Very Short-Term Solar Generation Forecasting Based on LSTM with Temporal Attention Mechanism

Abstract

Accuracy solar generation forecasting could avoid serious challenges to large scale PV grid-connected systems. Thus, a very short-term solar generation forecasting method based on the LSTM with the temporal attention mechanism (TA-LSTM) is proposed in this paper. In our method, partial autocorrelation is first utilized to determine the length of time series, which is used as input of the LSTM forecasting model. Then, the TA-LSTM is trained by the data to learn the forecasting model. The LSTM is used here to learn the forecasting model because it can make full use of the information of the past time and has stronger adaptability in time series data analysis. To further improve forecasting accuracy, the temporal attention mechanism is integrated into the LSTM prediction model. The experiments are carried out to verify the performance of the proposed method. The experimental results show that the proposed method is feasible and effective.