Técnicas de controlo não-linear baseadas em Redes Neuronais: do algoritmo à implementação

O presente trabalho analisa soluções de controlo não-linear baseadas em Redes Neuronais e apresenta a sua aplicação a um caso prático, desde o algoritmo de treino até à implementação física em hardware. O estudo inicial do estado da arte da utilização das Redes Neuronais para o controlo leva à proposta de soluções iterativas para a definição da arquitectura das mesmas e para o estudo das técnicas de Regularização e Paragem de Treino Antecipada, através dos Algoritmos Genéticos e à proposta de uma forma de validação dos modelos obtidos. Ao longo da tese são utilizadas quatro malhas para o controlo baseado em modelos, uma das quais uma contribuição original, e é implementado um processo de identificação on-line, tendo por base o algoritmo de treino Levenberg-Marquardt e a técnica de Paragem de Treino Antecipada que permite o controlo de um sistema, sem necessidade de recorrer ao conhecimento prévio das suas características. O trabalho é finalizado com um estudo do hardware comercial disponível para a implementação de Redes Neuronais e com o desenvolvimento de uma solução de hardware utilizando uma FPGA. De referir que o trabalho prático de teste das soluções apresentadas é realizado com dados reais provenientes de um forno eléctrico de escala reduzida.

Algoritmo Q-learning como estratégia de exploração e/ou explotação para metaheurísticas GRASP e algoritmo genético

Techniques of optimization known as metaheuristics have achieved success in the resolution of many problems classified as NP-Hard. These methods use non deterministic approaches that reach very good solutions which, however, don t guarantee the determination of the global optimum. Beyond the inherent difficulties related to the complexity that characterizes the optimization problems, the metaheuristics still face the dilemma of xploration/exploitation, which consists of choosing between a greedy search and a wider exploration of the solution space. A way to guide such algorithms during the searching of better solutions is supplying them with more knowledge of the problem through the use of a intelligent agent, able to recognize promising regions and also identify when they should diversify the direction of the search. This way, this work proposes the use of Reinforcement Learning technique - Q-learning Algorithm - as exploration/exploitation strategy for the metaheuristics GRASP (Greedy Randomized Adaptive Search Procedure) and Genetic Algorithm. The GRASP metaheuristic uses Q-learning instead of the traditional greedy-random algorithm in the construction phase. This replacement has the purpose of improving the quality of the initial solutions that are used in the local search phase of the GRASP, and also provides for the metaheuristic an adaptive memory mechanism that allows the reuse of good previous decisions and also avoids the repetition of bad decisions. In the Genetic Algorithm, the Q-learning algorithm was used to generate an initial population of high fitness, and after a determined number of generations, where the rate of diversity of the population is less than a certain limit L, it also was applied to supply one of the parents to be used in the genetic crossover operator. Another significant change in the hybrid genetic algorithm is the proposal of a mutually interactive cooperation process between the genetic operators and the Q-learning algorithm. In this interactive/cooperative process, the Q-learning algorithm receives an additional update in the matrix of Q-values based on the current best solution of the Genetic Algorithm. The computational experiments presented in this thesis compares the results obtained with the implementation of traditional versions of GRASP metaheuristic and Genetic Algorithm, with those obtained using the proposed hybrid methods. Both algorithms had been applied successfully to the symmetrical Traveling Salesman Problem, which was modeled as a Markov decision process

Análise e síntese de antenas e superfícies seletivas de frequência utilizando computação evolucionária e inteligência de enxames

The frequency selective surfaces, or FSS (Frequency Selective Surfaces), are structures consisting of periodic arrays of conductive elements, called patches, which are usually very thin and they are printed on dielectric layers, or by openings perforated on very thin metallic surfaces, for applications in bands of microwave and millimeter waves. These structures are often used in aircraft, missiles, satellites, radomes, antennae reflector, high gain antennas and microwave ovens, for example. The use of these structures has as main objective filter frequency bands that can be broadcast or rejection, depending on the specificity of the required application. In turn, the modern communication systems such as GSM (Global System for Mobile Communications), RFID (Radio Frequency Identification), Bluetooth, Wi-Fi and WiMAX, whose services are highly demanded by society, have required the development of antennas having, as its main features, and low cost profile, and reduced dimensions and weight. In this context, the microstrip antenna is presented as an excellent choice for communications systems today, because (in addition to meeting the requirements mentioned intrinsically) planar structures are easy to manufacture and integration with other components in microwave circuits. Consequently, the analysis and synthesis of these devices mainly, due to the high possibility of shapes, size and frequency of its elements has been carried out by full-wave models, such as the finite element method, the method of moments and finite difference time domain. However, these methods require an accurate despite great computational effort. In this context, computational intelligence (CI) has been used successfully in the design and optimization of microwave planar structures, as an auxiliary tool and very appropriate, given the complexity of the geometry of the antennas and the FSS considered. The computational intelligence is inspired by natural phenomena such as learning, perception and decision, using techniques such as artificial neural networks, fuzzy logic, fractal geometry and evolutionary computation. This work makes a study of application of computational intelligence using meta-heuristics such as genetic algorithms and swarm intelligence optimization of antennas and frequency selective surfaces. Genetic algorithms are computational search methods based on the theory of natural selection proposed by Darwin and genetics used to solve complex problems, eg, problems where the search space grows with the size of the problem. The particle swarm optimization characteristics including the use of intelligence collectively being applied to optimization problems in many areas of research. The main objective of this work is the use of computational intelligence, the analysis and synthesis of antennas and FSS. We considered the structures of a microstrip planar monopole, ring type, and a cross-dipole FSS. We developed algorithms and optimization results obtained for optimized geometries of antennas and FSS considered. To validate results were designed, constructed and measured several prototypes. The measured results showed excellent agreement with the simulated. Moreover, the results obtained in this study were compared to those simulated using a commercial software has been also observed an excellent agreement. Specifically, the efficiency of techniques used were CI evidenced by simulated and measured, aiming at optimizing the bandwidth of an antenna for wideband operation or UWB (Ultra Wideband), using a genetic algorithm and optimizing the bandwidth, by specifying the length of the air gap between two frequency selective surfaces, using an optimization algorithm particle swarm

Síntese de superfícies seletivas de frequência atráves de técnicas de computação natural

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior

Técnicas de computação natural para segmentação de imagens médicas

Image segmentation is one of the image processing problems that deserves special attention from the scientific community. This work studies unsupervised methods to clustering and pattern recognition applicable to medical image segmentation. Natural Computing based methods have shown very attractive in such tasks and are studied here as a way to verify it's applicability in medical image segmentation. This work treats to implement the following methods: GKA (Genetic K-means Algorithm), GFCMA (Genetic FCM Algorithm), PSOKA (PSO and K-means based Clustering Algorithm) and PSOFCM (PSO and FCM based Clustering Algorithm). Besides, as a way to evaluate the results given by the algorithms, clustering validity indexes are used as quantitative measure. Visual and qualitative evaluations are realized also, mainly using data given by the BrainWeb brain simulator as ground truth

Modulação quantizada para sistemas com codificação wavelet sujeitos ao desvanecimento rayleigh

Wavelet coding is an efficient technique to overcome the multipath fading effects, which are characterized by fluctuations in the intensity of the transmitted signals over wireless channels. Since the wavelet symbols are non-equiprobable, modulation schemes play a significant role in the overall performance of wavelet systems. Thus the development of an efficient design method is crucial to obtain modulation schemes suitable for wavelet systems, principally when these systems employ wavelet encoding matrixes of great dimensions. In this work, it is proposed a design methodology to obtain sub-optimum modulation schemes for wavelet systems over Rayleigh fading channels. In this context, novels signal constellations and quantization schemes are obtained via genetic algorithm and mathematical tools. Numerical results obtained from simulations show that the wavelet-coded systems derived here have very good performance characteristics over fading channels

Separação cega de fontes lineares e não lineares usando algoritmo genético, redes neurais artificiais RBF e negentropia de Rényi como medida de independência

Conventional methods to solve the problem of blind source separation nonlinear, in general, using series of restrictions to obtain the solution, often leading to an imperfect separation of the original sources and high computational cost. In this paper, we propose an alternative measure of independence based on information theory and uses the tools of artificial intelligence to solve problems of blind source separation linear and nonlinear later. In the linear model applies genetic algorithms and Rényi of negentropy as a measure of independence to find a separation matrix from linear mixtures of signals using linear form of waves, audio and images. A comparison with two types of algorithms for Independent Component Analysis widespread in the literature. Subsequently, we use the same measure of independence, as the cost function in the genetic algorithm to recover source signals were mixed by nonlinear functions from an artificial neural network of radial base type. Genetic algorithms are powerful tools for global search, and therefore well suited for use in problems of blind source separation. Tests and analysis are through computer simulations

Otimização em comitês de classificadores: uma abordagem baseada em filtro para seleção de subconjuntos de atributos

Traditional applications of feature selection in areas such as data mining, machine learning and pattern recognition aim to improve the accuracy and to reduce the computational cost of the model. It is done through the removal of redundant, irrelevant or noisy data, finding a representative subset of data that reduces its dimensionality without loss of performance. With the development of research in ensemble of classifiers and the verification that this type of model has better performance than the individual models, if the base classifiers are diverse, comes a new field of application to the research of feature selection. In this new field, it is desired to find diverse subsets of features for the construction of base classifiers for the ensemble systems. This work proposes an approach that maximizes the diversity of the ensembles by selecting subsets of features using a model independent of the learning algorithm and with low computational cost. This is done using bio-inspired metaheuristics with evaluation filter-based criteria

Dilema da diversidade-acurácia: um estudo empírico no contexto de multiclassificadores

Multi-classifier systems, also known as ensembles, have been widely used to solve several problems, because they, often, present better performance than the individual classifiers that form these systems. But, in order to do so, it s necessary that the base classifiers to be as accurate as diverse among themselves this is also known as diversity/accuracy dilemma. Given its importance, some works have investigate the ensembles behavior in context of this dilemma. However, the majority of them address homogenous ensemble, i.e., ensembles composed only of the same type of classifiers. Thus, motivated by this limitation, this thesis, using genetic algorithms, performs a detailed study on the dilemma diversity/accuracy for heterogeneous ensembles

Comitês de Classificadores para o Reconhecimento Multibiométrico em Dados Biométricos Revogáveis

This work discusses the application of techniques of ensembles in multimodal recognition systems development in revocable biometrics. Biometric systems are the future identification techniques and user access control and a proof of this is the constant increases of such systems in current society. However, there is still much advancement to be developed, mainly with regard to the accuracy, security and processing time of such systems. In the search for developing more efficient techniques, the multimodal systems and the use of revocable biometrics are promising, and can model many of the problems involved in traditional biometric recognition. A multimodal system is characterized by combining different techniques of biometric security and overcome many limitations, how: failures in the extraction or processing the dataset. Among the various possibilities to develop a multimodal system, the use of ensembles is a subject quite promising, motivated by performance and flexibility that they are demonstrating over the years, in its many applications. Givin emphasis in relation to safety, one of the biggest problems found is that the biometrics is permanently related with the user and the fact of cannot be changed if compromised. However, this problem has been solved by techniques known as revocable biometrics, which consists of applying a transformation on the biometric data in order to protect the unique characteristics, making its cancellation and replacement. In order to contribute to this important subject, this work compares the performance of individual classifiers methods, as well as the set of classifiers, in the context of the original data and the biometric space transformed by different functions. Another factor to be highlighted is the use of Genetic Algorithms (GA) in different parts of the systems, seeking to further maximize their eficiency. One of the motivations of this development is to evaluate the gain that maximized ensembles systems by different GA can bring to the data in the transformed space. Another relevant factor is to generate revocable systems even more eficient by combining two or more functions of transformations, demonstrating that is possible to extract information of a similar standard through applying different transformation functions. With all this, it is clear the importance of revocable biometrics, ensembles and GA in the development of more eficient biometric systems, something that is increasingly important in the present day