WNPS-LSTM-Informer: A Hybrid Stacking model for medium-term photovoltaic power forecasting with ranked feature selection

With the increasing adoption of photovoltaic (PV) technology in smart grids, accurate photovoltaic power forecasting (PVPF) is critical to grid stability. However, due to the stochastic nature of solar energy and the neglect of cyclical and trending patterns of PV power generation, existing methods struggle in mid-term forecasting. To address these challenges, this paper proposes a hybrid model, WNPS-LSTM-Informer, for medium-term PVPF based on Stacking ensemble algorithm. First, this paper optimizes the neural prophet (NP) by introducing the genetic and mutation operations of the genetic algorithm (GA) in the white shark optimizer (WSO). The NP is initially predicted and combined with meteorological data, and the relevant features are filtered to improve robustness using the ranked feature selection (RFS) algorithm proposed in this paper. The data are then fed into the Stacking model, which integrates the long short-term memory (LSTM) and the Informer model to handle medium-term data more efficiently than a single model. In this paper, the model is validated using historical data from the photovoltaic system in Uluru, Australia, and compared with five state-of-the-art forecasting models at different time scales and seasons. The results show that WNPS-LSTM-Informer outperforms existing methods in medium-term forecasting.