Forecasting output of a solar power plant using meteorological data

In this thesis, a model has been developed for forecasting solar energy production, one of the most prominent renewable energy sources. Although solar energy is characterized by low environmental impact, high sustainability, and long-term economic viability, it also exhibits a high degree of variability due to its natural characteristics. This variability causes intermittency in energy generation and poses significant challenges for energy planning and grid management. Therefore, reliable forecasting of solar energy production has become a critical requirement for energy management systems. Within the scope of the study, a deep learning model based on the Long Short-Term Memory (LSTM) architecture, which is highly capable of processing time series data, was developed. For the training and testing of the model, real-time solar panel production data from a specific region of Turkey were used, along with meteorological parameters such as temperature, humidity, wind speed, and solar radiation duration. In the data preprocessing stage, techniques such as normalization, missing data imputation, and time-shifted windowing were applied. To evaluate the model's performance, statistical metrics including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the coefficient of determination (R²) were employed. The results showed that the developed LSTM model generated highly accurate predictions with lower error rates compared to traditional statistical methods and some machine learning algorithms. This demonstrates that LSTM-based models are particularly successful in handling complex variability inherent in time series data such as solar energy.