Artficial intelligence – based energy generation estimating for rooftop photovoltaic plants

In this study, two different models, XGBoost and deep learning algorithm RNN, were used to predict energy production amounts in rooftop photovoltaic power plants. Our goal is to accurately predict future energy production amounts based on historical data, optimize plant performance, and develop energy management strategies. XGBoost is a popular machine learning technique based on gradient boosting and decision tree algorithms, providing high predictive power and speed advantages. Especially excelling in learning complex relationships and improving efficiency in energy production forecasting. Deep learning algorithms, on the other hand, are artificial intelligence methods that perform strongly with large data sets and offer data-driven solutions. In this study, especially deep learning model RNN were used. Recurrent neural networks are suitable for analyzing sequential data and are used to predict future data by considering past information. In the study, an evulation was made on the daily frequency data set between 01.03.2021 and 30.03.2024 of the solar power plant installed in Balıkesir provience with a capacity of 1 MW. This data includes the generation amounts recorded by solar power plants at different time intervals. The data set includes various features such as date, time, solar radiation, temperature, and weather conditions. These features are included in the model as fundamental factors affecting energy production amounts. In conclusion, the deep learning algorithm considered is the Renewable Neural Networks RNN, which is evaluated with another application model, the machine learning algorithm XGBoost. The performance of the dataset forecasting results was evaluated with 4 different error metrics, (NMSE, MAPE, R2, MedAe). The XGBoost model provided higher accuracy in solar power generation forecasting compared to RNN, demonstrating the importance of data-driven approaches in this field. This thesis provides valuable information in terms of strategic planning and resource management in the renewable energy sector. Increasing the accuracy of energy production forecasts will contribute to ensuring energy supply security and reducing energy costs. Additionally, these approaches can be applied to the management of other renewable energy sources, thus playing an important role in the transition to a more sustainable energy system.