Prediction and Monitoring of Solar Radiation Using Artificial Neural Networks for Renewable Energy Optimization

Accurate prediction of solar radiation is essential in the maximization of the output and planning of the renewable energy systems. The current paper proposes and tests the experimental daily global horizontal irradiance (GHI) forecasting of an Artificial Neural Network (ANN) model with a number of meteorological variables taken as input variables in the NASA POWER database. For the input features, ANN was trained with five input parameters: air temperature, relative humidity, wind speed, surface pressure, thermal range, and it utilized architecture with two hidden layers (128-64 neurons). Mean Absolute Error (MAE) calculation, Root Mean Square Error (RMSE) calculation and the coefficient of determination (R2) were used to assess the performance of models. Predictive capacity was high as indicated by a low MAE of 0.754 MJ/ m2/day, RMSE of 0.943 MJ/ m2/day, and R2 of 0.725 that interprets data to mean the model explains about 73% of GHI variation. Model stability and aids of monthly boxplots, visual diagnostics, and residual analysis were all in agreement in terms of the accuracy and stability of the model. The method is mathematically lean, interpretable, and is appropriate in data-scarce environments. This ANN model provided a viable and scalable solution to solar energy forecasting and helps in making decisions on planning and integrating photovoltaic systems into the grid.