Ultra-Short-Term Power Prediction for Distributed Photovoltaic Systems Based on Meteorological Resource Interpolation

As a clean and sustainable energy solution, photovoltaic (PV) power generation has gradually become an indispensable part of the energy structure. However, PV power generation is significantly affected by factors such as solar radiation intensity and weather conditions, leading to fluctuations and uncertainties in its output power, which poses challenges to the stable operation of the power grid. Accurate prediction of PV power generation is crucial for improving the efficiency of solar energy utilization, reducing energy costs, and achieving a green and low-carbon lifestyle. This paper proposes a PV power prediction method that combines Variational Mode Decomposition (VMD) and Aquila Optimizer (AO) with Long Short-Term Memory (LSTM) networks. First, meteorological resource data and PV data are preprocessed, and the required meteorological factors are determined through correlation analysis. Then, ordinary Kriging interpolation is used to enhance the spatial resolution of meteorological data, calculating the meteorological data at the location of distributed PV power stations. Finally, VMD processes the PV data, and the AO-optimized LSTM model is used to predict the output power of PV systems. Compared with the AO-LSTM model using wavelet decomposition to process data and the AO-LSTM model without data processing, the proposed VMD-AO-LSTM achieves the highest PV power prediction accuracy in all four seasons.