An Adaptive Pattern Prediction Framework for Solar PV Power Forecasting Using Cloud Image

The global energy crisis has hastened the development and adoption of renewable energy sources. The rapid expansion of photovoltaic (PV) installed capacity exacerbates the challenge of unit forecasting errors, particularly amidst instantaneous irradiance fluctuations caused by cloud movement. To address the limitations of individual models in accurately characterizing power fluctuations, this paper proposes a pattern prediction framework based on cloud image for refined enhancement of PV power forecasting. The framework incorporates a classifier for adaptive pattern classification of cloud images, aligning them with corresponding PV power periods, and a predictor comprising two models: an ultra-short-term PV power forecasting model and a pattern recognition model. Simulation results demonstrate the framework's ability to accurately predict target period patterns, thereby enhancing forecasting precision, thus providing crucial technical support for constructing a secure and manageable new power system.