INVESTIGATION OF THE MOST SUITABLE POWER OUTPUT PREDICTION METHODS WITH ARTIFICIAL INTELLIGENCE IN A ROOFTOP PHOTOVOLTAIC POWER PLANT

Abstract

Installing photovoltaic (PV) systems in buildings effectively achieves sustainable energy targets and reduces carbon emissions. Energy demand is increasing day by day. Accessing solar energy is preferred, especially in urban areas, because it is easier and more economical than other renewable energy sources. It is important to calculate the losses that occur in the integration of PV systems into the interconnected system and to select the appropriate material for the system. In the feasibility reports prepared before the system is installed, the selection of appropriate materials for the system, system cost, energy production and consumption, and amortization periods are calculated by considering the environmental and physical conditions. The dataset used in this study was obtained from two rooftop PV systems (each 200 kW) installed on separate buildings of Yüksek İhtisas Hospital in Bursa, Turkey, with production and ambient temperature data collected at 15-minute intervals throughout 2024. This study investigates the use of artificial intelligence techniques—Decision Tree, Random Forest, LSTM, and Linear Regression—for predicting photovoltaic (PV) power output using real data from two 200 kW rooftop PV power plants located at Yüksek İhtisas Hospital in Bursa, Turkey. One-year production, irradiance, and ambient temperature data recorded at 15-minute intervals were used. The aim was to forecast the expected power output of a 440 kW PV system to be installed on the BTU G Block under similar environmental and technical conditions. The effects of environmental and physical conditions on one-year production data were examined using various artificial intelligence methods such as Random Forest, Decision Tree, Linear Regression, and LSTM. The aim was to predict the production data that would arise when a power plant with similar environmental and physical conditions is established. According to the analysis results, the Decision Tree method was determined to be the highest-performing technique, providing a 99.6% R² accuracy value.