Hierarchical gated pooling and progressive feature fusion for short-term PV power forecasting

Abstract

In this paper, we propose a hierarchical gated pooling and progressive feature fusion model (HGP-PFF) for short-term photovoltaic (PV) power forecasting. HGP-PFF effectively overcomes the limitations of existing methods in multi-scale feature extraction and fusion by introducing a hierarchical gated pooling (HGP) module and a progressive feature fusion (PFF) module. This model replaces traditional convolution operations with a pooling gate mechanism for feature extraction, efficiently capturing features across different time scales. HGP-PFF also employs a PFF module to ensure the completeness and consistency of the fused feature information. The proposed HGP-PFF model is applied to three different PV power datasets collected from the Alice Springs PV power station. Compared to previous state-of-the-art (SOTA) models the proposed HGP-PFF model reduces the PV power forecasting error by more than 19.57%, 22.27% and 13.67% on these three PV power datasets.