In order to improve the accuracy of photovoltaic power prediction and reduce the impact of photovoltaic power generation on grid scheduling, a short-term photovoltaic power forecasting model based on similar day clustering, PCC-EEMD-BES-ELM, was proposed. The PCC (Pearson correlation coefficient) was used to select the main factors as the inputs for the forecasting model. EEMD(ensemble empirical mode decomposition) was employed to extract multi-level features of photovoltaic power generation, BES (bald eagle search) effectively selected important features and optimizes the model, while the optimized ELM (extreme learning machine) performs efficient predictions through its fast learning mechanism, forming a robust model capable of providing high-accuracy forecasts in complex environments. Finally, the similar day clustering method was used to categorize weather into sunny, rainy, and cloudy days for power generation forecasting. The model was validated using real photovoltaic station data, and the results show that the minimum root mean square error (RMSE) reaches 0.048 1, the mean absolute error (MAE) is 0.035 9, and the mean absolute percentage error(MAPE) is 0.947%. The model demonstrates the lowest forecasting errors under various seasons and weather conditions, proving its effectiveness and stability in short-term photovoltaic power prediction.