Creative Commons

The day-ahead Photovoltaic (PV) power forecasting has become increasingly important since it is a key factor to mitigate the economic dispatch issues caused by the extensive integration of PV systems into the Distribution System. In literature, many research papers have presented various methods that utilize statistical, Machine Learning, or Deep Learning techniques. These approaches aim to improve the PV power forecasting accuracy. The scope of this dissertation is the sizing of Battery Energy Storage Systems (BESS) in order to minimize the PV owners’ economic impact due to forecasting errors. The objective is to ensure the secure integration of PV power into the day ahead energy market. The sizing of the BESS will be determined by utilizing day-ahead forecasts generated by four forecasting models. These models include Long Short-Term Memory (LSTM), Feed-Forward Neural Networks (FFNN), Random Forest Regression (RFR), and Gradient Boosting (Gboost). Based on the BESS sizing the forecasting models’ efficiency will be assessed as well. The results indicate that the RFR model can improve forecasting accuracy since it outperformed the other models in terms of minimizing errors and achieving superior results in battery sizing.