Photovoltaic power interval prediction based on attention mechanism and quantile regression - convolutional neural network - bidirectional gated recurrent unit

Accurate photovoltaic (PV) power interval prediction allows effective quantification of PV power prediction uncertainty, which is critical for power system operation and scheduling. To strengthen the precision of PV power interval forecasting, this study presents a PV power interval forecasting method that employs an attention mechanism and a convolutional neural network (CNN) with bidirectional gated recurrent units (BiGRU). Firstly, the meteorological factors that are significantly associated with the PV power are selected based on the Pearson correlation coefficient; Second, a composite deep learning prediction model is established by leveraging the feature extraction capabilities of convolutional networks and the temporal feature modeling capabilities of bidirectional gated recurrent unit networks, in conjunction with attention mechanisms and quantile regression (QR); Lastly, the PV power prediction interval at a certain confidence level is obtained using quantile regression, and the arithmetic example is tested using data from a PV power plant in Xinjiang, China, confirming the superiority of the approach presented in this research.