Forecasting hourly solar irradiance in South Africa using machine learning models

Show simple item record

dc.contributor.advisor Sigauke, Caston
dc.contributor.advisor Mbuvha, Rendani
dc.contributor.author Mutavhatsindi, Tendani
dc.date 2020
dc.date.accessioned 2021-02-02T13:46:17Z
dc.date.available 2021-02-02T13:46:17Z
dc.date.issued 2020-08-12
dc.identifier.citation Mutavhatsindi Tendani (2020) Forecasting hourly solar irradiance in South Africa using machine learning models. University of Venda, South Africa,<http://hdl.handle.net/11602/1664>.
dc.identifier.uri http://hdl.handle.net/11602/1664
dc.description MSc (Statistics) en_ZA
dc.description Department of Statistics
dc.description.abstract Solar irradiance forecasting is essential in renewable energy grids amongst others for back-up programming, operational planning, and short-term power purchases. This research study focuses on forecasting hourly solar irradiance using data obtained from the Southern African Universities Radiometric Network at the University of Pretoria radiometric station. This research project compares the predictive performance of long short-term memory (LSTM) networks, support vector regression (SVR), and feed forward neural networks (FFNN) model for forecasting short-term solar irradiance. While all the models outperform principal component regression model, a benchmark model in this study, the FFNN yields the lowest mean absolute error (MAE) and root mean square error (RMSE) on the testing set. According to findings, among the three fitted machine learning models, the FFNN model produced the best forecast accuracy based on MAE and RMSE. Forecast combination of machine learning models’ forecasts is done using convex combination and quantile regression averaging (QRA). Based on MAE and average pinball losses, the QRA forecast combination model is the best forecast combination method, and also the best forecasting model compared with the individual machine learning models. Further analysis of the prediction interval widths (PIWs) based on the prediction interval coverage probabilities (PICPs), and prediction interval normalised average widths (PINAWs) including a count of the number of predictions below and above the PIs show inconsistency. The best model based on PINAW at 90% is QRA, FFNN at 95%, and SVR at 99%. There is inconsistency in the results of PICPs and PINAWs for different PINC values. The residual analysis shows FFNN as the best model with narrowest error distribution compared to other models, followed by QRA. en_ZA
dc.description.sponsorship NRF en_ZA
dc.format.extent 1 online resource (xiii, 75 leaves : color illustrations, color maps)
dc.language.iso en en_ZA
dc.rights University of Venda
dc.subject Forecast combination en_ZA
dc.subject Machine learning en_ZA
dc.subject Neural networks en_ZA
dc.subject Solar irradiance en_ZA
dc.subject Support vector regression en_ZA
dc.title Forecasting hourly solar irradiance in South Africa using machine learning models en_ZA
dc.type Dissertation en_ZA

Files in this item

This item appears in the following Collection(s)

Show simple item record

Search UnivenIR


My Account