Heavy landing of airplane is easy to cause airplane structural damage. Studying the prediction and assessment of heavy landing risks for airplane is very important and necessary for reducing the risk of heavy landing and improving the safety of civil aviation operations. A heavy landing risk prediction model is established by using QAR data and LSTM neural network. By calculating the probability density function of vertical acceleration, the possibility and severity of heavy landing are calculated to obtain the risk value; Based on the analysis of landing gear force during landing, vertical acceleration (landing load), descent rate, roll Angle, lateral acceleration and pitch Angle are selected as the influencing parameters of heavy landing. LSTM neural network is used to train the landing load of the flight, and the landing risk table is established. Through the parameter training of QAR data, the model is used to predict the landing load of the flight and verify its accuracy, and the risk of heavy landing is determined with reference to the risk grade table. The simulation results show that the RMSE and MAE of the predicted value and the actual value both reach the order of 10-3, and the quantitative heavy landing risk and heavy landing risk prediction are realized. The prediction model established by the research can provide theoretical basis for airplane landing safety risk management.