| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 10.08 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A crescente preocupação com a sustentabilidade energética tem conduzido a um aumento
significativo na adoção de fontes de energia renovável em detrimento das fontes não renováveis,
como os combustíveis fósseis. Este tópico tem ganhado cada vez mais destaque na atualidade,
refletindo um movimento global em direção a práticas mais ecológicas e eficientes no uso de
recursos energéticos. Os sistemas fotovoltaicos surgem como uma medida sustentável, permitindo
transformar energia solar em energia elétrica. Este tipo de sistemas é bastante vantajoso, para os
mais diversos setores, para além de permitir a redução de custos nas organizações, auxilia o
combate nas questões climáticas.
No presente trabalho, através do conjunto de dados fornecidos contendo várias métricas, foi
possível estudar uma instalação de painéis fotovoltaicos situada na zona do Porto. O principal
objetivo era criar um método de diagnóstico de falhas através de métodos de previsão em
linguagem Python. Para tal desenvolveu-se uma análise exploratória dos dados, com a finalidade
de obter insightsimportantes para a seleção do modelo a desenvolver. Desta forma, de acordo com
as conclusões da análise exploratória, foram selecionados dois modelos de previsão.
Inicialmente, foi implementado o modelo de previsão baseado em redes neuronais, no entanto,
apenas foi aplicado para uma variável. Desta forma, após uma nova pesquisa na literatura optou se por utilizar o modelo de previsão de séries temporais SARIMA, visto que, os dados apresentavam
sazonalidade, uma característica importante para a construção deste modelo. E desta forma, foi
possível também comparar os resultados dos dois modelos para uma variável em comum. O
principal desafio foi a robustez do conjunto de dados disponíveis. Apesar da granularidade dos
dados apresentados, apenas foram facultados dados relativos a dois anos, representando um
desafio substancial, uma vez que o modelo de previsão SARIMA depende da análise do
comportamento dos dados históricos para estabelecer previsões futuras confiáveis.
Os dois modelos permitiram desenvolver valores de previsões futuros sobre diversas variáveis
relevantes para a identificação de potenciais falhas no sistema fotovoltaico. Os resultados das
previsões indicam que, para o próximo ano, os valores das variáveis devem manter-se semelhantes
aos dados históricos e, portanto, não são esperadas falhas. Estes resultados eram esperados, uma
vez que os fornecedores garantem um alto desempenho no início da vida útil do sistema.
Os sistemas fotovoltaicos representam uma solução definitiva e sustentável e de grande relevância
na atualidade. Quando unimos esse tópico à previsão, ele torna-se ainda mais fascinante para
explorar. Além disso, a integração de questões relacionadas às energias renováveis com aspetos de
engenharia, como a programação de métodos de previsão, ofereceu uma abordagem muito interessante de desenvolver.
The growing concern about energy sustainability has led to a significant increase in the adoption of renewable energy sources to the detriment of non-renewable sources, such as fossil fuels. This topic has gained increasing prominence today, reflecting a global movement towards more ecological and efficient practices in the use of energy resources. Photovoltaic systems appear as a sustainable measure, allowing solar energy to be transformed into electrical energy. This type of systems is quite advantageous for the most diverse sectors, in addition to allowing cost reduction in organizations, it helps combat climate issues. In the present work, through the set of data provided containing several metrics, it was possible to study a photovoltaic panel installation located in the Porto area. The main objective was to create a fault diagnosis method using prediction methods in Python language. To this end, an exploratory analysis of the data was developed, with the purpose of obtaining important insights for the selection of the model to be developed. Therefore, according to the conclusions of the exploratory analysis, two forecast models were selected. Initially, the prediction model based on neural networks was implemented, however, it was only applied to one variable. Therefore, after a new search in the literature, it was decided to use the SARIMA time series forecasting model, since the data presented seasonality, an important characteristic for the construction of this model. This way, it was also possible to compare the results of the two models for a common variable. The main challenge was the robustness of the available data set. Despite the granularity of the data presented, data was only provided for two years, representing a substantial challenge, since the SARIMA forecast model relies on analyzing the behavior of historical data to establish reliable future forecasts. The two models made it possible to develop future prediction values on several variables relevant to identifying potential failures in the photovoltaic system. The forecast results indicate that, for the next year, the values of the variables should remain similar to historical data and, therefore, failures are not expected. These results were expected, as suppliers guarantee high performance at the beginning of the system's useful life. Photovoltaic systems represent a definitive and sustainable solution and are of great relevance today. When we combine this topic with prediction, it becomes even more fascinating to explore. Furthermore, the integration of issues related to renewable energy with engineering aspects, such as programming forecasting methods, offered a very interesting approach to develop.
The growing concern about energy sustainability has led to a significant increase in the adoption of renewable energy sources to the detriment of non-renewable sources, such as fossil fuels. This topic has gained increasing prominence today, reflecting a global movement towards more ecological and efficient practices in the use of energy resources. Photovoltaic systems appear as a sustainable measure, allowing solar energy to be transformed into electrical energy. This type of systems is quite advantageous for the most diverse sectors, in addition to allowing cost reduction in organizations, it helps combat climate issues. In the present work, through the set of data provided containing several metrics, it was possible to study a photovoltaic panel installation located in the Porto area. The main objective was to create a fault diagnosis method using prediction methods in Python language. To this end, an exploratory analysis of the data was developed, with the purpose of obtaining important insights for the selection of the model to be developed. Therefore, according to the conclusions of the exploratory analysis, two forecast models were selected. Initially, the prediction model based on neural networks was implemented, however, it was only applied to one variable. Therefore, after a new search in the literature, it was decided to use the SARIMA time series forecasting model, since the data presented seasonality, an important characteristic for the construction of this model. This way, it was also possible to compare the results of the two models for a common variable. The main challenge was the robustness of the available data set. Despite the granularity of the data presented, data was only provided for two years, representing a substantial challenge, since the SARIMA forecast model relies on analyzing the behavior of historical data to establish reliable future forecasts. The two models made it possible to develop future prediction values on several variables relevant to identifying potential failures in the photovoltaic system. The forecast results indicate that, for the next year, the values of the variables should remain similar to historical data and, therefore, failures are not expected. These results were expected, as suppliers guarantee high performance at the beginning of the system's useful life. Photovoltaic systems represent a definitive and sustainable solution and are of great relevance today. When we combine this topic with prediction, it becomes even more fascinating to explore. Furthermore, the integration of issues related to renewable energy with engineering aspects, such as programming forecasting methods, offered a very interesting approach to develop.
Description
Keywords
Photovoltaic Panels data analysis time series SARIMA Neural Networks