Estudo de segmentação e movimentação de nuvens utilizando imageadores do céu em Natal-RN, costa do nordeste do Brasil
Data
2024-05-17
Autores
Pires, Wendy Mary da Silveira 
Orientadores
Martins, Fernando Ramos
Tipo
Dissertação de mestrado
Título da Revista
ISSN da Revista
Título de Volume
Resumo
O Sol é a principal fonte de energia para os processos físicos, químicos e biológicos que acontecem em nosso planeta. Com o desenvolvimento tecnológico e o aumento da demanda por energia, as fontes renováveis de energia, incluindo a energia solar, são uma alternativa viável sob o ponto de vista técnico, econômico e ambiental. No entanto, devido às flutuações diárias da radiação solar incidente na superfície, é difícil modelar e simular a produção de energia solar, principalmente em uma região equato-rial considerando o conjunto de fenômenos meteorológicos que atuam e podem con-tribuir para os processos radiativos na atmosfera. A nebulosidade é o principal fator modulador da disponibilidade do recurso solar na superfície dada a sua variabilidade temporal e espacial, de modo que a previsão da cobertura futura de nuvens é uma informação fundamental para simulação e operação dos sistemas fotovoltaicos. Este estudo buscou analisar o movimento de nuvens em imagens coletadas em Natal-RN, cidade costeira localizada na região equatorial do Nordeste do Brasil (NEB). As ima-gens do céu foram coletadas por imageador all-sky com lente olho de peixe (fisheye) em operação na estação solarimétrica integrante da rede SONDA (INPE). O método de calibração das imagens foi desenvolvido com base na percurso solar com o intuito de localizar a posição das nuvens em relação à posição de operação do imageador. Dois algoritmos de classificação de pixels com relação a presença de nebulosidade foram desenvolvidos, um com base na razão vermelho-azul (RBR) em cada pixel e o outro algoritmo utiliza uma biblioteca de céu claro para minimizar o impacto da pre-sença do Sol não ocluso por nuvens nas imagens coletadas. O método de Farnebäck foi implementado para simular o fluxo óptico das nuvens e estimar a posição futura do campo de nuvens para três horizontes de previsão. Os resultados tornam evidentes o desafio da modelagem de movimento de nuvens na costa equatorial NEB em razão das características climáticas regionais: ventos constantes e processos de formação/-dissipação de nuvens em escalas de tempo muito reduzidas. As métricas adotadas para avaliação das previsões realizadas foram RMSE, Viés e MAE. A previsão por persistência foi adotada como referência para comparação com a modelagem de fluxo ótico. Os resultados mostram que para horizontes de 1 minuto, a modelagem de Far-nebäck apresentou desvios do RMSE abaixo de 50%. Em contrapartida os resultados para 10 minutos foram inferiores ao observado em 5 minutos, com um RMSE má-ximo de 55%. Diante disso, a previsão para intervalos de 5 minutos foram as que apresentaram maiores problemas, tendo em vista, as mudanças rápidas das nebulosi-dades sobre as lentes da câmera. Em contrapartida, nos intervalos de 1 e 10 minutos Farnebäck apresentou resultados satisfatórios. As previsões de Farnebäck mostram pequena vantagem em relação ao método de persistência uma vez que os desvios de RMSE foram similares em ambos os métodos, mas valores de viés são reduzidos em condições meteorológicas específicas.
The Sun is the main source of energy for physical, chemical, and biological processes that occur on our planet. With technological development and the increase in energy demand, renewable energy sources, including solar energy, are a viable alternative from a technical, economic, and environmental point of view. However, due to daily fluctuations in solar radiation incident on the surface, it is difficult to model and simulate solar energy production, especially in an equatorial region, considering the set of meteorological characteristics that act and can contribute to radiative processes in the atmosphere. Cloudiness is the main factor modulating the availability of solar resources on the surface given its temporal and spatial variability, making the prediction of future cloud cover fundamental for the simulation and operation of photovoltaic systems.This study sought to analyze the movement of clouds in images collected in Natal-RN, a regional city located in the equatorial region of Northeast Brazil (NEB). The sky images were collected by an all-sky imager with a fisheye lens, operating at the integrated solarimetric station of the SONDA network (INPE). The image processing method was developed based on the solar path to locate the position of the clouds relative to the imager’s operating position.Two pixel classification algorithms regarding the presence of cloudiness were developed: one based on the red-blue ratio (RBR) at each pixel, and another algorithm that uses a clear sky library to minimize the impact of the Sun not being occluded by clouds in the collected images. The Farnebäck method was implemented to simulate the optical flow of clouds and estimate the future position of the cloud field for three forecast horizons.The results highlight the challenge of modeling cloud movement on the NEB equatorial coast due to regional climatic characteristics: constant winds and cloud formation/dissipation processes on very short time scales. The assessments adopted to evaluate the corrections made were RMSE, Bias, and MAE. The persistence forecast was adopted as a reference for comparison with optical flow modeling. The results show that for 1-minute horizons, Farnebäck presented RMSE deviations below 50%. In contrast, the results for 10 minutes were lower than those seen in 5 minutes, with a maximum RMSE of 55%. Given this, the forecast for 5-minute intervals presented the most serious problems due to rapid changes in cloud cover information on the camera lens. On the other hand, in the 1 and 10-minute intervals, Farnebäck showed superior results. Farnebäck’s analysis shows a small advantage over the persistence method since the RMSE deviations were similar in both methods, but the bias values were reduced under specific adverse conditions.
The Sun is the main source of energy for physical, chemical, and biological processes that occur on our planet. With technological development and the increase in energy demand, renewable energy sources, including solar energy, are a viable alternative from a technical, economic, and environmental point of view. However, due to daily fluctuations in solar radiation incident on the surface, it is difficult to model and simulate solar energy production, especially in an equatorial region, considering the set of meteorological characteristics that act and can contribute to radiative processes in the atmosphere. Cloudiness is the main factor modulating the availability of solar resources on the surface given its temporal and spatial variability, making the prediction of future cloud cover fundamental for the simulation and operation of photovoltaic systems.This study sought to analyze the movement of clouds in images collected in Natal-RN, a regional city located in the equatorial region of Northeast Brazil (NEB). The sky images were collected by an all-sky imager with a fisheye lens, operating at the integrated solarimetric station of the SONDA network (INPE). The image processing method was developed based on the solar path to locate the position of the clouds relative to the imager’s operating position.Two pixel classification algorithms regarding the presence of cloudiness were developed: one based on the red-blue ratio (RBR) at each pixel, and another algorithm that uses a clear sky library to minimize the impact of the Sun not being occluded by clouds in the collected images. The Farnebäck method was implemented to simulate the optical flow of clouds and estimate the future position of the cloud field for three forecast horizons.The results highlight the challenge of modeling cloud movement on the NEB equatorial coast due to regional climatic characteristics: constant winds and cloud formation/dissipation processes on very short time scales. The assessments adopted to evaluate the corrections made were RMSE, Bias, and MAE. The persistence forecast was adopted as a reference for comparison with optical flow modeling. The results show that for 1-minute horizons, Farnebäck presented RMSE deviations below 50%. In contrast, the results for 10 minutes were lower than those seen in 5 minutes, with a maximum RMSE of 55%. Given this, the forecast for 5-minute intervals presented the most serious problems due to rapid changes in cloud cover information on the camera lens. On the other hand, in the 1 and 10-minute intervals, Farnebäck showed superior results. Farnebäck’s analysis shows a small advantage over the persistence method since the RMSE deviations were similar in both methods, but the bias values were reduced under specific adverse conditions.
Descrição
Citação
PIRES, Wendy Mary da Silveira. Estudo de segmentação e movimentação de nuvens utilizando imageadores do céu em Natal-RN, costa do nordeste do Brasil. 2024. 100 f. Dissertação (Mestrado Interdisciplinar em Ciência e Tecnologia do Mar) - Universidade Federal de São Paulo, Instituto do Mar, Santos, 2024.