In order to solve the problem that the quick access recorder (QAR) data contains a lot of noise, which affects the flight data analysis, a network based on long short-term memory (LSHM) and Kalman filtering (KF) is proposed. In this paper, the data is preprocessed by using the Leida criterion, and the state equation of the model is established based on LSTM, and the real-time online estimation of QAR data is carried out by combining with the Kalman filter, and the simulation experiment is carried out by using the flight data of the domestic ARJ21 aircraft. The results show that the adaptability of the proposed method to real-time data is better than that of the LSTM method, the dependence on the dynamic model is less than that of the traditional filtering method, and the proposed method has higher accuracy and better noise reduction effect for QAR data.