Prediction of electrical power and illuminance in an off-grid photovoltaic system using artificial neural networks

This master’s thesis focuses on the design and experimental validation of predictive models based on artificial neural networks for short-term forecasting of photovoltaic system power output and illuminance. As part of the work, an off-grid photovoltaic system is implemented to enable the collection of electrical and meteorological data. The acquired data are subsequently analyzed and preprocessed for the training of several selected neural network models. These models are then compared in terms of prediction accuracy for different input window lengths and forecasting horizons. The evaluation is carried out using the MAE, RMSE, and R² metrics. The results indicate that prediction accuracy depends on the choice of model, input data parameters, and the nature of the predicted variable. The conclusion discusses the suitability of the models for practical applications and the limitations arising from the nature of the problem.