Machine learning supported web-based production forecasting using weather forecast service data for solar power plants

As the global population continues to grow and technological advancements progress, energy demand is becoming increasingly noticeable worldwide. Solar energy, which is among the sustainable energy sources to reduce the environmental impact of fossil fuels, has a critical role in the global energy transition. Energy generation forecasts play a vital role in supply-demand balance, grid stability and cost optimization. Moreover, accurate and reliable generation forecasts are essential to facilitate the integration of renewable energy sources and improve the efficiency of energy systems. In this study, two years of production data from a solar power plant in Denizli province and weather data obtained from Solcast API service are used. The performance of machine learning algorithms such as XGBoost, Extra Trees, k-Nearest Neighbors, Gradient Boosting, Random Forest and Linear Regression are evaluated. According to the results obtained, the Extra Trees model demonstrated the best performance in one- day predictions with a Mean Square Error value of 24.290, Root Mean Square Error value of 4.928, Mean Absolute Error value of 3.123, and an R² score of 0.993. For five-day and one-week predictions, Extra Trees and Random Forest models showed similar performance, producing more successful results than other algorithms. This study contributes to a more reliable estimation of solar power generation, facilitating the integration of renewable energy sources and offering significant potential for the optimization of energy management systems.