parameters, and to detect damage (total errors equal to 4.4%).
Acesso aberto (Open Access)
PredictmodelGUI: ferramenta para classificação de genes essenciais através de técnicas de aprendizado de máquina
(Universidade Federal do Pará, 2025-06-06) MOIA, Gislenne da Silva; SILVA, Cleison Daniel; http://lattes.cnpq.br/1445401605385329; HTTPS://ORCID.ORG/0000-0001-8280-2928; VERAS, Adonney Allan de Oliveira; http://lattes.cnpq.br/2201652617167877; https://orcid.org/0000-0002-7227-0590
DNA sequencing technologies have provided significant advances in the understanding of the genetic content of numerous organisms, ranging from microorganisms to humans. Among the analyses performed in the Omics Sciences, Annotation stands out as one of the most important. Conceptually, this process consists of inferring biological information from genomic sequences, which allows researchers to understand the function of genetic products, such as Genes — the Basic Units of Heredity responsible for the physical and hereditary characteristics of an organism. Some Genes perform vital functions by encoding Proteins or RNAs essential for processes such as Cellular Metabolism, which participate in crucial pathways like Glycolysis and the Tricarboxylic Acid Cycle. Sequencing Platforms have started to generate large volumes of data, which has driven advances in the Omics fields and fostered the development of computational methods aimed at diverse analyses. More recently, Machine Learning and Artificial Intelligence techniques have been applied to these data, with studies demonstrating the effectiveness of biology-inspired approaches. These models do not require rule-based programming, although their creation still demands advanced skills in Programming and Computing. To contribute toward solving this challenge, this study presents PredictModelGUI, a graphical interface developed in Python that implements nine models to classify Essential Genes. The interface allows importing datasets, re-training models, and adjusting parameters. The information is stored in the software database, which ensures traceability and provides a simple and intuitive tool to test different configurations. Available
Acesso aberto (Open Access)
Previsão da irradiação solar utilizando método ensemble para seleção de atributos e algoritmos de aprendizado de máquina
(Universidade Federal do Pará, 2023-06-20) MEJIA, Edna Sofia Solano; AFFONSO, Carolina de Mattos; http://lattes.cnpq.br/2228901515752720
Accurate forecasting of solar irradiance is essential for effective management of power systems with significant photovoltaic generation. Machine learning algorithms, which leverage historical data and patterns to make predictions, play a crucial role in this task. One key aspect is the use of ensemble models that combine the predictions of multiple algorithms to improve forecast accuracy and reliability. In this study, ensemble models are utilized to enhance the forecasting performance by aggregating the predictions of different algorithms. Moreover, the paper proposes an ensemble feature selection method, which involves identifying the most relevant input parameters and their related past observations. This approach aims to optimize the input features used by the machine learning algorithms, ensuring that only the most pertinent information is considered for accurate solar irradiance forecasts. By leveraging the strengths of multiple algorithms and selecting the most informative features, the ensemble approach offers a robust framework for improving the accuracy of solar irradiance predictions. The performance of several machine learning algorithms, including ensemble models, is compared for solar irradiance forecasting on days with different weather patterns using endogenous and exogenous inputs. The algorithms considered are AdaBoost, SVR, RF, XGBT, CatBoost, VOA, and VOWA. The proposed ensemble feature selection relies on the RF, IM, and Relief algorithms. The forecast accuracy is evaluated based on several metrics using a real database of the city of Salvador, Brazil. Different weather forecasts are considered: 1 hour, 2 hours, 3 hours, 6 hours, 9 hours, and 12 hours in advance. Numerical results show that the proposed ensemble feature selection improves forecast accuracy, and that the VOWA model selected with the best-performing algorithms presents forecasts with higher accuracy than the other algorithms at different forecast time horizons. This research demonstrates the effectiveness of ensemble models and feature selection techniques in enhancing solar irradiance forecasting, providing valuable insights for efficient power system management.
Acesso aberto (Open Access)
Previsão de séries temporais no sistema elétrico brasileiro utilizando preditores baseados em aprendizado de máquina: uma análise empírica
(Universidade Federal do Pará, 2024-04-05) CONTE, Thiago Nicolau Magalhães de Souza; OLIVEIRA, Roberto Célio Limão de; http://lattes.cnpq.br/4497607460894318; https://orcid.org/0000-0002-6640-3182
The overview of electric energy in Brazil is influenced by a variety of complex factors and nonlinear relationships, making forecasting challenging. With the increasing demand for energy and growing environmental concerns, it is crucial to seek solutions based on clean and renewable energy practices, aiming to make the energy market more sustainable. These practices aim to reduce waste and optimize the efficiency of processes involved in the operation of electricity distribution and generation technologies. A promising approach to enable sustainable energy is the application of forecasting techniques for various variables in the energy market. This thesis proposes an empirical analysis of the use of regressors to make predictions in the databases of the Price of Settlement Differences (PLD) in the Brazilian market and wind speed in wind turbines in Northeast Brazil, through principal component analysis. We aim to provide significant information about machine learning techniques that can be employed as effective tools for time series prediction in the electric sector. The results obtained may encourage the implementation of these techniques to extract knowledge about the behavior of the Brazilian energy system. This is particularly relevant, given that energy prices often exhibit seasonality, high volatility, and peaks, and wind power generation is widely influenced by weather conditions. To model the prediction of these two time series, we use the database on the Price of Settlement Differences (PLD), focusing especially on the average energy price of the Brazilian National System. The most relevant variables are related to hydrological conditions, electrical load, and fuel prices for thermal units. For collecting variables related to wind energy, two distinct locations in the Northeast region of Brazil were considered: Macau and Petrolina. For the prediction study, we use a Multilayer Perceptron Neural Network (MLP), a Long Short Term Memory (LSTM), Auto-Regressive Integrated Moving Average (ARIMA), and Support Vector Machine (SVM) to determine baseline results in prediction. To enhance the results of these regressors, we employ two different prediction approaches. One approach involves combining deep artificial neural network techniques based on the Canonical Genetic Algorithm (AG) meta-heuristic to adjust the hyperparameters of MLP and LSTM regressors. The second strategy focuses on machine committees, which include MLP, decision tree, linear regression, and SVM in one committee, and MLP, LSTM, SVM, and ARIMA in another. These approaches consider two types of voting, voting average (VO) and voting weighted average (VOWA), to assess the impact on the performance of the machine committee.