Using Artificial Intelligence Methods for Power Estimation in Photovoltaic Panels

The limited reserves of fossil resources, the fluctuations in their prices and the damage they cause to the environment have led countries to seek alternatives to primary energy resources. Solar energy, which is an unlimited and environmentally friendly resource, is a powerful alternative to other energy sources. The majority of the European Union countries offer various opportunities to consumers in electricity generation from solar energy with many incentive mechanisms and ensure their widespread use. In many parts of the world, interest in renewable energy sources such as solar, wind, hydrogen and geothermal is also growing. In addition to all these, researches are continuing to use alternative energy sources and to make energy production more efficient. The radiation value required to obtain electricity from solar energy varies according to the weather conditions during the day and seasonal characteristics. The climatic conditions in the area where solar power plants are installed directly affect the output power and energy cost to be obtained from photovoltaic panels. Estimating the output power produced from photovoltaic panels according to environmental conditions, guiding companies in the installation of solar energy systems, obtaining maximum energy, energy management and efficient operation of the system are of great importance. In this study, feedforward back propagation artificial neural networks and KNN (K-Nearest Neighbors) methods were used to estimate power values using the data (Temperature, Humidity, Pressure, Radiation) obtained from the installed photovoltaic panels. Thus, the panel values obtained under real field conditions were trained with both methods and the results were compared. As a result, the power values of the panel were classified using the artificial neural network model developed with the highest accuracy of 98.7945%. It has been seen that the machine learning models used for solar energy estimation developed within the scope of this study have high performance and can produce results very close to the real values. In addition, it was concluded that both artificial intelligence models developed in locations with different characteristics according to the determined load demand can be used.