Loading...
Publication
Validação de uma metodologia de cálculos de estimativas de produção de energia em parques fotovoltaicos em operação
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 9.22 MB | Adobe PDF |
Authors
Abstract(s)
A transição energética é um dos grandes desafios do nosso tempo, e a energia solar fotovoltaica tem-se afirmado como catalisador desta transformação. Torna-se, por isso, indispensável dispor de metodologias robustas para estimar a produção a longo prazo em parques já em operação. Ao contrário das previsões de pré-construção, as estimativas pós-operacionais permitem recalibrar as projeções futuras de produção anual de energia, incorporando dados operacionais reais. Esta abordagem possibilita a redução das incertezas associadas a estudos
energéticos preliminares, decisões de investimento, Power Purchase Agreements (PPAs) e avaliações de risco.
Neste enquadramento, a presente investigação pretende validar a aplicabilidade da metodologia Measure–Correlate–Predict (MCP), originalmente desenvolvida no setor eólico, ao setor fotovoltaico, adaptando-a a dados mensais de produção real e a séries de radiação de longo prazo provenientes de bases reconhecidas. A abordagem assentou na regressão linear simples entre a produção medida e a série de radiação de referência, após uma análise comparativa de três bases (Solargis, PVGIS e ERA5) com o intuito de identificar a mais adequada para caracterizar cada parque. Esta etapa foi complementada por um processo de deteção de outliers, assegurando uma maior robustez das correlações. Em paralelo, foram explorados modelos alternativos como: regressão múltipla e algoritmos de machine learning (Random Forest e XGBoost) para comparar o respetivo desempenho preditivo com o MCP clássico. Os resultados evidenciam que, quando suportado por seleção criteriosa da base de radiação e por tratamento sistemático de outliers, o MCP apresenta desempenho consistente e fiável em contexto pós-operacional. As abordagens de machine learning revelam ganhos pontuais em cenários de maior variabilidade, constituindo um complemento pertinente à clareza e parcimónia da regressão linear. Em síntese, o estudo reforça a confiança nas estimativas de longo prazo em parques fotovoltaicos em operação, contribui para a redução de incertezas e informa decisões de investimento.
The energy transition is one of the great challenges of our time, and photovoltaic solar energy has established itself as a catalyst for this transformation. It is therefore essential to have robust methodologies for estimating long-term production in parks already in operation. Unlike preconstruction forecasts, post-operational estimates allow future annual energy production projections to be recalibrated, incorporating actual operational data. This approach reduces the uncertainties associated with preliminary energy studies, investment decisions, power purchase agreements (PPAs), and risk assessments. In this context, the present research validates the applicability of the Measure–Correlate–Predict (MCP) methodology, originally developed in the wind sector, to the photovoltaic sector, adapting it to monthly actual production data and long-term radiation series from recognized databases. The approach was based on simple linear regression between measured production and the reference radiation series, after a comparative analysis of the three databases (Solargis, PVGIS, and ERA5) to identify the most appropriate one to characterize each farm. This step was complemented by a process of outlier detection and management, ensuring a greater robustness of the correlations. In parallel, alternative models were explored, such as multiple regression and machine learning algorithms (Random Forest and XGBoost), to compare their predictive performance with that of the classic MCP. The results show that, when supported by careful selection of the radiation base and systematic treatment of outliers, MCP performs consistently and reliably in a post-operational context. Machine learning approaches reveal occasional gains in scenarios of greater variability, constituting a relevant complement to the clarity and parsimony of linear regression. In summary, the study reinforces confidence in long-term estimates for photovoltaic parks in operation, contributes to reducing uncertainties, and informs investment decisions.
The energy transition is one of the great challenges of our time, and photovoltaic solar energy has established itself as a catalyst for this transformation. It is therefore essential to have robust methodologies for estimating long-term production in parks already in operation. Unlike preconstruction forecasts, post-operational estimates allow future annual energy production projections to be recalibrated, incorporating actual operational data. This approach reduces the uncertainties associated with preliminary energy studies, investment decisions, power purchase agreements (PPAs), and risk assessments. In this context, the present research validates the applicability of the Measure–Correlate–Predict (MCP) methodology, originally developed in the wind sector, to the photovoltaic sector, adapting it to monthly actual production data and long-term radiation series from recognized databases. The approach was based on simple linear regression between measured production and the reference radiation series, after a comparative analysis of the three databases (Solargis, PVGIS, and ERA5) to identify the most appropriate one to characterize each farm. This step was complemented by a process of outlier detection and management, ensuring a greater robustness of the correlations. In parallel, alternative models were explored, such as multiple regression and machine learning algorithms (Random Forest and XGBoost), to compare their predictive performance with that of the classic MCP. The results show that, when supported by careful selection of the radiation base and systematic treatment of outliers, MCP performs consistently and reliably in a post-operational context. Machine learning approaches reveal occasional gains in scenarios of greater variability, constituting a relevant complement to the clarity and parsimony of linear regression. In summary, the study reinforces confidence in long-term estimates for photovoltaic parks in operation, contributes to reducing uncertainties, and informs investment decisions.
Description
Keywords
Photovoltaic energy Measure?Correlate?Predict (MCP) Long-term production estimation post-operational data Radiation bases Machine learning (ML) Energia fotovoltaica Estimativa de produção de longo prazo Dados pós-operacionais Bases de radiação