A Regional Distributed Photovoltaic Power Forecasting Method Based on Cluster Division and Selection of Representative Plants

As the proportion of distributed photovoltaic (DPV) power generation in the energy structure increases, accurate forecasting of its power output is crucial for ensuring the stability and reliability of the power grid. The crux of DPV power forecasting lies in the effective division of DPV plant clusters and the selection of representative plants. To address these issues, the geographical location distribution information and power characteristics of DPV plants are utilized for cluster division to ensure that the power characteristics of DPV plants within the clusters are similar. Following this, the maximum difference algorithm is used to identify representative plants from each cluster, thereby reducing calculational load and enhancing forecasting efficiency. Subsequently, a Convolutional Neural Network (CNN)- Bidirectional Gated Recurrent Unit model (BiGRU) is constructed, which combines meteorological data and historical power data, to forecast the power of selected representative plants, and then aggregates these forecasts to get the overall forecasting results for the region. This model leverages the strengths of CNN in capturing spatial features and BiGRU in capturing temporal dynamics, thereby significantly improving forecasting accuracy compared to traditional methods. The proposed method demonstrated a high coefficient of determination (R² > 0.91) across all four seasons, highlighting its superior forecasting performance. Compared to CNN-GRU, the proposed CNN-BiGRU model achieves higher accuracy of 4.5%. The main innovation of this paper is the systematic division of regional DPV plants cluster and the selection of representative plants. This approach offers an efficient and dependable technical solution for the power forecasting of DPV plants, advancing the field with its innovative methodology.