A ramp prediction model based on principal component analysis, time series decomposition and correction of long short-term memory (LSTM) network was proposed, in order to improve the accuracy of ramp prediction of new energy represented by wind power and photovoltaic power. In order to fully consider the time series characteristics of power, the power was decomposed into period, trend and residual by the time series decomposition method, and the trend and residual prediction model based on LSTM was established by combining the principal components of several characteristic factors, to realize the mapping relationship between the time characteristics of power and the principal components of influencing factors. Based on the preliminary prediction of trend and residual terms by LSTM, an error correction algorithm was introduced to calculate the dynamic error of the fitting prediction model and construct a new non-stationary time series to obtain the trend and residual predicted values with better accuracy. The final power prediction was obtained by fusing the trend, residual terms and the period value obtained by using the naive method. Combined with the definition of wind power and photovoltaic ramp event, the proposed model was used to predict the wind power and photovoltaic ramp event respectively. Experimental results show that the proposed model has better accuracy than other forecasting methods in direct power prediction and indirect ramp event prediction, and it can provide a more reliable basis for power grid dispatching.