Research on Ultra-Short-Term Photovoltaic Power Forecasting Method Considering Complex Meteorological Factors

Complex meteorological factors pose challenges to photovoltaic power forecasting. In response, this paper proposes an ultra-short-term photovoltaic power forecasting method tailored for the complex meteorological environment. We introduce a method for decomposing complex meteorological factors, breaking down the time series of meteorological data into multiple modal components and trend terms. To analyze the correlations among these sub-components, we construct a dynamic graph to extract coupling features. Furthermore, we propose an improved ISAM-Trans-GNN hybrid neural network model to learn from the dynamic graph features. Experimental results demonstrate that our approach optimizes the model's selfattention mechanism, aggregating node features, edge features, and graph structural information, thereby providing better support for accurate ultra-short-term photovoltaic power predictions.