Comparison of production estimates with actual production values using solar power plant design programs and artificial intelligence assistance: Civril example

Solar energy systems attract the attention of investors by standing out in the national market with their accessibility and cost-effectiveness. Prior to installation, the system power and geographic data of the region are processed in simulation software to make annual production estimates. However, these software may cause deviations in the estimates due to the fact that they cannot keep the data up-to-date and do not take into account the power plant losses. In this case, AI models can step in and provide more accurate production forecasts. Processing of meteorological data by artificial intelligence with machine learning and deep learning methods increases forecast performance. In this thesis, electricity production estimates of photovoltaic (PV) systems were analyzed using extreme gradient augmentation (XGBoost), support vector regression (SVR), multilayer perceptron (MLP) and long short-term memory (LSTM) models. The results obtained were compared with common simulation software such as PVSOL and PVSYST. Evaluations with actual production data have shown that the MLP and XGBoost models offer more reliable results with low error rates and a high coefficient of accuracy. While machine learning methods make predictions that are more suitable for field conditions thanks to their sensitivity to environmental variables, it has been determined that PVSOL and PVSYST software work with a deviation of 2-10%. These findings reveal that artificial intelligence methods are an effective alternative for more accurate production predictions.