Geração e Simplificação da Base de Conhecimento de um Sistema Híbrido Fuzzy- Genético propõe uma metodologia para o desenvolvimento da base de conhecimento de sistemas fuzzy, fundamentada em técnicas de computação evolucionária. Os sistemas fuzzy evoluídos são avaliados segundo dois critérios distintos: desempenho e interpretabilidade. Uma metodologia para a análise de problemas multiobjetivo utilizando a Lógica Fuzzy foi também desenvolvida para esse fim e incorporada ao processo de avaliação dos AGs. Os sistemas fuzzy evoluídos foram avaliados através de simulações computacionais e os resultados obtidos foram comparados com os obtidos por outros métodos em diferentes tipos de aplicações. O uso da metodologia proposta demonstrou que os sistemas fuzzy evoluídos possuem um bom desempenho aliado a uma boa interpretabilidade da sua base de conhecimento, tornando viável a sua utilização no projeto de sistemas reais. / Genetic-Fuzzy Systems Generation and Simplification of a Knowledge Base proposes a methodology to develop a knowledge base for fuzzy systems through the utilization of evolutionary computational techniques. The evolved fuzzy systems are evaluated considering two distinct criteria: performance and interpretability. Another Fuzzy Logic-based methodology for multiobjective problem analysis was also developed in this work and incorporated in GAs fitness evaluation process. The aforementioned systems were analyzed through computational simulations, and the results were compared to those obtained through other methods, in some applications. The proposed methodology demonstrated that the evolved fuzzy systems are capable of not only good performance, but also good interpretation of their knowledge base, thus showing that they can be effectively used in real world projects.

Orientadores: Christiano Lyra Filho, Celso Cavellucci / Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia Elétrica e de Computação / Made available in DSpace on 2018-08-18T02:44:14Z (GMT). No. of bitstreams: 1 VizcainoGonzalez_JoseFederico_D.pdf: 9526841 bytes, checksum: 9fc3b73df592526f56693f9b329fc70c (MD5) Previous issue date: 2011 / Resumo: O trabalho desenvolve alternativas de otimização combinatória para a redução de perdas técnicas e melhoria das condições de operação de sistemas de distribuição de energia elétrica. Sua principal contribuição é na área de redução dos fluxos de reativos através da instalação e controle de bancos de capacitores. Duas alternativas de otimização são desenvolvidas. A primeira, propõe um algoritmo genético híbrido com buscas locais nas representações fenotípicas e genotípicas das soluções. A segunda alternativa utiliza conceitos de programação dinâmica no projeto de algoritmos que encontram soluções ótimas globais para o problema de localização, dimensionamento e controle de capacitores. Outro algoritmo genético híbrido, para a instalação de reguladores de tensão, complementa a possibilidade de melhoria nos perfis de tensão proporcionada pelos capacitores. Os algoritmos baseados em programação dinâmica são de complexidade polinomial; adicionalmente, suas complexidades são lineares para instâncias reais. As características desses algoritmos estabelecem novas referências para a área de localização e controle de capacitores em sistemas de distribuição de energia elétrica, hoje povoada por métodos heurísticos / Abstract: This work develops combinatorial optimization alternatives for technical loss reduction and improvements on operational conditions of power distribution networks. Its main contribution is in the area of loss reduction by decreasing reactive flows, through allocation and control of shunt capacitors banks. Two optimization strategies are proposed. The first one develops a hybrid genetic algorithm with local searches in both genotypical and fenotypical representations of solutions. The second alternative uses dynamic programming concepts in the design of algorithms that unveil global optimal solutions for capacitor location, sizing and control. Another hybrid genetic algorithm for allocation of voltage regulators complements the improvement in voltage profiles obtained with the allocation of capacitors. The algorithms based on dynamic programming concepts have polynomial-time complexity; further, they have linear-time complexity for practical applications. Therefore, these algorithms establish a new reference for the area of shunt capacitors allocation and control on power distribution systems, which is today populated by heuristic methods / Doutorado / Energia Eletrica / Doutor em Engenharia Elétrica

Orientadores: Akebo Yamakami, Jose Luis Verdegay Galdeano / Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia Eletrica e de Computação / Made available in DSpace on 2018-08-14T06:44:19Z (GMT). No. of bitstreams: 1 Silva_RicardoCoelho_D.pdf: 1144878 bytes, checksum: 38379443fb6892fd6eda74c55c3b99dc (MD5) Previous issue date: 2009 / Resumo: O objetivo deste trabalho é buscar, estudar e estabelecer as condições de otimali-dade para resolver problemas de programação multi-objetivo irrestritos e restritos em um ambiente impreciso. Essas imprecisões estão presentes nos problemas da vida real e existem muitas formas de tratá-las, mas nesse trabalho será usado a teoria de conjuntos nebulosos. Utilizando como base a otimização nebulosa, foram desenvolvidas duas abordagens para resolver problemas multi-objetivo nebulosos. A primeira abordagem transforma um problema nebuloso em um problema clássico paramétrico com um número maior de funções objetivo, a qual é chamada de paramétrica. A segunda abordagem, chamada de possibilística, usa a teoria de possibilidade como um índice de comparação entre números nebulosos com a finalidade de garantir condições de otimalidade em um ambiente nebuloso. Alguns exemplos numéricos são resolvidos usando um algoritmo genético chamado NSGA-II elitista, com algumas modificações para a comparação de números nebulosos, e depois feita uma análise dos resultados encontrados por ambos os enfoques. / Abstract: The main goal of this work is to search, study and present the optimality conditions to solve the unconstraint and constraint multiobjetive programming problems in imprecise environment. These imprécisions can be found in the real-world optimization problems and there are utmost ways for dealing with them, but in this work will be used the theory of fuzzy sets. Using as a basis the fuzzy optimization, two approaches were developed to solve fuzzy multiobjective problems. The first approach transforms a fuzzy problem into a parametric classic multiobjective programming problem with many more objective functions, which is called parametric approach. The second one, called possibilistic, uses the possibility theory as a comparison index between two fuzzy numbers in order to ensure optimality conditions in a fuzzy environment. Some numerical examples are solved by using a genetic algorithm called elitist NSGA-II with some modifications to compare fuzzy numbers, and then the results obtained with both approaches are analysed. / Doutorado / Automação / Doutor em Engenharia Elétrica

Orientadores: Eliane Martins, Fabiano Luis de Sousa / Dissertação (mestrado) - Universidade Estadual de Campinas, Instituto de Computação / Made available in DSpace on 2018-08-08T18:05:18Z (GMT). No. of bitstreams: 1 Abreu_BrunoTeixeirade_M.pdf: 1214826 bytes, checksum: 85afe48e3c8196abe877dc2ea2efa102 (MD5) Previous issue date: 2006 / Resumo: O teste é uma atividade importante do processo de desenvolvimento de software, e automatizar a geração de dados de teste contribui para a redução dos esforços de tempo e recursos. Recentemente foi mostrado que os algoritmos evolutivos, tal como os Algoritmos Genéticos (AGs), são ferramentas valiosas para a geração de dados. Este trabalho avalia pela primeira vez o desempenho de um algoritmo evolutivo proposto recentemente, a Otimização Extrema Generalizada (em inglês, Generalized Extremal Optimization, GEO), na geração de dados de teste para cobrir um subconjunto de caminhos de um programa, com ou sem loops. Sete programas muito conhecidos e utilizados como benchmarks por outros autores foram escolhidos como estudos de caso, e o desempenho do GEO foi comparado com o de um AG e o Random-Test (RT). Uma aplicação real do Instituto Nacional de Pesquisas Espaciais (INPE) também foi testada para validar a pesquisa, e as comparações de desempenho incluíram uma variação do AG utilizado nos benchmarks. Para os benchmarks e a aplicação real, o uso do GEO exigiu muito menos esforço computacional para gerar os dados do que os AGs, e a cobertura média de caminhos obtida por ele foi muito semelhante à dos AGs. Além disso, o GEO também exigiu muito menos esforço computacional no ajuste interno de parâmetros do que os AGs. Estes resultados indicam que o GEO é uma opção muito atraente a ser utilizada na geração de dados de teste / Abstract: Software testing is an important activity of the software development process and automating test data generation contributes to reduce cost and time efforts. It has recently been shown that evolutionary algorithms, such as the Genetic Algorithms (GAs), are valuable tools for test data generation. This work assesses for the first time the performance of a recently proposed evolutionary algorithm, the Generalized Extremal Optimization (GEO), on generating test data to cover a subset of paths of a program, with or without loops. Seven well known benchmark programs were used as study cases, and the performance of GEO was compared to the one of a GA and Random-Test (RT). A real application from Instituto Nacional de Pesquisas Espaciais (INPE) was also tested in order to validate the research, and the performance comparisons included one variation of the GA used in the benchmarks. For the benchmark programs and the real application, using GEO required much less computational effort to generate test data than using the GAs, and GEO¿s average coverage was very similar to GA¿s. Besides this, it also required much less computational effort on internal parameter setting than the GAs. These results indicate that GEO is a very attractive option to be used for test data generation / Mestrado / Mestre em Ciência da Computação

This master thesis describes the design and the implementation of the algorithm solving the domain- independent partial order simple task network planning problem using the tree-based genetic programming. The work contains comparison of several possible approaches to the problem --- it compares different representations, ways of evaluation and approaches to the partial ordering. It defines heuristics to improve the efficiency of the algorithm, including the distance heuristic, the local search and the individual equivalency. The implementation was tested on several experiments to show the abilities, strengths and weaknesses of the algorithm. Powered by TCPDF (www.tcpdf.org)

Optimization problems are vital in physical sciences, commercial and finance matters. In a nutshell, almost everyone is the stake-holder in certain optimization problems aiming at minimizing the cost of production and losses of system, and also maximizing the profit. In control systems, the optimal configuration problems are essential that have been solved by various newly developed methods. The literature is exhaustively explored for an appropriate optimization method to solve such kind of problems. Particle Swarm Optimization is found to be one of the best among several optimization methods by analysing the experimental results. Two novel PSO variants are introduced in this thesis. The first one is named as N State Markov Jumping Particle Swarm Optimization, which is based on the stochastic technique and Markov chain in updating the particle velocity. We have named the second variant as N State Switching Particle Swarm Optimization, which is based on the evolutionary factor information for updating the velocity. The proposed algorithms are then applied to some widely used mathematical benchmark functions. The statistical results of 30 independent trails illustrate the robustness and accuracy of the proposed algorithms for most of the benchmark functions. The better results in terms of mean minimum evaluation errors and the shortest computation time are illustrated. In order to verify the satisfactory performance and robustness of the proposed algorithms, we have further formulated some basic applications in power system operations. The first application is about the static Economic Load Dispatch and the second application is on the Dynamic Economic Load Dispatch. These are highly complex and non-linear problems of power system operations consisting of various systems and generator constraints. Basically, in the static Economic Load Dispatch, a single load is considered for calculating the cost function. In contrast, the Dynamic Economic Load Dispatch changes the load demand for the cost function dynamically with time. In such a challenging and complex environment the proposed algorithms can be applied. The empirical results obtained by applying both of the proposed methods have substantiated their adaptability and robustness into the real-world environment. It is shown in the numerical results that the proposed algorithms are robust and accurate as compared to the other algorithms. The proposed algorithms have produced consistent best values for their objectives, where satisfying all constraints with zero penalty.

Games have been investigated as computationally complex problems since the inception of artificial intelligence in the 1950’s. Originally, search-based techniques were applied to create a competent (and sometimes even expert) game player. The search-based techniques, such as game trees, made use of human-defined knowledge to evaluate the current game state and recommend the best move to make next. Recent research has shown that neural networks can be evolved as game state evaluators, thereby removing the human intelligence factor completely. This study builds on the initial research that made use of evolutionary programming to evolve neural networks in the game learning domain. Particle Swarm Optimisation (PSO) is applied inside a coevolutionary training environment to evolve the weights of the neural network. The training technique is applied to both the zero sum and non-zero sum game domains, with specific application to Tic-Tac-Toe, Checkers and the Iterated Prisoners Dilemma (IPD). The influence of the various PSO parameters on playing performance are experimentally examined, and the overall performance of three different neighbourhood information sharing structures compared. A new coevolutionary scoring scheme and particle dispersement operator are defined, inspired by Formula One Grand Prix racing. Finally, the PSO is applied in three novel ways to evolve strategies for the IPD – the first application of its kind in the PSO field. The PSO-based coevolutionary learning technique described and examined in this study shows promise in evolving intelligent evaluators for the aforementioned games, and further study will be conducted to analyse its scalability to larger search spaces and games of varying complexity. / Dissertation (MSc)--University of Pretoria, 2005. / Computer Science / unrestricted

Orientador: Christiano Lyra Filho / Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Engenharia Eletrica e de Computação / Made available in DSpace on 2018-08-04T02:20:44Z (GMT). No. of bitstreams: 1 VizcainoGonzalez_JoseFederico_M.pdf: 972282 bytes, checksum: 652889ccc97d9102a9333902b4169885 (MD5) Previous issue date: 2003 / Resumo: As perdas técnicas de energia nas redes primarias de distribuição são decorrentes das resistências elétricas nas linhas. Pela natureza indutiva de algumas cargas e reatâncias das linhas, parte da energia dissipada é devida aos fluxos de potências reativas entre a subestação e os pontos de carga. Capacitores instalados próximo às cargas podem fornecer energia reativa local, diminuindo as perdas de energia na rede. Este trabalho apresenta inovações conceituais e de implementação que permitem o resgate da abordagem por programação dinâmica para a solução do problema de instalação e dimensionamento de capacitores fixos em redes de distribuição, para perfis de cargas fixos ou variáveis. O trabalho também aborda o problema de controle de capacitores chaveados, propondo duas novas abordagens. A primeira abordagem desenvolve uma versão de sistemas complexos adaptativos, também chamados sistemas classificadores. A segunda abordagem corresponde a uma especialização dos conceitos de programação dinâmica esenvolvidos para o problema de instalação de capacitores. Estudos de casos em redes reais de grande porte ilustram as possibilidades das metodologias desenvolvidas / Abstract: Technical energy losses in primary distribution networks are due to electrical resistances in lines. Due to reactance of power lines and inductive nature of some loads, part of the energy dissipated is due to reactive power that travels back and forth in lines, all the way from power sources to load points. Capacitors installed near load points can provide local complementary reactive power that decrease losses. This work presents conceptual and implementation innovations that allows to rescue the dynamic programming approach for the capacitors sizing and allocation problem in distribution networks, for fixes and variables loads. The work also presents two approaches to the capacitors control problem. The first approach is the development of a complex adaptive system (a classifier systems). The second approach to the capacitor control problem is a specialization of dynamic programming concepts, developed for the capacitors placement problem. Case studies in large real networks illustrate the possibilities of the developed methodologies / Mestrado / Automação / Mestre em Engenharia Elétrica

Reservoir Computing (RC) is an emerging machine learning paradigm where a fixed kernel, built from a randomly connected "reservoir" with sufficiently rich dynamics, is capable of expanding the problem space in a non-linear fashion to a higher dimensional feature space. These features can then be interpreted by a linear readout layer that is trained by a gradient descent method. In comparison to traditional neural networks, only the output layer needs to be trained, which leads to a significant computational advantage. In addition, the short term memory of the reservoir dynamics has the ability to transform a complex temporal input state space to a simple non-temporal representation. Adaptive real-time systems are multi-stage decision problems that can be used to train an agent to achieve a preset goal by performing an optimal action at each timestep. In such problems, the agent learns through continuous interactions with its environment. Conventional techniques to solving such problems become computationally expensive or may not converge if the state-space being considered is large, partially observable, or if short term memory is required in optimal decision making. The objective of this thesis is to use reservoir computers to solve such goal-driven tasks, where no error signal can be readily calculated to apply gradient descent methodologies. To address this challenge, we propose a novel reinforcement learning approach in combination with reservoir computers built from simple Boolean components. Such reservoirs are of interest because they have the potential to be fabricated by self-assembly techniques. We evaluate the performance of our approach in both Markovian and non-Markovian environments. We compare the performance of an agent trained through traditional Q-Learning. We find that the reservoir-based agent performs successfully in these problem contexts and even performs marginally better than Q-Learning agents in certain cases. Our proposed approach allows to retain the advantage of traditional parameterized dynamic systems in successfully modeling embedded state-space representations while eliminating the complexity involved in training traditional neural networks. To the best of our knowledge, our method of training a reservoir readout layer through an on-policy boot-strapping approach is unique in the field of random Boolean network reservoirs.

The high cost of processor fabrication plants and approaching physical limits have started a new wave research in alternative computing paradigms. As an alternative to the top-down manufactured silicon-based computers, research in computing using natural and physical system directly has recently gained a great deal of interest. A branch of this research promotes the idea that any physical system with sufficiently complex dynamics is able to perform computation. The power of networks in representing complex interactions between many parts make them a suitable choice for modeling physical systems. Many studies used networks with a homogeneous structure to describe the computational circuits. However physical systems are inherently heterogeneous. We aim to study the effect of heterogeneity in the dynamics of physical systems that pertains to information processing. Two particularly well-studied network models that represent information processing in a wide range of physical systems are Random Boolean Networks (RBN), that are used to model gene interactions, and Liquid State Machines (LSM), that are used to model brain-like networks. In this thesis, we study the effects of function heterogeneity, in-degree heterogeneity, and interconnect irregularity on the dynamics and the performance of RBN and LSM. First, we introduce the model parameters to characterize the heterogeneity of components in RBN and LSM networks. We then quantify the effects of heterogeneity on the network dynamics. For the three heterogeneity aspects that we studied, we found that the effect of heterogeneity on RBN and LSM are very different. We find that in LSM the in-degree heterogeneity decreases the chaoticity in the network, whereas it increases chaoticity in RBN. For interconnect irregularity, heterogeneity decreases the chaoticity in LSM while its effects on RBN the dynamics depends on the connectivity. For {K} < 2, heterogeneity in the interconnect will increase the chaoticity in the dynamics and for {K} > 2 it decreases the chaoticity. We find that function heterogeneity has virtually no effect on the LSM dynamics. In RBN however, function heterogeneity actually makes the dynamics predictable as a function of connectivity and heterogeneity in the network structure. We hypothesize that node heterogeneity in RBN may help signal processing because of the variety of signal decomposition by different nodes.