Intermittent solar power hybrid forecasting system based on pattern recognition and feature extraction

Solar energy, with its abundance and accessibility, occupies an irreplaceable position in the shift in global energy consumption patterns. The difficulties of managing solar energy on the grid, caused by its highly volatile and intermittent nature, necessitate an accurate and stable forecasting system. However, existing studies have focused more on the accuracy of prediction without considering the reactive power caused by intermittency, whose prevalence leads to the solar power series having both discrete and continuous characteristics, making the prediction problem more challenging. To fill this gap, a multistep ahead hybrid forecasting system has been constructed in this study that contains feature extraction, pattern recognition, forecasting, and integrated optimization modules. The system integrates pattern recognition algorithms into the prediction models to appropriately deal with the inconsistent data features caused by intermittency and introduces a data decomposition strategy to achieve feature extraction, such that it could be adapted to forecast situations where relevant weather information is not available. Finally, the multi-objective optimization algorithm allows selecting, weighting, and integrating all submodels and yields a hybrid model with excellent accuracy and stability. The results showed that the hybrid forecasting system achieved the best performance in all aspects of the four forecast scenarios and also performed well in the multistep advance forecast. Taking Site 1 as an example, the mean absolute scale errors of the one-step, two-step, and three-step advance forecasts are 2.9953%, 3.7095%, and 3.4945%, respectively; and the standard deviations of the errors are 3.1086, 3.6210, and 3.2133, respectively. Furthermore, the hybrid forecasting system achieves a performance improvement in accuracy and stability of more than 90%, and this performance improvement is significant at the 10% significance level.