In recent years, large-scale photovoltaic(PV) grid connection has seriously affected the safe and stable operation of regional power grid.Ultra-short-term prediction of PV power can provide necessary data support for regional power dispatching and promote the realization of new energy consumption goal.However, the fluctuation characteristics of PV power make it difficult to improve the accuracy of power prediction.Therefore, PV power prediction model based on autoregressive integrated moving average(ARIMA) and support vector regression(SVR) considering power correction is proposed.Firstly, the ARIMA model is established using time series power data collected by PV power monitoring system, and preliminary prediction results can be obtained.Secondly, the prediction residuals of the previous meteorological similar day are used to establish SVR model to obtain the residuals of the prediction day.Finally, the preliminary prediction results are revised by prediction residuals.The typical PV power prediction models of different weather conditions are established by using the measured data.Test results show that the prediction accuracy is obviously improved after the residual error correction.