A Short-Term Power Prediction Method for Photovoltaic Power Generation Based on GRU-Transformer Model

To optimize the utilization of solar energy resources and ensure the power system's safety and stability during photovoltaic (PV) grid integration, precise forecasting of PV power generation becomes imperative. This paper addresses the low prediction accuracy stemming from the inability of existing methods to thoroughly analyze temporal information. It introduces a GRU-Transformer model that combines the Gate Recurrent Unit (GRU) and Transformer for enhanced short-term PV power prediction. The model initiates with a correlation analysis to identify significant factors influencing PV power generation and integrates various conditions affecting PV output into a unified embedding. To improve the prediction accuracy, the model adopts a dual-layer architecture. The initial layer employs GRU's gated units to distill essential temporal features from the data sequence. The subsequent layer utilizes the Transformer's self-attention mechanism to discern global information and contextual interrelations within the temporal data, pinpointing pivotal features for the current timeframe. This process refines the sequence data and augments the model's representational ability, leading to precise short-term PV power forecasts. Empirical validation using actual measurement data corroborates that the GRU-Transformer model surpasses existing methods in prediction accuracy.