https://www.anmb.ro/buletinstiintific/buletine/2023_Issue1/02_EEA/169-177.pdf

The Photovoltaic (PV) systems are more present in the communities’ landscape providing energy to the consumers, public buildings, municipalities, and industry, smoothen the electricity prices fluctuations and reducing the dependency on the public grid. They are reliable energy sources for boats and ships as some of the PV technologies are flexible and can be located on plane surfaces or even on the water surface especially when the ships dock at sea or at the seashore. However, the operation of PV systems depends on several weather factors, and it is important to predict their operation to manage the controllable load. Furthermore, it is essential to know if the PV systems generate in surplus or additional energy is required to cover the load. The surplus can be offered for local trading or aggregated and offered for centralized markets. Therefore, in this paper, we aim to predict the output of the PV systems using machine learning algorithms and recurrent neural networks (RNN), especially a multivariate Long Short-Term Memory (LSTM) model. Data extraction, feature engineering, and forecast of the PV power are depicted and the simulations are performed using 4 PV systems located in Constanta County. The results are assessed with prediction performance metrics such as RMSE, MAPE, etc.