Doctor of Philosophy / Department of Electrical and Computer Engineering / Shelli K. Starrett / Monitoring and control of modern power systems have become very complex tasks due to the interconnection of power grids. These large-scale power grids confront system operators with a huge set of system inputs and control parameters. This work develops and compares intelligent systems-based algorithms which may be considered by power system operators or planners to help manage, process, and evaluate large amounts of data due to varying conditions within the system. The methods can be used to provide assistance in making operational control and planning decisions for the system in a timely manner. The effectiveness of the proposed algorithms is tested and validated on four different power systems. First, Artificial Neural Network (ANN) models are developed and compared for two different voltage collapse indices and utilizing two different-sized sets of inputs. The ANNs monitor and evaluate the voltage profile of a system and generate intelligent conclusions regarding the status of the system from a voltage stability perspective. A feature reduction technique, based on the analysis of generated data, is used to decrease the number of inputs fed to the ANN, decreasing the number of physical quantities that need to be measured. The major contribution of this work is the development of four different algorithms to control the VAR resources in a system. Four different objectives were also considered in this part of the work, namely: minimization of the number of control changes needed, minimization of the system power losses, minimization of the system's voltage deviations, and consideration of the computational time required. Each of the algorithms is iterative in nature and is designed to take advantage of a method of decoupling the load flow Jacobian matrix to decrease the time needed per iteration. The methods use sensitivity information derived from the load flow Jacobian and augmented with equations relating the desired control and dependent variables. The heuristic-sensitivity based method is compared to two GA-based methods using two different objective functions. In addition, a FL algorithm is added to the heuristic-sensitivity algorithm and compared to a PS-based algorithm. The last part of this dissertation presents the use of one of the GA-based algorithms to identify the size of shunt capacitor necessary to enhance the voltage profile of a system. A method is presented for utilizing contingency cases with this algorithm to determine required capacitor size.

Installation of reactive compensators is widely used for improving power system voltage stability. Reactive compensation also improves the system loading margin resulting in more stable and reliable operation. The improvements in system performance are highly dependent on the location where the reactive compensation is placed in the system. This paper compares three load flow analysis methods - PV curve analysis, QV sensitivity analysis, and Continuation Load Flow - in identifying system weak buses for placing reactive compensation. The methods are applied to three IEEE test systems, including modified IEEE 14-bus system, IEEE 30-bus system, and IEEE 57-bus system. Locations of reactive compensation and corresponding improvements in loading margin and voltages in each test system obtained by the three methods are compared. The author also analyzes the test systems to locate the optimal placement of reactive compensation that yields the maximum loading margin. The results when compared with brute force placement of reactive compensation show the relationship between effectiveness of the three methods and topology of the test systems.

Made available in DSpace on 2014-06-11T19:30:50Z (GMT). No. of bitstreams: 0 Previous issue date: 2009-03-13Bitstream added on 2014-06-13T20:40:35Z : No. of bitstreams: 1 isoda_ly_dr_ilha.pdf: 519614 bytes, checksum: 8efa9d6eaa776e3e613a4da6439527c1 (MD5) / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / Nesta tese apresenta-se uma proposta para análise da estabilidade estática de tensão de sistemas de energia elétrica utilizando uma rede neural baseada na arquitetura ART (Adaptive Resonance Theory), designada rede neural ARTMAP Fuzzy. As redes neurais ARTdescendentes apresentam as características de estabilidade e plasticidade, as quais são propriedades imprescindíveis para a realização do treinamento e execução da análise de forma rápida e confiável. A versão ARTMAP Fuzzy é uma rede neural supervisionada, ou seja, a extração do conhecimento se processa por estímulos de entrada e de saída. O problema da análise de estabilidade de tensão é formulado considerando-se o estímulo de entrada composto pelas potências ativa e reativa nodais. O estímulo de saída é adotado como sendo a margem de segurança, a qual representa a “distância” entre o ponto de operação do sistema e a fronteira da estabilidade estática de tensão. Esta margem de segurança é calculada, via análise de sensibilidade e álgebra matricial de Kronecker, a partir da função determinante da matriz jacobiana relativa ao problema do fluxo de potência de Newton-Raphson. A operacionalidade das redes neurais é constituída por três fases principais: treinamento (ou aprendizado), análise e treinamento continuado. A fase de treinamento requer uma grande quantidade de processamento, enquanto que a fase de análise é realizada, efetivamente, sem esforço computacional. Esta é, por conseguinte, a principal justificativa para o uso das redes neurais para a resolução de problemas complexos que exigem soluções rápidas, como é o caso de aplicações em tempo real. Na fase de treinamento, o perfil de geração e de carga do sistema elétrico é gerado empregando-se uma distribuição aleatória (ou pseudo-aleatória) e a respectiva saída (margem de segurança) calculada via execução... / This work develops a methodology to effectuate static voltage stability of electrical power systems by neural network. The neural network used is based on the ART (Adaptive Resonance Theory) architecture, named ARTMAP Fuzzy neural network. The ART descendent neural networks present the characteristics of stability and plasticity, which are important properties to execute the training and the analysis fast and reliable. The ARTMAP Fuzzy version is a supervised neural network, i.e. the extraction of the knowledge is processed by input/output stimulus. The voltage stability analysis problem is formulated considering the input stimulus composed by the active and reactive nodal power. The output stimulus is adopted as the security margin, which represents the distance with the operation point and the static voltage stability frontier. The security margin is calculated by sensitivity analysis and Kronecker algebra from the determinant function of the Jacobian matrix related to the power flow problem by Newton-Raphson. Neural Network operation is constituted by three principal phase: training (or learning), analysis and continuous training. The training phase needs great processing effort, while the analysis is effectuated without computational effort. This is the principal advantage to use neural networks to solve complex problems that need fast solutions as the real time applications. On the training phase, the generation and load profile is generated using a random (or pseudo random) distribution and the respective output (security margin) is calculated by executing a conventional power-flow with adequate adaptations. The procedure proposed is independent of how is defined the generation dispatch and how the system load evolves. This is a more realistic approach, when compared to the most of the proposals found on the specialized literature that considers the load increasing linearly... (Complete abstract click electronic access below)

Orientador: Carlos Roberto Minussi / Banca: Francisco Villarreal Alvarado / Banca: Jozué Vieira Filho / Banca: Osvaldo Ronald Saavedra Mendez / Banca: Eduardo Nobuhiro Asada / Resumo: Nesta tese apresenta-se uma proposta para análise da estabilidade estática de tensão de sistemas de energia elétrica utilizando uma rede neural baseada na arquitetura ART (Adaptive Resonance Theory), designada rede neural ARTMAP Fuzzy. As redes neurais ARTdescendentes apresentam as características de estabilidade e plasticidade, as quais são propriedades imprescindíveis para a realização do treinamento e execução da análise de forma rápida e confiável. A versão ARTMAP Fuzzy é uma rede neural supervisionada, ou seja, a extração do conhecimento se processa por estímulos de entrada e de saída. O problema da análise de estabilidade de tensão é formulado considerando-se o estímulo de entrada composto pelas potências ativa e reativa nodais. O estímulo de saída é adotado como sendo a margem de segurança, a qual representa a "distância" entre o ponto de operação do sistema e a fronteira da estabilidade estática de tensão. Esta margem de segurança é calculada, via análise de sensibilidade e álgebra matricial de Kronecker, a partir da função determinante da matriz jacobiana relativa ao problema do fluxo de potência de Newton-Raphson. A operacionalidade das redes neurais é constituída por três fases principais: treinamento (ou aprendizado), análise e treinamento continuado. A fase de treinamento requer uma grande quantidade de processamento, enquanto que a fase de análise é realizada, efetivamente, sem esforço computacional. Esta é, por conseguinte, a principal justificativa para o uso das redes neurais para a resolução de problemas complexos que exigem soluções rápidas, como é o caso de aplicações em tempo real. Na fase de treinamento, o perfil de geração e de carga do sistema elétrico é gerado empregando-se uma distribuição aleatória (ou pseudo-aleatória) e a respectiva saída (margem de segurança) calculada via execução... (Resumo completo, clicar acesso eletrônico abaixo) / Abstract: This work develops a methodology to effectuate static voltage stability of electrical power systems by neural network. The neural network used is based on the ART (Adaptive Resonance Theory) architecture, named ARTMAP Fuzzy neural network. The ART descendent neural networks present the characteristics of stability and plasticity, which are important properties to execute the training and the analysis fast and reliable. The ARTMAP Fuzzy version is a supervised neural network, i.e. the extraction of the knowledge is processed by input/output stimulus. The voltage stability analysis problem is formulated considering the input stimulus composed by the active and reactive nodal power. The output stimulus is adopted as the security margin, which represents the distance with the operation point and the static voltage stability frontier. The security margin is calculated by sensitivity analysis and Kronecker algebra from the determinant function of the Jacobian matrix related to the power flow problem by Newton-Raphson. Neural Network operation is constituted by three principal phase: training (or learning), analysis and continuous training. The training phase needs great processing effort, while the analysis is effectuated without computational effort. This is the principal advantage to use neural networks to solve complex problems that need fast solutions as the real time applications. On the training phase, the generation and load profile is generated using a random (or pseudo random) distribution and the respective output (security margin) is calculated by executing a conventional power-flow with adequate adaptations. The procedure proposed is independent of how is defined the generation dispatch and how the system load evolves. This is a more realistic approach, when compared to the most of the proposals found on the specialized literature that considers the load increasing linearly... (Complete abstract click electronic access below) / Doutor