부산대학교 도서관

With increased global demands for renewable energy, the prediction of photovoltaic (PV) power generation becomes more necessary for efficient power scheduling and energy usage. Since PV generation is heavily dependent on weather conditions, using historical weather data and machine learning techniques to predict solar power generation is a popular approach to solving the problem. In this paper, we focus on the Support Vector Machine (SVM) to create a regression model capable of forecasting PV generation one hour ahead with two cases: fair and cloudy weather. The SVM model is trained by minimizing the one-step prediction (15-minute ahead) error, and then its hyperparameters are tuned using GridSearchCV and a validation dataset. The final performance of the resulting model is evaluated on the testing dataset for one-hour ahead prediction.