Short-term power dynamic prediction method for high-density distributed PV clusters

Accurate prediction of the power generation of regional distributed photovoltaic (PV) systems under complex operating conditions is of great significance for enhancing PV accommodation and hosting capacity, as well as optimizing regional power resource dispatching. To address the limitations of poor representativeness in single-point prediction and feature dilution in global prediction for regions with high-density distributed PV systems, this paper proposes a regional distributed PV cluster power prediction strategy based on weighted dynamic time warping (WDTW) and the S-Transformer model. This method employs dynamic time series distance with Gaussian weights to analyze the output characteristics of different PV systems, grouping distributed PV units with similar output characteristics into the same sub-cluster. The clustering results are then used as features and input into the S-Transformer model with a seasonal module for regional distributed PV power prediction. Case study results demonstrate that, compared with traditional prediction models, the proposed improved clustering method and prediction network achieve higher prediction accuracy.