Ultra Short-term Forecast of Photovoltaic Generation Based on Time-division Long Short-term Memory Neural Networks

It is significant for power system operation and dispatching to forecast photovoltaic (PV) power generation output accurately. Based on Spearman correlation coefficient analysis and time-division long short-term memory network, this paper proposed a new ultra short-term forecasting method for PV generation output. First, the Spearman correlation coefficient was utilized to analyze the factors influencing the PV power generation at every moment, and the influencing factors with high correlation degrees were selected as the input variables of the long short-term memory network. Secondly, for each moment a forecasting model based on the long short-term memory network was established to carry the forecasting of the output of time-division PV power generation. Finally, utilizing the data from historical generated power of actual PV power station and the data from numerical weather prediction (NWP) the case study was performed. Results of calculating examples show that using the proposed method, a more accurate forecasting can be obtained than the results obtained by only using the forecasting model based on the time-division long short-term memory network.