Hard landing risk prediction of civil aircraft based on GBDT-GS method

Hard landing may cause structural damage of aircraft or other potential accident causes and even crash and fatal flight accidents. In view of the lack of physical nature analysis in current hard landing risk assessment, combined with flight status analysis, a hard landing risk prediction model based on gradient boosting decision tree (GBDT) and grid search (GS) was proposed to effectively implement hard landing risk identification and grade criteria and improve pilots’ landing operation quality. Firstly, the flight kinematics equation of landing was established through the force analysis of aircraft, and five flight status parameters closely related to hard landing were determined. Then, flight status data was extracted from onboard quick access recorder (QAR) data to construct a data set. According to QAR parameter characteristics, the hard landing risk prediction model was constructed by the GBDT algorithm, and model parameters were optimized by GS. Finally, taking the Chengdu-Shenyang route operation of an airline as an example, the study selected 530 pieces of QAR data to train and test the model and compared the result of the model with those of random forest, recurrent neural networks (RNN), and Logistic multiple regression. The results show that the GBDT-GS method is better than other algorithms in predicting hard landing risk, and its prediction accuracy reaches 92%, which verifies the objective validity of the constructed model.