Abstract:Aiming at the problem of missing data in distributed PV power prediction, the study proposes an innovative cluster accumulation method. Firstly, through the K-mean clustering algorithm, based on the forward power and reverse power of PV power generation customers, distributed PV power plants are categorized into different clusters, forming multiple sub-clusters with similar power generation characteristics. Then, for each sub-cluster, a Long Short-Term Memory (LSTM) neural network model is used to predict the average power over a specific time period in the future. By accumulating the predicted power values of each sub-cluster, the total predicted power of the whole PV plant cluster is finally obtained. The experimental result shows that the cluster accumulation method not only effectively solves the efficiency and accuracy problems in distributed PV prediction, but also performs well in terms of mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE). At the same time, it has a coefficient of determination (R2) close to 1, showing the strong fitting ability and generalization performance of the model. This research provides a new technical approach to improve the management efficiency of distributed PV systems.