Developing AI-assisted energy generation estimation methods in solar power plants

Renewable energy sources are playing an increasingly significant role in meeting global energy demand. Solar energy is one of the most abundant and clean renewable energy sources. Turkey is a country with considerable potential for solar energy. Accordingly, the installation and use of solar power plants in Turkey have increased rapidly in recent years. This thesis aims to investigate the feasibility of a grid-connected, unlicensed photovoltaic solar power plant at the Bursa Technical University Mimar Sinan Campus and to develop intelligent methods for energy forecasting. In the design process, factors such as the campus's sunshine duration, system size, and cost were taken into account. Four different models were applied for energy production forecasting: Random Forest (RF), Long Short-Term Memory (LSTM), Linear Regression, and Decision Tree (DT). These models were trained using parameters such as solar irradiance, temperature, humidity, and historical production data. The performance of the models was evaluated using benchmark metrics including Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and the Coefficient of Determination (R²). Additionally, this study comprehensively addresses the legal procedures and regulatory steps required for implementing energy projects in Turkey. This section aims to provide not only the technical and economic aspects of energy investments but also their legal framework, thereby offering a holistic perspective on the processes through which energy projects are executed in Turkey. Steps ranging from unlicensed production applications to institutional approvals and project endorsements are explained in detail to provide guiding information to decision-makers and investors regarding the feasibility and sustainability of such projects. The results reveal that the LSTM and Random Forest models outperform others in terms of high accuracy and low error rates. In particular, the LSTM model demonstrated the highest performance by effectively capturing the complex relationships in time series data. Through this study, the economic feasibility and production forecasting of a campus-scale PV system have been successfully demonstrated. Moreover, by evaluating the technical, economic, and legal dimensions of energy projects in Turkey, the study offers a comprehensive approach for planning renewable energy systems.