The aircraft landing phase is one of the most risky phases in each flight phase, and may encounter unsafe events such as gusts, runway incursions, hard landings, and runway overruns. A longer landing distance will increase the risk of unsafe events. In order to reduce the risk of unsafe incidents, this paper uses long short term memory (LSTM) neural network to capture the dependence of time series flight data on time, and studies a multi-step rolling prediction strategy to predict the aircraft landing distance for real-time early warning. The aircraft landing prediction model is used to predict the landing distance. The results show that compared with single-step prediction, this method can better capture the time variation of flight parameters. The accuracy and effectiveness of the multi-step rolling prediction method based on LSTM neural network model are verified by multiple sets of simulation experiments.