Improving the accuracy of photovoltaic power prediction is crucial for grid scheduling planning and is essential for the safe, stable, and economic operation of power systems. Based on the statistical characterization of the data, a two-stage PV power prediction model with error correction is developed. First, an Elman neural network model optimized by a small habitat genetic algorithm is introduced; subsequently, a more accurate model for the preliminary prediction error probability distribution is established, based on its distribution characteristics. This model aims to achieve error correction of the preliminary prediction results. The empirical results, derived from actual PV power curves and meteorological data, demonstrate the effectiveness of the proposed method.