Work of evaluating flight training quality of trainer aircraft pilot on left-boundary, faces practical needs of obvious subjective intention in manual judgment, in accurately locating focus of flight training, lack of practical design of current digital and intelligent means, and difficulty in meeting full process evaluation of flight training. Hence, with flight training outline requirements, typical flight parameters data were selected for left-boundary flight training subjects, and a-loop evaluation system of "action identification-risk identification-capability analysis-personal statistics" was constructed. Through DTW compound distance, kurtosis, skewness, 3D space coordinate system was established, and combined with clustering algorithm, flight action features were analyzed, and ultimately flight action recognition was realized. Flight risk identification was done using naive Bayes algorithm and maximum likelihood. Flight ability analysis and personality characteristic statistics were realized by flight deviation, deviation and kurtosis. At last, actual flight training effect of specific pilot was evaluated by an example of minimum speed level flight action, and effectiveness and practicability of proposed method was verified.