Very Short-Term Photovoltaic Power Forecasting Using Stochastic Factors

Abstract

This paper proposes a photovoltaic (PV) power forecasting model, using the application of a Gaussian blur algorithm filtering technique to estimate power output and the creation of a stochastic forecasting model. As a result, affected power can be forecasted from stochastic factors with machine learning and an artificial neural network. This model focuses on very short-term forecasting over a five minute period. As it uses only endogenous data, no exogenous data is needed.
 To evaluate the model, results were compared to the persistence model, which has good short-term forecasting accuracy. This proposed PV forecasting model gained higher accuracy than the persistence model using stochastic factors.