Distributed photovoltaic power prediction considering spatiotemporal correlation and dual Attention-LSTM

Predicting the power output of photovoltaic clusters is crucial for optimizing regional solar power scheduling. To enhance the accuracy of distributed photovoltaic station power forecasts, a method incorporating spatiotemporal correlation and dual Attention-LSTM is introduced. The K-means algorithm is first employed to classify the distributed photovoltaic power station clusters in the area. The reference station for the target photovoltaic station is determined by calculating the Euclidean distance between the target station and the typical daily power profiles of other stations in the cluster. Simultaneously, pivotal weather features that influence photovoltaic output are ascertained through computation of the Pearson correlation coefficient. Subsequently, an Attention-LSTM-based power prediction and error correction model is constructed, utilizing both meteorological and power traits as input variables to finalize the photovoltaic power generation forecast. To validate the approach, a simulation analysis is conducted using empirical data from Arizona, USA. The experimental results indicate that the suggested method greatly improves the accuracy of predictions for distributed photovoltaic power.