Harnessing Deep Learning and Time Series Models for Accurate Global Solar Radiation Prediction

For applications such as weather forecasting, power grid management, and solar energy integration, accurate solar radiation prediction is essential. Using real-time data from the EU Science Hub’s Climate Monitoring Satellite Application Facility (CM SAF), this research examines the efficacy of several time series models for hourly solar radiation prediction. The data are subjected to stationarity testing using the Augmented Dickey-Fuller (ADF) test prior to the models being applied. This study uses well-known time series models, including Vector Auto-regression (VAR), Linear Regression, and Autoregressive Integrated Moving Average (ARIMA). The performance of the proposed model is evaluated using Mean Square Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), R-squared, and Mean Absolute Percentage Error (MAPE). It is found that Vector Auto-regression (VAR) is encouraging for predicting solar radiation on hourly basis with least amount of overall error. This analysis has used residual error inspection, error distribution analysis and sample accuracy evaluation using MAPE. This paper highlights the importance of VAR on using time series model for predicting solar radiation.