Short-term power prediction of distributed photovoltaic based on CNN-LSTM and meteorological interpolation

Abstract

The photovoltaic power generation has strong randomness and fluctuation. Considering the traditional photovoltaic power prediction methods mainly focus on centralized photovoltaic and rely heavily on meteorological data, while distributed photovoltaic usually does not have perfect meteorological data, so the existing photovoltaic power prediction methods are difficult to apply to distributed photovoltaic. Based on the known wide-area meteorological resource data, the model using the Kriging method and the inverse distance weighting (IDW) method to interpolate the meteorological data of the of the distributed photovoltaic power stations by its longitude and latitude positions, using the CNN network to extract the hidden feature information, and constructs a prediction model based on LSTM. The model verifies the effectiveness of the method through actual data.