A novel causal graph attention network with convergent cross mapping for short-term photovoltaic power forecasting

Short-term photovoltaic power forecasting (SPPF) is vital for grid integration of renewable energy but remains challenging due to meteorological influences. While graph neural networks (GNNs) excel at capturing spatiotemporal correlations in SPPF, existing methods overlook causal relationships between weather factors and photovoltaic (PV) output, limiting accuracy. Thus, this paper proposes a causal graph attention network (CGAT) that employs convergent cross mapping (CCM) to calculate the causal intensity among factors related to photovoltaic power. Initially, we consider the causal relationship between PV power generation and meteorological factors, constructing a dynamic causal graph where photovoltaic power and meteorological factors are represented as nodes, and causal intensity serves as edge attributes. Subsequently, we employed graph attention network (GAT) and long short-term memory network (LSTM) to extract spatial and temporal characteristics between meteorological factors and photovoltaic power at each time stamp, thereby achieving accurate SPPF. Finally, the experimental results obtained from the actual operation of the photovoltaic power station show that under extremely challenging in summer, the forecasting performance of the proposed model is still superior to that of the existing state-of-the-art models, and its root mean square error (RMSE) has been reduced by up to 5.6%.