ANN and Random Forest based Prediction of Power Output of Solar Panel

The globally upcoming demand for renewable resources of energy has increased the importance of predictive modelling of solar energy. This paper states development of two machine learning techniques: Artificial Neural Networks using Levenberg- Marquart (LM), and Random Forests (RF), in MATLAB, for estimating power generated by a solar panel. This study focuses on efficient models for amplifying the prediction accuracy of solar energy panels. A dataset, comprising of numerous operational and environmental variables affecting performance of solar panels, forms the base for development of model. The essential parameters include ambient temperature, panel temperature, solar irradiance, wind speed, panel efficiency, and dust deposition. The data pre-processing steps involved include handling missing data (rows), normalization of data and splitting of dataset into training and testing data for efficient model estimation. The results illustrated that both the models have robust performance, ANN marginally outperformed RF. As lower RMSE and MAE values and comparatively higher accuracy values, signifying a better fit to data. ANNs better performance is due to its capability to handle nonlinear complex relationships in data, crucial for accurate prediction of solar panel output. While RF model is interpretable and robust but suffers from limitations like unable to handle complicated and non-linear patterns.