Accurate hourly forecasting of photovoltaic energy production using artificial intelligence algorithms: An approach based on the use of high-certainty inputs

Accurate hourly forecasting of photovoltaic energy production using artificial intelligence algorithms: An approach based on the use of high-certainty inputs

Soha Sami1 Mohammad Hajian2 Tayyeb Nazghelichi3 Jafar Massah4 seyed Reza Hassan-beygi5

1) Master s student in Biosystems Mechanics, Renewable Energy Orientation, University of Tehran, Iran
2) Master s student in Biosystem Mechanics, Renewable Energy Orientation, Gorgan University of Agricultural Sciences and Natural Resources, Iran
3) Assistant Professor, Gorgan University of Agricultural Sciences and Natural Resources, Iran
4) Professor, Faculty of Agricultural Technology, University of Tehran, Iran
5) Professor, Faculty of Agricultural Technology, University of Tehran, Iran

IOI Code : XKBA-AAAGDE
Publication : 3rd International Conference on Recent Advances in Engineering, Innovation and Technology - Belgium(eitconf.com/3rd)
Abstract :
    Accurate forecasting of photovoltaic (PV) power generation is of utmost importance for the effective integration of renewable energy sources and improvement in efficiency of power grids. However, PV power forecasting is a complex task because it depends on erratic weather conditions and hourly changes. These complexities show the need for advanced algorithms and better quality data to enhance the accuracy of the forecast. Moreover, much of the available meteorological data has been questioned due to its inherent uncertainty for use in short-term forecasting. The objective of this study was to forecast the hourly output of photovoltaic power using parameters characterized by high reliability and availability. To achieve this objective, the base dataset included hourly photovoltaic generation data from an industrial site in Konstanz, Germany, which was obtained via the Open Power System Data (OPSD) project. The temporal and climatic features of the area were then compiled to create a complete dataset. The availability and reliability of the model inputs were then assessed as criteria for selecting and determining appropriate inputs, thereby aiding in the generation of predictions with high accuracy and acceptable confidence levels. The model inputs were selected with great care following extensive data correlation analysis and examination of interdependencies between inputs, and different forecasting models were compared and contrasted. As a result, the LSTM + LGBR hybrid model showed outstanding performances, with the following results on four evaluation indicators: R² Score: 0.9864, MAE: 0.0053, MSE: 0.0003, RMSE: 0.0159. The predictions made by the model, when analyzed against similar meteorological inputs, proved

Keywords : short term forecasting temporal parameters hybrid model input certainty renewable energy