Amplify seasonality, prioritize meteorological: Strengthening seasonal correlation in photovoltaic forecasting with dual-layer hierarchical attention

Abstract

Overloading beyond the grid’s capacity poses a serious threat to grid security. In 2023, photovoltaic power generation accounted for 75 % of the total increase in renewable energy generation. However, due to the significant fluctuations in photovoltaic power output, forecasting photovoltaic generation has become a crucial tool for ensuring grid security. A key challenge in practical applications remains the deep mining of hidden features in photovoltaic data and their correlation with meteorological data to improve prediction accuracy. To address this, this study proposes a photovoltaic prediction strategy called “Amplify Seasonality, Prioritize Meteorological". This strategy aims to leverage meteorological information to connect with the seasonal component of photovoltaic power data while preventing meteorological factors from affecting the trend component, thereby effectively reducing the impact of short-term seasonal meteorological fluctuations on the trend component of photovoltaic data. Additionally, this study proposes a seasonal component prediction unit with a dual-layer hierarchical attention mechanism, which enhances the focus on the connections between meteorological features, key time nodes, and the seasonal component. These innovations enable the proposed AspmNet model to achieve superior prediction accuracy. The model was validated using Australian photovoltaic data through experiments with forecast lengths of 1 day, 2 days, and 4 days. In terms of Mean Absolute Error, the model demonstrated over a 10 % improvement compared to other benchmark models.