| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 4.86 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
Esta dissertação teve como propósito desenvolver metodologias de análise de
correlações estatísticas entre séries de dados de vento, com diversos métodos Measure Correlate-Predict (MCP), com o objetivo de selecionar o método mais adequado para a
extrapolação dos dados de longo termo com a menor incerteza associada. Pretendeu-se,
também, analisar o tempo mínimo necessário de uma campanha de medição do vento a
utilizar na aplicação desta metodologia.
Para a realização deste trabalho foram selecionadas 15 estações de medição local
de vento. O processo de seleção teve em conta a maior extensão possível do período da
campanha de medição, com disponibilidades de dados de velocidade e de direção de vento
superiores a 90% e com variadas orografias, rugosidades e regime de ventos. De seguida,
selecionou-se a série de reanálise de dados de vento de longo termo que melhor se
correlacionou com os dados de vento medidos nas estações de medição. Posteriormente,
realizaram-se vários testes, com diferentes períodos simultâneos entre a série de dados
medidos e a série de longo termo onde se testaram, para cada período concorrente, 15
algoritmos de correlação: 13 regressões, 1 matriz, e 1 rede neural artificial. O desempenho
de cada modelo testado foi avaliado através da caraterização dos parâmetros de
performance RMSE (Root Mean Square Error) e MBE (Mean Bias Error) associados a
cada MCP.
A análise dos erros obtidos permitiu identificar que os períodos de medição com
menor erro associado se compreendem entre 1 e 5 anos. Para cada um destes anos foi feita
uma análise de variância ANOVA de fator único com o objetivo de perceber a
significância do ano de medição. Por fim, realizaram-se testes de significância t para os
anos selecionados, com o objetivo de obter os algoritmos de MCP com médias diferentes.
Este trabalho permitiu concluir que para a amostragem selecionada, o algoritmo 4
Seasons 360s DN apresentou-se como sendo o melhor método de MCP a ser usado, por
apresentar o menor valor de RMSE e MBE e conseguir captar adequadamente a
sazonalidade de cada local. Por outro lado, concluiu-se que o número mínimo de anos
necessários para uma campanha de medição local deve situar-se entre os 2 e os 3 anos.
This dissertation aimed to develop methodologies for analyzing statistical correlations between series of wind data, with different Measure-Correlate-Predict (MCP) methods, with the aim of selecting the most appropriate method for extrapolating long term data with the lower associated uncertainty. It was also intended to analyze the minimum time required for a wind measurement campaign to be used in the application of this methodology. To carry out this work, 15 local wind measurement stations were selected. The selection process took into account the largest possible extension of the measurement campaign period, with availability of wind speed and direction data greater than 90% and with varied orographies, roughness and wind regime. Next, the long-term wind data reanalysis series that best correlated with the wind data measured at the measuring stations was selected. Subsequently, several tests were carried out, with different simultaneous periods between the measured data series and the long-term series, where 15 correlation algorithms were tested, for each concurrent period: 13 regressions, 1 matrix, and 1 artificial neural network. The performance of each model tested was evaluated through the characterization of the performance parameters RMSE (Root Mean Square Error) and MBE (Mean Bias Error) associated with each MCP. The analysis of the errors obtained allowed us to identify that the measurement periods with the lowest associated error are between 1 and 5 years. For each of these years, a single-factor ANOVA analysis was carried out with the aim of understanding the significance of the year of measurement. Finally, t-significance tests were carried out for the selected years, with the aim of obtaining MCP algorithms with different means. This work allowed us to conclude that for the selected sampling, the 4 Seasons 360s DN algorithm was the best MCP method to be used, as it presented the lowest RMSE and MBE value and was able to adequately capture the seasonality of each location. On the other hand, it was concluded that the minimum number of years required for a local measurement campaign should be between 2 and 3 years.
This dissertation aimed to develop methodologies for analyzing statistical correlations between series of wind data, with different Measure-Correlate-Predict (MCP) methods, with the aim of selecting the most appropriate method for extrapolating long term data with the lower associated uncertainty. It was also intended to analyze the minimum time required for a wind measurement campaign to be used in the application of this methodology. To carry out this work, 15 local wind measurement stations were selected. The selection process took into account the largest possible extension of the measurement campaign period, with availability of wind speed and direction data greater than 90% and with varied orographies, roughness and wind regime. Next, the long-term wind data reanalysis series that best correlated with the wind data measured at the measuring stations was selected. Subsequently, several tests were carried out, with different simultaneous periods between the measured data series and the long-term series, where 15 correlation algorithms were tested, for each concurrent period: 13 regressions, 1 matrix, and 1 artificial neural network. The performance of each model tested was evaluated through the characterization of the performance parameters RMSE (Root Mean Square Error) and MBE (Mean Bias Error) associated with each MCP. The analysis of the errors obtained allowed us to identify that the measurement periods with the lowest associated error are between 1 and 5 years. For each of these years, a single-factor ANOVA analysis was carried out with the aim of understanding the significance of the year of measurement. Finally, t-significance tests were carried out for the selected years, with the aim of obtaining MCP algorithms with different means. This work allowed us to conclude that for the selected sampling, the 4 Seasons 360s DN algorithm was the best MCP method to be used, as it presented the lowest RMSE and MBE value and was able to adequately capture the seasonality of each location. On the other hand, it was concluded that the minimum number of years required for a local measurement campaign should be between 2 and 3 years.
Description
Keywords
MCP correlation algorithms local measurement station wind regime orography reanalysis series