Short - Term Solar Radiation Forecasting for Adaptive Solar Cell Using a Hybrid D-FFNN Model

Abstract

Variations in the value of solar radiation affect the efficiency of solar panels and thus cannot be predicted. This causes instability in the generated power and negatively impacts the reliability of the solar power plant. This research formulates the prediction of solar radiation at short time intervals based on decomposition techniques and Artificial Neural Network - Feed Forward Multi-Layer Perceptron. The prediction method consists of the following steps: data pre-processing, application of the modified FFN model, and evaluation through RMSE, MSE, MAPE, and MAE. The experimental results show that the prediction model performs well, with MAE of 3.3823, MSE of 37.0858, RMSE of 6.089, and MAPE of 0.29%. Although there is variation between the predicted and actual values, the overall trend of the data yields good accuracy, with a deviation of only 5%. This shows that the combination of decomposition and FFNN techniques can improve the prediction accuracy of solar radiation which is crucial for adaptive solar panel system applications. The next research direction is to build more complicated models using larger data shares and other decomposition methods, which will improve the prediction accuracy under rapidly changing weather conditions.