A Photovoltaic Power Generation Prediction Method Combining STL Decomposition and Attention Mechanism

As global demand for renewable energy increases, photovoltaic (PV) power generation has become crucial for grid management and energy optimization. This paper introduces a novel PV power forecasting method that combines Seasonal-Trend Decomposition using LOESS (STL) with a Self-Attention Mechanism (STL-SA). This approach utilizes STL to break down time series data, enhancing the model's capacity to identify intrinsic photovoltaic patterns, while the self-attention mecha-nism focuses on learning long-term dependencies within the time series. Experimental validation on five public datasets shows that the STL-SA method excels in multivariate multi-step forecasting tasks, delivering remarkable results with real photovoltaic data.