English abstract
The worldwide generation of solar energy has grown at an unprecedented rate over the
last decade. However, several challenges are still associated with solar energy. One of
which is the unreliability of this renewable energy source due to the imperfect prediction
of weather conditions. As such, accurate solar irradiance predictors are necessary to
improve the reliability of solar energy. Such an accurate predictor can be embedded into
an energy management system for renewable power generation to perform early
prediction of Solar irradiance. This helps during power generation planning and
supporting cost reduction strategies in the energy industry.
This thesis proposes an accurate and fast solar irradiance predictor using a data-driven
deep learning model. The model is based on the Long Short-Term Memory (LSTM)
architecture. The model was developed using a complex machine learning algorithm
with three inputs (i.e., Wind Sustained Speed, Temperature Dry, and Solar irradiance)
and one output (Solar irradiance). The model was trained and tested using actual weather
measurements from the Thumrait area, Sultanate of Oman. Furthermore, an additional
dataset from Marmul was used to explore if the proposed model can be used to predict
irradiance at locations other than Thumrait.
The simulation results show that the proposed model outperforms the ARIMA model in
terms of prediction accuracy, where the former scored RMSE of 24 W/m2
compared to
279 W/m2
scored by the latter. Additionally, the proposed model runs faster compared
to other contending models found in the literature. More specifically, the model can
predict the solar irradiance of one year in only 23 seconds. Overall, the results show that
the proposed model is more effective in dealing with time-series datasets.
