Based on CNN-LSTM for photovoltaic output prediction of multi-scale method

Abstract To maximize power generating efficiency, accurate forecast of generation power is the very important job for the photovoltaic generation system. In this paper, photovoltaic production is forecasted using a understanding way called multi-scale based on convolutional neural network-long short-term memory network (CNN-LSTM) for photovoltaic output prediction of multi-scale method. However, existing methods are often challenged by the multi-scale nature of the data when dealing with PV output prediction, so a more efficient method is needed to overcome difficulties. Therefore, this paper put forward an idea of a multi-scale method based on CNN-LSTM, which combines CNN and LSTM to accurately predict photovoltaic output at different time scales. The CNN-LSTM model can achieve the precision of 0.9097 for 15 minutes, the prediction accuracy of 0.8750 for 1 hour, and the prediction accuracy of 0.7987 for 4 hours, which can prove that the model can meet the scheduling rules.