Global temperature variations between 1861 and 1984 are forecast usingsregularization networks, multilayer perceptrons and linearsautoregression. The regularization network, optimized by stochasticsgradient descent associated with colored noise, gives the bestsforecasts. For all the models, prediction errors noticeably increasesafter 1965. These results are consistent with the hypothesis that thesclimate dynamics is characterized by low-dimensional chaos and thatsthe it may have changed at some point after 1965, which is alsosconsistent with the recent idea of climate change.s

On-line negotiation and bargaining systems can work effectively on the Internet based on the prerequisite that user utility functions are known while undergoing transactions. However, this prerequisite is hard to meet due to the variety and anonymous nature of Internet surfing. Therefore, how to rapidly and precisely construct a user¡¦s utility function is an essential issue. This research proposes a radial basis function (RBF) network, a neural network, to model a user¡¦s utility function in order to rapidly and precisely model user utility function. We verify the feasibility of the method through experiments, and compare the performance of RBF networks in prediction performance, time expenses, and subjects¡¦ perceptions with the Multiple Regression (MR), SMARTS, and SMARTER methods. The results show that the RBF network method is feasible in these criteria. Not only the RBF network needs less time to construct the users¡¦ utility function than the SMARTS method does, but also it can model user utility functions more precisely than the MR, SMARTS, and SMARTER methods.

The aim is to study a continuous chemical process, and then analyze the hold process of the reactor and build the models which could be trained to realize the fault diagnosis and localization in the process. An experimental system has been built to be the research base. That includes experiment part and record system. To the diagnosis and localization methods, the work presented the methods with the data-based approach, mainly the Bayesian network and RBF network based on GAAPA (Genetic Algorithm with Auto-adapted of Partial Adjustment). The data collected from the experimental system are used to train and test the models.

A meshless direct pressure-velocity coupling procedure is presented to perform Direct Numerical Simulations (DNS) and Large Eddy Simulations (LES) of turbulent incompressible flows in regular and irregular geometries. The proposed method is a combination of several efficient techniques found in different Computational Fluid Dynamic (CFD) procedures and it is a major improvement of the algorithm published in 2007 by this author. This new procedure has very low numerical diffusion and some preliminary calculations with 2D steady state flows show that viscous effects become negligible faster that ever predicted numerically. The fundamental idea of this proposal lays on several important inconsistencies found in three of the most popular techniques used in CFD, segregated procedures, streamline-vorticity formulation for 2D viscous flows and the fractional-step method, very popular in DNS/LES. The inconsistencies found become important in elliptic flows and they might lead to some wrong solutions if coarse grids are used. In all methods studied, the mathematical basement was found to be correct in most cases, but inconsistencies were found when writing the boundary conditions. In all methods analyzed, it was found that it is basically impossible to satisfy the exact set of boundary conditions and all formulations use a reduced set, valid for parabolic flows only. For example, for segregated methods, boundary condition of normal derivative for pressure zero is valid only in parabolic flows. Additionally, the complete proposal for mass balance correction is right exclusively for parabolic flows. In the streamline-vorticity formulation, the boundary conditions normally used for the streamline function, violates the no-slip condition for viscous flow. Finally, in the fractional-step method, the boundary condition for pseudo-velocity implies a zero normal derivative for pressure in the wall (correct in parabolic flows only) and, when the flows reaches steady state, the procedure does not guarantee mass balance. The proposed procedure is validated in two cases of 2D flow in steady state, backward-facing step and lid-driven cavity. Comparisons are performed with experiments and excellent agreement was obtained in the solutions that were free from numerical instabilities. A study on grid usage is done. It was found that if the discretized equations are written in terms of a local Reynolds number, a strong criterion can be developed to determine, in advance, the grid requirements for any fluid flow calculation. The 2D-DNS on parallel plates is presented to study the basic features present in the simulation of any turbulent flow. Calculations were performed on a short geometry, using a uniform and very fine grid to avoid any numerical instability. Inflow conditions were white noise and high frequency oscillations. Results suggest that, if no numerical instability is present, inflow conditions alone are not enough to sustain permanently the turbulent regime. Finally, the 2D-DNS on a backward-facing step is studied. Expansion ratios of 1.14 and 1.40 are used and calculations are performed in the transitional regime. Inflow conditions were white noise and high frequency oscillations. In general, good agreement is found on most variables when comparing with experimental data.

In recent history, the area of physics-based engineering simulation has seen rapid increases in both computer workstation performance as well as common model complexity, both driven largely in part by advances in memory density and availability of clusters and multi-core processors. While the increase in computation time due to model complexity has been largely offset by the increased performance of modern workstations, the increase in model setup time due to model complexity has continued to rise. As such, the major time requirement for solving an engineering model has transitioned from computation time to problem setup time. This is due to the fact that developing the required mesh for complex geometry can be an extremely complicated and time consuming task. Consequently, new solution techniques which are capable of reducing the required amount of human interaction are desirable. The subject of this thesis is the development of a novel meshless method that promises to eliminate the need for structured meshes, and thus, the need for complicated meshing procedures. Although the savings gain due to eliminating the meshing process would be more than sufficient to warrant further study, the proposed method is also capable of reducing the computation time and memory footprint compared to similar models solved using more traditional finite element, finite difference, finite volume, or boundary element methods. In particular, this thesis will outline the development of an interactive, meshless, physically accurate modeling environment that provides an extensible framework which can be applied to a multitude of governing equations encountered in computational mechanics and thermofluids. Additionally, through the development of tailored preprocessing routines, efficiency and accuracy of the proposed meshless algorithms can be tested in a more realistic and flexible environment. Examples are provided in the areas of elasticity, heat transfer and computational fluid dynamics.

No description available.

A filtragem, na sua forma mais geral, tem estado presente na vida do homem há muito tempo. Com o surgimento de novas tecnologias (surgimento da eletricidade e a sua evolução) e o desenvolvimento da computação, as técnicas de filtragem (separação) de sinais elétricos. Normalmente, os sistemas de comunicação (telefonia móvel e fixa, sinais recebidos de satélites e outros sistemas) contém sinais indesejáveis responsáveis pela degradação do sinal original. Dentro desse contexto, este projeto de pesquisa apresenta um estudo do algoritmo Filtro Duplo de Kalman Estendido, onde um filtro e Kalman e duas redes neurais são empregadas para a redução de ruídos em sinais de voz. O algoritmo estudado foi aplicado ao processamento de um sinal corrompido por dois tipos de ruídos diferentes: ruído branco e ruído gaussiano e ruído branco não estacionário, conseguindo-se bons resultados. Uma melhora sensível do sinal filtrado pode ser conseguida com técnicas de pré-filtragem do sinal. Neste trabalho foi utilizado o filtro de médias para a pré-filtragem, obtendo um sinal filtrado com ruído musical de baixa intensidade. / Filtering in it\'s most general kind has been present in men\'s life for a long time. With the appearance of new technologies (appearance of electricity and it\'s evolution) and the deyelopment of the computer science, the filtering techniques started to be widely used in engineering to the filtering (separation) of electric signals. Normally the communication systems (fixed and mobile telephony, signals sent from satellites and other systems) bring undesired results responsible for the degradation of the original signal. Within this context, this research project shows a study of the algorithm Dual Extended Kalman Filtering, in which a Kalman filter and two neural networks are used for the reduction of noise in speech signals. The algorithm studied was applied to the processing of a signal corrupted by two types of different noises: gaussian white noise and non stationary white noise obtaining good results. A significant improvement of the filtered noise can be obtained with techniques of pre-filtering of the signal. In this research the average filter for a pre-filtering was used, obtaining a filtered signal with musical noise oflow intensity.

Com base em exames como a videolaringoscopia, que é considerado um procedimento médico invasivo e desconfortável, diagnósticos têmsido realizados visando detectar patologias na laringe. Geralmente, esse tipo de exame é realizado somente com solicitação médica e quando alterações na fala já são marcantes, ou há sensação de dor. Nessa fase, muitas vezes a doença está em grau avançado, dificultando o seu tratamento. Com o objetivo de realizar um pré-diagnóstico computacional de tais patologias, este trabalho apresenta uma técnica não invasiva na qual são testados e comparados três classificadores: a Distância Euclidiana, a Rede Neural RBF com o kernel Gaussiano e a Rede Neural RBF com o kernel Gaussiano modificado. Testes realizados com uma base de dados de vozes normais e aquelas afetadas por diversas patologias demonstram a eficácia da técnica proposta, que pode, inclusive, ser implementada em tempo-real. / Based on examinations such as laryngoscopy, which is considered an invasive and uncomfortable procedure, diagnosis have been performed aiming at the detection of larynx pathologies. Usually, this type of test is carried out upon medical request and when the speech changes are notable or are causing pain. At this point, the disease is possibly at an advanced degree, complicating its treatment. In order to perform a computational pre-diagnosis of such conditions, this work proposes a noninvasive technique in which three classifiers are tested and compared: the Euclidean distance, the RBF Neural Network with the Gaussian kernel and RBF Neural Network with a modified Gaussian kernel. Tests carried out with a database of normal voices and those affected by various pathologies demonstrate the effectiveness of the technique that may even be implemented to work in real time.

A filtragem, na sua forma mais geral, tem estado presente na vida do homem há muito tempo. Com o surgimento de novas tecnologias (surgimento da eletricidade e a sua evolução) e o desenvolvimento da computação, as técnicas de filtragem (separação) de sinais elétricos. Normalmente, os sistemas de comunicação (telefonia móvel e fixa, sinais recebidos de satélites e outros sistemas) contém sinais indesejáveis responsáveis pela degradação do sinal original. Dentro desse contexto, este projeto de pesquisa apresenta um estudo do algoritmo Filtro Duplo de Kalman Estendido, onde um filtro e Kalman e duas redes neurais são empregadas para a redução de ruídos em sinais de voz. O algoritmo estudado foi aplicado ao processamento de um sinal corrompido por dois tipos de ruídos diferentes: ruído branco e ruído gaussiano e ruído branco não estacionário, conseguindo-se bons resultados. Uma melhora sensível do sinal filtrado pode ser conseguida com técnicas de pré-filtragem do sinal. Neste trabalho foi utilizado o filtro de médias para a pré-filtragem, obtendo um sinal filtrado com ruído musical de baixa intensidade. / Filtering in it\'s most general kind has been present in men\'s life for a long time. With the appearance of new technologies (appearance of electricity and it\'s evolution) and the deyelopment of the computer science, the filtering techniques started to be widely used in engineering to the filtering (separation) of electric signals. Normally the communication systems (fixed and mobile telephony, signals sent from satellites and other systems) bring undesired results responsible for the degradation of the original signal. Within this context, this research project shows a study of the algorithm Dual Extended Kalman Filtering, in which a Kalman filter and two neural networks are used for the reduction of noise in speech signals. The algorithm studied was applied to the processing of a signal corrupted by two types of different noises: gaussian white noise and non stationary white noise obtaining good results. A significant improvement of the filtered noise can be obtained with techniques of pre-filtering of the signal. In this research the average filter for a pre-filtering was used, obtaining a filtered signal with musical noise oflow intensity.

Com base em exames como a videolaringoscopia, que é considerado um procedimento médico invasivo e desconfortável, diagnósticos têmsido realizados visando detectar patologias na laringe. Geralmente, esse tipo de exame é realizado somente com solicitação médica e quando alterações na fala já são marcantes, ou há sensação de dor. Nessa fase, muitas vezes a doença está em grau avançado, dificultando o seu tratamento. Com o objetivo de realizar um pré-diagnóstico computacional de tais patologias, este trabalho apresenta uma técnica não invasiva na qual são testados e comparados três classificadores: a Distância Euclidiana, a Rede Neural RBF com o kernel Gaussiano e a Rede Neural RBF com o kernel Gaussiano modificado. Testes realizados com uma base de dados de vozes normais e aquelas afetadas por diversas patologias demonstram a eficácia da técnica proposta, que pode, inclusive, ser implementada em tempo-real. / Based on examinations such as laryngoscopy, which is considered an invasive and uncomfortable procedure, diagnosis have been performed aiming at the detection of larynx pathologies. Usually, this type of test is carried out upon medical request and when the speech changes are notable or are causing pain. At this point, the disease is possibly at an advanced degree, complicating its treatment. In order to perform a computational pre-diagnosis of such conditions, this work proposes a noninvasive technique in which three classifiers are tested and compared: the Euclidean distance, the RBF Neural Network with the Gaussian kernel and RBF Neural Network with a modified Gaussian kernel. Tests carried out with a database of normal voices and those affected by various pathologies demonstrate the effectiveness of the technique that may even be implemented to work in real time.