Comparing Machine Learning Methods to Predict Photovoltaic Power Output

The purpose of this paper is to develop the models to predict 24-hour ahead photovoltaic (PV) power generation and compare the performance of the developed models. To develop the models for predicting PV power output, we first develop a support vector regression (SVR) based model to predict solar isolation and a model to predict cloudiness as sub models. The model to predict cloudiness uses data for sky condition and the model to predict solar isolation uses the weather forecast, the actual measured data, and the derived variables from the actual data. Second we develop a SVR based model and artificial neural network (ANN) based models to predict PV power output with weather forecast data, actual measured weather data, and some derived data. The performances of these models are compared in terms of mean absolute error and mean absolute percentage error. The experimental result shows that the SVR based model has superior performance comparing with the ANN based models. Also, we analyze the factors to decrease the performance of the model and suggest the directions to improve the model.