Research on short-term photovoltaic power point-interval prediction method based on multi-scale similar day and EVO-TABiGRU

Abstract Photovoltaic (PV) power generation, known for its environmental benefits and renewability, plays a critical role in advancing sustainable energy. However, the inherent randomness and volatility of PV generation challenge the stable operation of power systems with high PV penetration. Accurate PV power prediction is essential for ensuring safe grid integration and reliable power system operation. This study introduces an advanced short-term PV power prediction framework, combining multi-scale similar days (MSSD) selection and a trend-aware bidirectional gated recurrent unit (TABiGRU). First, MSSD is employed to select historical data with meteorological conditions similar to the predicted day as training samples, reducing the impact of meteorological randomness on the model. Then, to enhance the model’s ability to capture the trends in meteorological dynamics, a TABiGRU model is proposed, which introduces meteorological change rate features and dynamic weight adjustment to improve the model’s adaptability to meteorological fluctuations. In addition, an energy valley optimization algorithm is used to tune the hyperparameters of TABiGRU, preventing performance degradation due to improper parameter settings. Furthermore, to mitigate the cumulative error issue of point prediction under uncertain meteorological conditions, adaptive bandwidth kernel density estimation is used to generate high-quality prediction intervals, providing more robust decision support for power system scheduling. Finally, experimental results demonstrate that the proposed method achieves high prediction accuracy and stability under various meteorological conditions, particularly showing significant advantages in complex meteorological fluctuation scenarios, providing strong support for the safe and stable operation of the power grid.