Made available in DSpace on 2017-07-10T18:08:07Z (GMT). No. of bitstreams: 1 Tulio Klassen.pdf: 1461464 bytes, checksum: 02d111496a486b94ea73232034f2aeb6 (MD5) Previous issue date: 2008-02-11 / The Artificial Neural Networks have been used with success for the description and modeling of processes in the most several areas of the knowledge, from economy, administration, artificial intelligence and even control of complex industrial processes. The process of chilling of chickens for immersion in cold water ("chillers") is complex and difficult to be modeled phenomenologicaly, because it involves transfer of heat, mass and transient regime, besides a great number of variables. In this work several architectures of artificial neural networks were used in the description and modeling of the process of chilling of the chickens, foreseeing the final temperature and the growth of weight of the carcasses. Also for comparison effect they were used an empiric model proposed by CARCIOFI & LAURINDO (2007) to describe the absorption of the water for the carcasses and the chilling model according to Newton's Law for the temperature of the carcasses. Different situations were tested changing the numbers of neurons of the entrance and hidden layers, and the number of layers. The data used were supplied by the SADIA - Toledo company for training and validation of the net. For the model twenty-five entrance variables were selected, as weight of the carcass, temperature before the chillers, temperature of the propilenoglicol shirt, flow of water in each module of the tanks, time of chilling and temperature of the renewal water, bubble intensity and amount of ice. The results obtained by the neural network and for Newton's Law they were not efficient to represent the final temperature of the carcass. The neural networks and the empiric model of CARCIOFI & LAURINDO (2007) went very efficient to esteem the amount of water absorbed for the carcasses. The obtained results showed that the net type with 4 x 12 x 4 neurons in the entrance layer, first and second hidden layers respectively was the best to represent the investigated system. / As Redes Neurais Artificiais têm sido empregadas com sucesso para a descrição e modelagem de processos nas mais diversas áreas do conhecimento, desde economia, administração, inteligência artificial e até controle de processos industriais complexos. O processo de resfriamento de frangos por imersão em água gelada ( chillers ) é complexo e difícil de ser modelado fenomenologicamente, pois envolve transferência de calor, massa e regime transiente, além de um grande número de variáveis. Neste trabalho foram empregadas diversas arquiteturas de redes neurais artificiais na descrição e modelagem do processo de resfriamento dos frangos, prevendo a temperatura final e o ganho de peso das carcaças. Também para efeito de comparação foram empregados um modelo empírico proposto por CARCIOFI & LAURINDO (2007) para descrever a absorção da água pelas carcaças e o modelo de resfriamento segundo a Lei de Newton para a temperatura das carcaças. Foram testadas diferentes situações alterando-se os números de neurônios das camadas de entrada e intermediária, e o número de camadas. Foram utilizados dados fornecidos pela empresa SADIA Toledo para treinamento e validação da rede. Para o modelo foram selecionadas vinte e cinco variáveis de entrada, como peso da carcaça, temperatura antes do resfriamento, temperatura da camisa de propilenoglicol, vazão de água em cada módulo dos tanques, tempo de resfriamento e temperatura da água de renovação, borbulhamento e quantidade de gelo. Os resultados obtidos pelas redes neurais e pela Lei de Newton não foram eficientes para representar a temperatura de saída da carcaça. As redes neurais e o modelo empírico de CARCIOFI & LAURINDO (2007) foram muito eficientes para estimar a quantidade de água absorvida pelas carcaças. Os resultados obtidos mostraram que a rede tipo 4 x 12 x 4 neurônios na camada de entrada, primeira intermediária e segunda intermediária respectivamente foi a que melhor representou o sistema investigado.

Made available in DSpace on 2016-08-17T14:52:37Z (GMT). No. of bitstreams: 1 Ricardo Luis da Rocha Ataide.pdf: 1712992 bytes, checksum: 27d451c245e151370c1c17a8e89cf8bb (MD5) Previous issue date: 2007-12-27 / Most of the existing software of wireless intrusion detection identify behaviors obtrusive only taking as a basis the exploitation of known vulnerabilities commonly called of attack signatures. They analyze the activity of the system, watching sets of events that are similar to a pre-determined pattern that describes an intrusion known. Thus, only known vulnerabilities are detected, leading to the need for new techniques for detecting intrusions be constantly added to the system. It is necessary to implement a wireless IDS that can identify intrusive behaviors also based on the observation of the deflection normal behaviour of the users, hosts or network connections. This normal behaviour should be based on historical data, collected over a long period of normal operation. This present work proposes an architecture for a system to intrusion detection in wireless networks by anomalies, which is based on the application of technology to artificial neural networks, both in the processes of intrusion detection, as making countermeasures. The system can be adapted to the profile of a new community of users, and can recognize attacks with characteristics somewhat different from the already known by the system, relying only on deviations in behaviour of this new community. A prototype has been implemented and various simulations and tests were performed on it, with three denial of service attacks. The tests were to verify the effectiveness of the application of neural networks in the solution of the problem of wireless network intrusion detection, and concentrated its focus on the power of generalization of neural networks. This ensures the system detects attacks though these features slightly different from those already known. / A maioria dos sistemas de detecção de intrusos para redes wireless existentes identificam comportamentos intrusivos apenas tomando como base a exploração de vulnerabilidades conhecidas, comumente chamadas de assinaturas de ataques. Eles analisam a atividade do sistema, observando conjuntos de eventos que sejam semelhantes a um padrão pré-determinado que descreva uma intrusão conhecida. Com isso, apenas vulnerabilidades conhecidas são detectadas, trazendo a necessidade de que novas técnicas de intrusão sejam constantemente adicionadas ao sistema. Torna-se necessária a implementação de um WIDS (Wireless Intrusion Detection System) que possa identificar comportamentos intrusivos baseandose também na observação de desvios do comportamento normal dos usuários, computadores pessoais ou conexões da rede. Esse comportamento normal deve se basear em dados históricos, coletados durante um longo período normal de operação. Este trabalho propõe uma arquitetura para um sistema de detecção de intrusos em redes wireless por anomalias, que tem como base a aplicação da tecnologia de redes neurais artificiais, tanto nos processos de detecção de intrusões quanto de tomada de contramedidas. O sistema pode se adaptar ao perfil de uma nova comunidade de usuários, bem como pode reconhecer ataques com características um pouco diferentes das já conhecidas pelo sistema, baseando-se apenas nos desvios de comportamento dessa nova comunidade. Um protótipo foi implementado e várias simulações e testes desse protótipo foram realizadas, para três ataques de negação de serviço. Os testes tiveram o objetivo de verificar a eficácia da aplicacação de redes neurais na solução do problema da detecção de intrusos em redes wireless, concentrando seu foco no poder de generalização das redes neurais. Isto garante que o sistema detecte ataques ainda que estes apresentem características ligeiramente diferentes das já conhecidas. Redes Neurais Artificiais.

Made available in DSpace on 2016-08-17T14:52:58Z (GMT). No. of bitstreams: 1 Raimundo Nonato Rosa Filho.pdf: 1125835 bytes, checksum: 91689e7b58443f6d0eb73d752860ce37 (MD5) Previous issue date: 2005-03-31 / In this work a methodology for reduction of maintenance cost in the on-load tap changers (OLTC) of extra high voltage is proposed. The methodology is based on the use of Artificial Neural Networks (ANN) for the intelligent processing of input signals of the commutator. The neural nets adequately trained allow to create an information system and dedicated diagnosis of the OLTC. This system can interpret and diagnosis the components through the real time input signals in order to delay the power transformer maintenance intervals, foreseeing when the OLTC is going to maintenance have intervention based on its condition. It has been adopted a multiperceptron ANN architecture in which the input vector has 22 components and the output considers only one component with the status of the OLTC condition in function of its operation time. This output information is used to determine the periods of maintenance of the commutators. It is reported an application of the proposed system considering the on load tap changer of an autotransformer bank of 500/230/13.8 kV, 600MVA of Centrais Elétricas do Norte do Brasil S/A (ELETRONORTE). The results indicate the advantages of the maintenance based on the condition using ANN. / Neste trabalho é proposta uma metodologia para redução de custo de manutenção nos comutadores de tap sob carga (OLTC) dos transformadores de potência de extra alta tensão. A metodologia está baseada na utilização de redes neurais artificiais (RNA) para o processamento inteligente dos sinais de entrada dos comutadores. As redes neurais adequadamente treinadas permitem criar um sistema de informação e diagnóstico dedicado a OLTC que podem interpretar e diagnosticar os componentes através das entradas em tempo real de forma a, postergar os intervalos de manutenção, prevendo quando o OLTC deverá sofrer intervenção de manutenção baseada na condição do OLTC. Foi adotada uma arquitetura de RNA de multiperceptron na qual a entrada considera um vetor com 22 entrada e apenas uma saída com o status da condição do OLTC em função do tempo de operação. Essa informação de saída é utilizada para determinar os períodos de manutenção dos comutadores de tap. É realizada uma aplicação do sistema proposto considerando o comutador de tap sob carga de um banco de autotransformador de 500/230/13.8kV, 600MVA da Centrais Elétricas do Norte do Brasil S/A( ELETRONORTE) e os resultados indicam as vantagens da manutenção baseada na condição usando RNA.

Made available in DSpace on 2016-08-17T14:53:19Z (GMT). No. of bitstreams: 1 dissertacao Raimundo.pdf: 3146531 bytes, checksum: 40d7a999c6dda565c6701f7cc4a171aa (MD5) Previous issue date: 2011-07-28 / The defense mechanisms expansion for cyber-attacks combat led to the malware evolution, which have become more structured to break these new safety barriers. Among the numerous malware, Botnet has become the biggest cyber threat due to its ability of controlling, the potentiality of making distributed attacks and because of the existing structure of control. The intrusion detection and prevention has had an increasingly important role in network computer security. In an intrusion detection system, information about the current situation and knowledge about the attacks contribute to the effectiveness of security process against this new cyber threat. The proposed solution presents an Intrusion Detection System (IDS) model which aims to expand Botnet detectors through active objects system by proposing a technology with collect by sensors, preprocessing filter and detection based on signature and anomaly, supported by the artificial intelligence method Particle Swarm Optimization (PSO) and Artificial Neural Networks. / A ampliação dos mecanismos de defesas no uso do combate de ataques ocasionou a evolução dos malwares, que se tornaram cada vez mais estruturados para o rompimento destas novas barreiras de segurança. Dentre os inúmeros malwares, a Botnet tornou-se uma grande ameaça cibernética, pela capacidade de controle e da potencialidade de ataques distribuídos e da estrutura de controle existente. A detecção e a prevenção de intrusão desempenham um papel cada vez mais importante na segurança de redes de computadores. Em um sistema de detecção de intrusão, as informações sobre a situação atual e os conhecimentos sobre os ataques tornam mais eficazes o processo de segurança diante desta nova ameaça cibernética. A solução proposta apresenta um modelo de Sistema de Detecção de Intrusos (IDS) que visa na ampliação de detectores de Botnet através da utilização de sistemas objetos ativos, propondo uma tecnologia de coleta por sensores, filtro de pré-processamento e detecção baseada em assinatura e anomalia, auxiliado pelo método de inteligência artificial Otimização de Enxame da Partícula (PSO) e Redes Neurais Artificiais.

Made available in DSpace on 2016-08-17T14:53:21Z (GMT). No. of bitstreams: 1 dissertacao Simone Cristina.pdf: 915238 bytes, checksum: 6eb097a7ebfb66da176cd431d9883ba3 (MD5) Previous issue date: 2012-07-24 / The ovarian cancer is difficult to diagnose in the early stages of development. In this work we bring a study of a new method that gave us great accuracy rates based on a bioinformatics tool called surface enhanced for laser desorption and ionization (SELDI-TOF) used to generate proteomic patterns which is one of the technologies advanced in the diagnosis. Our goal is to contribute to effectiveness of this tool, which already helps diagnosis earlier, our methodology uses independent component analysis (ICA) for feature extraction and neural networks to classify between malignancy and no malignancy in a database of the research center cancer in the U.S.A. Our work rates obtained acurracy 97%, 98% specificity and 96% sensitivity. / O câncer de ovário possui difícil diagnóstico nas primeiras fases de desenvolvimento. Neste trabalho trazemos um estudo de um novo método que nos deu ótimas taxas de precisão baseado em uma ferramenta da bio-informática chamada superfície mehorada a laser para ionização e dessorção (SELDI-TOF) usada para geração de padrões proteômicos que é uma das tecnologias mais avançada no auxílio ao diagnóstico. Nosso objetivo é contribuir para eficácia desta esta ferramenta, que já auxilia o dignóstico precoce, nossa metodologia usa análise de componentes independentes (ICA) para extração de caractéristicas e redes neurais para classificar entre malignidade e não malignidade em uma base de dados do centro de pesquisa do câncer nos EUA. Nosso trabalho obteve taxas de 97% de acurária, 98% de especifidade e 96 % de sensibilidade.

Orientador: Carlos Alberto de Castro Jr / Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Engenharia Eletrica e de Computação / Made available in DSpace on 2018-08-04T04:11:24Z (GMT). No. of bitstreams: 1 JimenezCifuentes_Alberto_M.pdf: 868822 bytes, checksum: 18c5f2e3c043eab417e1404db98cf0d0 (MD5) Previous issue date: 2005 / Resumo: As mudan¸cas no setor eletrico tem influido de maneira dr'astica nas condicoes de operacao dos sistemas, levando-os a operar proximos da sua maxima capacidade. Por esta razao, estimar a proximidade do ponto de operacao da rede aos limites do sistema de forma eficiente tem se tornado uma tarefa fundamental na operacao de sistemas el'etricos de pot¿encia. Do ponto de vista da estabilidade de tensao e interessante conhecer a margem de carregamento, ja que ela fornece uma id'eia precisa da proximidade do sistema ao colapso de tensao.Este trabalho propoe o treinamento de redes neurais artificiais para avaliar a margem de carregamento sob condicoes de operacao normais e de contingencia a partir das informacoes coletadas da previs¿ao de carga e usando diferentes grandezas mensuraveis diretamente do sistema ou atraves de calculos simples, visando a aplicacao desta metodologia num ambiente de tempo real / Abstract: Deregulation of electricity industry has drastically changed the operating conditions of power systems (PS). This fact leads them to operate close to their maximum capability. Thus, estimating efficiently the proximity of the operating point to its limits has become an essential task in PS. In static voltage stability analysis it is interesting to know the load margin (LM), which gives the proximity to the collapse point. This work proposes a neural network training in order to evaluate the LM not only under normal operating conditions, but also disturbance conditions, using either collected data of several measured parameters available from the system or through simple computations toward applying this methodology on real systems under on-line analysis / Mestrado / Energia Eletrica / Mestre em Engenharia Elétrica

La spectrométrie de plasma induit par laser, plus connue sous le nom de LIBS (l’acronyme du terme en anglais Laser-Induced Breakdown Spectroscopy) est une technique analytique qui permet la détection de l’ensemble des éléments du tableau périodique avec des limites de détection de l’ordre du ppm et ceci sur tous types d’échantillons qu’ils soient liquides, solides ou gazeux. Sa simplicité de mise en œuvre, sa rapidité et sa versatilité en font une technique très attractive avec un fort potentiel en termes d’applications que ce soit pour le contrôle en ligne, l’environnement ou l’exploration spatiale. Son point faible reste cependant son manque de fiabilité dans l’analyse quantitative, en particulier lors de l’étude d’échantillons hétérogènes ou de matrices complexes telles que les matrices organiques. Ce travail de thèse propose une étude des propriétés des plasmas induit par laser sur différentes familles de polymères. Une étude du plasma au temps court (~ns) par ombroscopie est tout d’abord présentée, ceci pour différents paramètres expérimentaux (énergie laser, durée d’impulsion, longueur d’onde). Un diagnostic complet du plasma par spectrométrie d’émission est ensuite détaillé pour différents délais de détection et montre que la mesure des températures des différentes espèces du plasma (atomique, ionique et moléculaire) permet de vérifier, dans certaines conditions, les hypothèses d’homogénéité et de l’équilibre thermodynamique local. Ceci permet alors la mise en place de procédures quantitatives telles que la méthode dite sans calibration (calibration free LIBS) tout en optimisant le rapport signal sur bruit de la mesure LIBS. Dans nos expériences cette optimisation est mise à profit pour l’identification de différentes familles de polymères en utilisant, pour le traitement des données de la spectroscopie LIBS, la méthode chimiométrique des réseaux de neurones artificiels. Les résultats obtenus, très prometteurs, permettent d’envisager l’utilisation de la LIBS pour l’identification en temps réel des matières plastiques sur chaine de tri. Par ailleurs et de manière plus générale, ce travail pourrait constituer une base solide pour aller étudier d’autres matériaux organiques plus complexes tels que des tissus biologiques. / Laser-Induced Breakdown Spectroscopy (LIBS) is an analytical technique that has the potential to detect all the elements present in the periodic table. The limit of detection can go below a few ppm and this regardless of the physical phase of the analyzed sample (solid, liquid or gas). Its simplicity of use, its rapidity to get results and its versatility provide this technique with attractive features. The technique is currently developed for applications in a large number of domains such as online control, spatial explorations and the environment. However the weakness of the LIBS technique, compared to other more conventional ones, is still its difficulty in providing reliable quantitative results, especially for inhomogeneous and complex matrix such as organic or biological materials. The work presented in this thesis includes a study of the properties of plasma induced from different organic materials. First, a study of the plasma induced on the surface of a Nylon sample at short time delays (~ns) was carried out using the time-resolved shadowgraph technique for different experimental parameters (laser energy, pulse duration, wavelength). Then, a complete diagnostics of the plasma was performed using the plasma emission spectroscopy. A detailed analysis of the emission spectra at different detection delays allowed us to determine the evolution of the temperatures of the different species in the plasma (atoms, ions and molecules). The homogeneity and the local thermodynamic equilibrium within the plasma was then experimentally checked and validated. We demonstrated that the optimisation of the signalto- noise ratio and a quantitative procedure, such as the calibration-free LIBS, can be put in place within a properly chosen detection window. In our experiments, such optimised detection configuration was further employed to record LIBS spectra from different families of polymer in order to identify and classify them. For this purpose, the chemometrics procedure of artificial neural networks (ANN) was used to process the recorded LIBS spectroscopic data. The promising results obtained in this thesis makes LIBS stand out as a potentially useful tool for real time identification of plastic materials. Finally, this work can also be considered as a base for the further studies of more complex materials such as biological tissues with LIBS.

[pt] Esta dissertação de mestrado analisa os efeitos de previsão de séries temporais com redes neurais em conjunto com a técnica de poda, denominada de Regularização Bayesiana. Utilizam-se diversas séries simuladas cujo processo gerador é de fato linear para comparar as previsões feitas por meio de modelos auto-regressivos lineares e redes neurais. Apresenta-se,ao final, uma comparação entre os modelos citados acima, segundo à eficiência preditiva de cada um. / [en] This paper studies the performance of neural networks estimated with Bayesian regularization to model and forecast time series where the data generations process is in fact linear. A simulation experiment is carried out to compare the forecast made by linear autoregressive models and neural networks.

雲端的盛行使得人們做任何事都要透過網路，但是總會有些有心人士使用一些惡意程式來創造攻擊或通過網絡連接竊取資料。為了防止這些網路惡意攻擊，我們必須不斷檢查網路流量資料，然而現在這個雲端時代，網路的資料是非常龐大且複雜，若要檢查所有網路資料不僅耗時而且非常沒有效率。 本研究使用TensorFlow與多個圖形處理器(Graphics Processing Unit, GPU)來實作類神經網路(Artificial Neural Networks, ANN)機制，用以分析網路流量資料，並得到一個可以判斷正常與異常網路流量的偵測規則，也設計一個實驗來驗證我們提出的類神經網路機制是否符合網路流向異常偵測的時效性和有效性。 在實驗過程中，我們發現使用更多的GPU可以減少訓練類神經網路的時間，並且在我們的實驗設計中使用三個GPU進行運算可以達到網路流量異常偵測的時效性。透過該方法得到的初步實驗結果，我們提出機制的結果優於使用反向傳播算法訓練類神經網路得到的結果。 / The prosperity of the cloud makes people do anything through the Internet, but there are people with bad intention to use some malicious programs to create attacks or steal information through the network connection. In order to prevent these cyber-attacks, we have to keep checking the network traffic information. However, in the current cloud environment, the network information is huge and complex that to check all the information is not only time-consuming but also inefficient. This study uses TensorFlow with multiple Graphic Processing Units (GPUs) to implement an Artificial Neural Networks (ANN) mechanism to analyze network traffic data and derive detection rules that can identify normal and malicious traffics, and we call it Network Traffic Anomaly Detection (NTAD). Experiments are also designed to verify the timeliness and effectiveness of the derived ANN mechanism. During the experiment, we found that using more GPUs can reduce training time, and using three GPUs to do the operation can meet the timeliness in NTAD. As a result of this method, the experiment result was better than ANN with back propagation mechanism.

SYRACO est un SYstème de Reconnaissance Automatisée des COccolithes, développé à son origine par Luc Beaufort et Denis Dollfus à partir de 1995 et plus récemment avec Yves Gally. L'utilité d'un tel système est de permettre aux spécialistes un gain de temps majeur dans l'acquisition et le traitement des données. Dans ce travail, le système a été amélioré techniquement et sa reconnaissance a été étendue aux nannofossiles calcaires du Cénozoïque. Ce système fait le tri entre les nannofossiles et les non-nannofossiles avec une efficacité respectivement estimée à 75% et 90 %. Il s'appuie sur une nouvelle base d'images de référence d'espèces datant de l'Eocène Supérieur aux espèces vivantes, ce qui représente des centaines d'espèces avec une forte variabilité morphologique. Il permet de réaliser une classification en 39 morphogroupes par la combinaison de réseaux de neurones artificiels avec des modèles statistiques. Les résultats sont présentés sous forme de comptages automatisés, de données morphométriques (taille, masse...) et de mosaïques d'images. Il peut ainsi être utilisé pour des analyses biostratigraphiques et paléocéanographiques. / SYRACO is an automated recognition system of coccoliths, originally developed since 1995 by Luc Beaufort and Denis Dollfus, and more recently with the help of Yves Gally. The main purpose of this system is for specialists to save time in the acquisition and treatment of data. By this recent work, the system has been technically improved and its ability of recognition has been extended to calcareous nannofossils of the Cenozoic Era. It sorts nannofossils and non-nannofossils with a reliability respectively estimated to 75% and 90%. It is based on a new reference images database of species from the Upper Eocene up to living species. This represents hundreds of species with a high morphological variability. It leads to the establishment of a classification arranged in 39 morphogroups, combining artificial neural networks to statistical models. The results are presented as automated counting, morphometrical data (size, mass...) and mosaics of images. Those results can be valuable in biostratigraphical and paleoceanographical analyses.