FORECASTING THE OUTPUT POWER OF A PHOTOVOLTAIC SYSTEM BASED ON NEURAL NETWORKS

The research study conducted a detailed evaluation of the power forecasting performance of PV systems using two different models, namely LSTM and XGBoost. The experiments involved evaluating models using a variety of metrics, including MAE, R², and RMSPE, based on time data limited to one day. The results obtained confirm the high accuracy of both models, despite the limited data amount. It is particularly important to note that the XGBoost model demonstrated an impressive coefficient of determination (R²) of 0.99 in predicting solar panel power, while the LSTM model performed satisfactorily with an R² of 0.06. The radiation prediction analysis also revealed that the XGBoost model achieved a high R² of 0.97, while the LSTM model performed well with an R² of 0.67. These results highlight the successful ability of deep models in predicting PV production, ensuring stability and reliability of the predictions. The full research analysis is presented in this article.