Artificial neural network model with different backpropagation algorithms and meteorological data for solar radiation...

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Environmental Science

Background:

• Solar energy is a crucial clean alternative to fossil fuels.
• Accurate solar radiation (SR) forecasting minimizes costs in solar energy development.
• Existing SR forecasting methods include support vector machines, autoregressive moving average, and artificial neural networks (ANNs).

Purpose of the Study:

• To comprehensively study meteorological data and backpropagation (BP) algorithms for developing optimal SR predicting ANN models.
• To compare the predictive abilities of ANNs trained with different BP algorithms and meteorological inputs.
• To identify the best combination of meteorological data and BP algorithm for superior SR prediction.

Main Methods:

• Collected meteorological data (temperature, relative humidity, wind speed) from Kuala Terengganu, Malaysia.
• Employed three distinct BP algorithms for ANN training: Levenberg-Marquardt, Scaled Conjugate Gradient, and Bayesian Regularization (BR).
• Evaluated and compared the performance of various ANN models based on predictive accuracy.

Main Results:

• Temperature and relative humidity exhibit a strong correlation with SR.
• Wind speed has a minimal influence on SR.
• ANN models trained with the BR algorithm achieved a maximum R of 0.8113 and a minimum RMSE of 0.2581, outperforming other models.

Conclusions:

• The study successfully identified optimal parameters for SR prediction using ANNs.
• Meteorological data, particularly temperature and humidity, are significant predictors of SR.
• The Bayesian Regularization algorithm provides superior performance for training ANN models in SR forecasting.