| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 11.04 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
Atualmente, o mundo enfrenta um processo de transição energética, de uma matriz
focada em combustíveis fósseis para uma matriz baseada em fontes renováveis, o qual se
impôs face à necessidade de proteger o planeta da emissão de gases com efeito de estufa,
em parte, responsáveis pelo aquecimento global. Na verdade, a queima de demasiados
combustíveis fósseis, o abate de demasiadas árvores e a emissão de demasiados gases
com efeito de estufa está a aumentar a temperatura do planeta, o que é uma ameaça para
a sua sustentabilidade. Embora as sociedades tenham assistido a outras transições energéticas, no passado, nunca o planeta esteve tanto em risco. A atual transição energética,
para além de ter impacto no clima terá forte influência na economia e na forma como a
sociedade está organizada. Como prova disso, temos a atual guerra desencadeada pela
Rússia na Ucrânia que ao congestionar o fornecimento de gás natural à Europa mostrou
a necessidade, urgente, de reduzir a dependência energética entre os diferentes países.
Portanto, para além de permitirem reduzir a pegada de carbono, é aqui que as fontes renováveis como a energia solar fotovoltaica têm vindo a ter uma contribuição significativa,
nos últimos anos, e que necessariamente tem de continuar a crescer para se alcançar uma
sociedade com zero ou baixas emissões de carbono.
Em resposta a esta necessidade energética a capacidade fotovoltaica instalada mundialmente tem mostrado um crescimento significativo, o qual por sua vez leva à necessidade
de sistemas fotovoltaicos eficientes e confiáveis. Deste modo, insere-se aqui o tema da
presente dissertação, a qual tem como objetivo propor um novo algoritmo de procura do
ponto de máxima potência em sistemas fotovoltaicos. Para extrair a máxima potência disponível num sistema fotovoltaico o novo algoritmo de procura do ponto de máxima potência proposto combina as potencialidades do Particle Swarm Optimization Estocástico e
do Particle Swarm Optimization Determinístico. Além disso, o algoritmo proposto utiliza
uma Radial Basis Function Neural Network para construir um modelo aproximado das
curvas caraterísticas corrente-tensão e potência-tensão. Para isso, numa primeira fase,
o Particle Swarm Optimization Estocástico obriga as partículas a percorrer o espaço de
procura de forma aleatória, fornecendo a informação proveniente da sua execução à Radial Basis Function Neural Network. Posteriormente, são criados modelos aproximados
das curvas caraterísticas corrente-tensão e potência-tensão, permitindo identificar as regiões do espaço de procura promissoras. Uma vez identificadas e definidas essas regiões,
as partículas são reposicionadas, e é acionado o Particle Swarm Optimization Determinístico para evitar um comportamento das partículas divergente e repetitivo. Após a convergência do Particle Swarm Optimization Determinístico é realizado um processo de
refinamento das curvas caraterísticas corrente-tensão e potência-tensão.
Para testar e validar o algoritmo proposto foram usados quatro casos de estudo em
ambiente de simulação e dois casos de estudo em ambiente experimental. O desempenho
do algoritmo proposto foi comparado com o método convencional de procura do ponto de máxima potência Perturba e Observa, bem como, com diferentes métodos de procura
do ponto de máxima potência meta-heurísticos, nomeadamente o Differential Evolution,
o Grey Wolf Optimizer e o Particle Swarm Optimization clássico. Os resultados mostraram que o algoritmo proposto superou os métodos considerados ao encontrar o ponto
de máxima potência global em termos de taxa de sucesso, tempo de procura e eficiência, além disso, apresentou menor número de oscilações respondendo com robustez às
rápidas transições de irradiância perante condições de sombreamento parcial complexas.
Currently, the world is facing an energy transition process, from a matrix focused on fossil fuels to a matrix based on renewable sources, which has imposed itself in the face of the need to protect the planet from the emission of greenhouse gases, partly responsible by global warming. In fact, burning too many fossil fuels, cutting too many trees and emitting too many greenhouse gases is increasing the temperature of the planet, which is a threat to its sustainability. While societies have seen other energy transitions in the past, the planet has never been more at risk. The current energy transition, in addition to having an impact on the climate, will have a strong influence on the economy and the way society is organized. As proof of this, we have the current war unleashed by Russia in Ukraine which, by congesting the supply of natural gas to Europe, showed the urgent need to reduce energy dependence between different countries. Therefore, in addition to reducing the carbon footprint, it is here that renewable sources such as photovoltaic solar energy have been making a significant contribution in recent years, and which necessarily must continue to grow to achieve a society with zero or low carbon emissions. In response to this energy need, the photovoltaic capacity installed worldwide has shown significant growth, which in turn leads to the need for efficient and reliable photovoltaic systems. Thus, the theme of the present dissertation is inserted here, which aims to propose a new algorithm to search for the maximum power point in photovoltaic systems. To extract the maximum available power in a photovoltaic system, the new proposed maximum power point search algorithm combines the potential of Stochastic Particle Swarm Optimization and Deterministic Particle Swarm Optimization. Furthermore, the proposed algorithm uses a Radial Basis Function Neural Network to build an approximate model of the current-voltage and power-voltage characteristic curves. For that, in a first phase, the Stochastic Particle Swarm Optimization forces the particles to traverse the search space in a random way, supplying the information from its execution to the Radial Basis Function Neural Network. Subsequently, approximate models of the current-voltage and power-voltage characteristic curves are created, allowing the identification of promising regions of the search space. Subsequently, approximate models of the current-voltage and power-voltage characteristic curves are created, allowing the identification of promising regions of the search space. Once these regions are identified and defined, the particles are repositioned, and the Deterministic Particle Swarm Optimization is activated to avoid a divergent and repetitive behavior of the particles. After the convergence of the Deterministic Particle Swarm Optimization, a refinement process of the current-voltage and power-voltage characteristic curves is carried out. To test and validate the proposed algorithm, four case studies were used in a simulation environment and two case studies in an experimental environment. The performance of the proposed algorithm was compared with the conventional maximum power point tracking method (Perturb and Observe) as well as with different meta-heuristic maximum power point tracking methods, namely Differential Evolution, Gray Wolf Optimizer and Classic Particle Swarm Optimization. The results showed that the proposed algorithm outperformed the considered methods in finding the global maximum power point in terms of success rate, search time and efficiency, in addition, it presented a lower number of oscillations responding with robustness to fast irradiance transitions under complex partial shading conditions.
Currently, the world is facing an energy transition process, from a matrix focused on fossil fuels to a matrix based on renewable sources, which has imposed itself in the face of the need to protect the planet from the emission of greenhouse gases, partly responsible by global warming. In fact, burning too many fossil fuels, cutting too many trees and emitting too many greenhouse gases is increasing the temperature of the planet, which is a threat to its sustainability. While societies have seen other energy transitions in the past, the planet has never been more at risk. The current energy transition, in addition to having an impact on the climate, will have a strong influence on the economy and the way society is organized. As proof of this, we have the current war unleashed by Russia in Ukraine which, by congesting the supply of natural gas to Europe, showed the urgent need to reduce energy dependence between different countries. Therefore, in addition to reducing the carbon footprint, it is here that renewable sources such as photovoltaic solar energy have been making a significant contribution in recent years, and which necessarily must continue to grow to achieve a society with zero or low carbon emissions. In response to this energy need, the photovoltaic capacity installed worldwide has shown significant growth, which in turn leads to the need for efficient and reliable photovoltaic systems. Thus, the theme of the present dissertation is inserted here, which aims to propose a new algorithm to search for the maximum power point in photovoltaic systems. To extract the maximum available power in a photovoltaic system, the new proposed maximum power point search algorithm combines the potential of Stochastic Particle Swarm Optimization and Deterministic Particle Swarm Optimization. Furthermore, the proposed algorithm uses a Radial Basis Function Neural Network to build an approximate model of the current-voltage and power-voltage characteristic curves. For that, in a first phase, the Stochastic Particle Swarm Optimization forces the particles to traverse the search space in a random way, supplying the information from its execution to the Radial Basis Function Neural Network. Subsequently, approximate models of the current-voltage and power-voltage characteristic curves are created, allowing the identification of promising regions of the search space. Subsequently, approximate models of the current-voltage and power-voltage characteristic curves are created, allowing the identification of promising regions of the search space. Once these regions are identified and defined, the particles are repositioned, and the Deterministic Particle Swarm Optimization is activated to avoid a divergent and repetitive behavior of the particles. After the convergence of the Deterministic Particle Swarm Optimization, a refinement process of the current-voltage and power-voltage characteristic curves is carried out. To test and validate the proposed algorithm, four case studies were used in a simulation environment and two case studies in an experimental environment. The performance of the proposed algorithm was compared with the conventional maximum power point tracking method (Perturb and Observe) as well as with different meta-heuristic maximum power point tracking methods, namely Differential Evolution, Gray Wolf Optimizer and Classic Particle Swarm Optimization. The results showed that the proposed algorithm outperformed the considered methods in finding the global maximum power point in terms of success rate, search time and efficiency, in addition, it presented a lower number of oscillations responding with robustness to fast irradiance transitions under complex partial shading conditions.
Description
Keywords
Inteligência Artificial Particle Swarm Optimization Procura do Ponto de Máxima Potência Radial Basis Function Neural Network Sistema Fotovoltaico