Simulation of grid-connected photovoltaic solar power plant and energy production estimation with machine learning algorithms

This thesis aims to compare the actual electrical energy production values of a grid-connected photovoltaic (PV) solar power plant (SPP) installed in the Kemalpaşa district of Izmir with data obtained in a simulation environment, and to develop methods for increasing system efficiency based on this comparison. Using solar radiation data obtained through the PV Geographic Information System (PVGIS) software developed by the European Commission, the existing system was modeled using the "solarVis" simulation software. Subsequently, the obtained simulation outputs were compared with the actual production data to analyze the system's production consistency. Additionally, various parameters affecting PV system efficiency, such as panel tilt angle, orientation, cell shading, and dust accumulation, were evaluated to determine the impact of these parameters on the system. In the second phase of the study, machine learning algorithms such as K-Nearest Neighbor, Linear Regression, Decision Tree, Support Vector Machines, Ensemble Methods, and Gaussian Processes were applied using the existing system's production data and meteorological data. Using these algorithms, prospective production forecasts were performed, and the most appropriate method was determined by analyzing the prediction accuracies of the algorithms. In this context, this study has made significant contributions both in terms of increasing the physical efficiency of PV-SPPs and in terms of showing the applicability of new methods in production estimation. Keywords: renewable energy, energy, solar power plant, photovoltaic solar power plants, machine learning