Loading...
Publication
Aplicação de algoritmos da inteligência artificial à otimização de painéis fotovoltaicos usando MPPT
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 22.66 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
Esta dissertação aborda a aplicação de algoritmos de inteligência artificial na otimização
de painéis fotovoltaicos através de algoritmos MPPT. Desenvolveram-se modelos
de simulação no MATLAB/Simulink para sistemas fotovoltaicos, integrando várias
abordagens de monitorização do ponto de máxima potência, com foco principal nas
redes neuronais e Deep Reinforcement Learning, com o objetivo de maximizar a
eficiência energética em diferentes condições de irradiância e temperatura.
O sistema proposto incluiu o dimensionamento de um conversor DC/DC buck e a
validação em diferentes cenários meteorológicos, com análises quantitativas baseadas
em métricas de avaliação.
Foram implementados e comparados diferentes algoritmos MPPT, incluindo
métodos convencionais (Perturb and Observe, Incremental Conductance), lógica
difusa (FLC), algoritmos genéticos (GA) e técnicas baseadas em inteligência artificial.
Os resultados obtidos demonstraram que os métodos baseados em inteligência
artificial, em particular as redes neuronais e o agente Deep Q-Network, apresentaram
melhor desempenho em termos de eficiência média e tempo de convergência, especialmente
em cenários com condições de sombra parcial (Partial Shading Conditions
(PSC)). As conclusões reforçam a viabilidade da aplicação destas abordagens em
sistemas fotovoltaicos, contribuindo para uma produção energética mais eficiente
perante condições não uniformes.
This dissertation explores the application of artificial intelligence algorithms to the optimization of photovoltaic panels through MPPT algorithms. Simulation models were developed in MATLAB/Simulink for photovoltaic systems, integrating various approaches to maximum power point tracking, with a primary focus on neural networks and Deep Reinforcement Learning, aiming to maximize energy efficiency under different irradiance and temperature conditions. The proposed system included the design of a DC/DC buck converter and validation under different weather scenarios, with quantitative analyses based on evaluation metrics. Different MPPT algorithms were implemented and compared, including conventional methods (Perturb and Observe, Incremental Conductance), fuzzy logic control (FLC), genetic algorithms (GA), and techniques based on artificial intelligence. The results demonstrated that artificial intelligence-based methods, particularly neural networks and the Deep Q-Network agent, achieved superior performance in terms of average efficiency and convergence time, especially in scenarios with partial shading conditions (PSC). The conclusions support the feasibility of applying these approaches to photovoltaic systems, contributing to more efficient energy production under non-uniform conditions.
This dissertation explores the application of artificial intelligence algorithms to the optimization of photovoltaic panels through MPPT algorithms. Simulation models were developed in MATLAB/Simulink for photovoltaic systems, integrating various approaches to maximum power point tracking, with a primary focus on neural networks and Deep Reinforcement Learning, aiming to maximize energy efficiency under different irradiance and temperature conditions. The proposed system included the design of a DC/DC buck converter and validation under different weather scenarios, with quantitative analyses based on evaluation metrics. Different MPPT algorithms were implemented and compared, including conventional methods (Perturb and Observe, Incremental Conductance), fuzzy logic control (FLC), genetic algorithms (GA), and techniques based on artificial intelligence. The results demonstrated that artificial intelligence-based methods, particularly neural networks and the Deep Q-Network agent, achieved superior performance in terms of average efficiency and convergence time, especially in scenarios with partial shading conditions (PSC). The conclusions support the feasibility of applying these approaches to photovoltaic systems, contributing to more efficient energy production under non-uniform conditions.
Description
Keywords
Photovoltaic System Maximum Power Point (MPP) Maximum Power Point Tracking (MPPT) Buck Converter Energy Efficiency Artificial Intelligence Neural Networks Deep Reinforcement Learning Sistema fotovoltaico Conversor Buck Eficiência energética Inteligência artificial Redes neuronais
Pedagogical Context
Citation
Publisher
CC License
Without CC licence