Enhanced day-ahead multi-step PV power forecasting using informer-ensemble learning

With the increasing integration of renewable energy into power systems, accurate photovoltaic (PV) power forecasting has become crucial for maintaining grid stability, optimizing energy storage, and supporting electricity market operations. This paper proposes a novel day-ahead PV power forecasting model based on an ensemble of Informer networks to improve multi-step prediction accuracy. Influential factors affecting PV output are identified using the Pearson correlation coefficient for effective feature selection. K-means clustering is employed to group similar daily PV output patterns, improving the time series representation, while a Gaussian filter is applied to smooth the dcata, reduce noise, and accelerate model convergence. The proposed ensemble Informer model is evaluated using real-world datasets across various time resolutions and forecasting horizons. Comparison results demonstrate that the model outperforms conventional machine learning approaches, achieving a minimum root mean square error (RMSE) of 0.029 for 24-h forecasts. These results underscore the model’s high accuracy and low error, showcasing its potential to improve the reliability of PV forecasting and support more efficient day-ahead energy scheduling with solar power integration.