Pilot Safety Performance Evaluation Based On Operational Data

The safety performance evaluation of pilots by airlines is mainly based on the identification of unsafe events,and the over-the-limit monitoring in flight quality monitoring is one of the important means to evaluate the safety performance of pilots.However,it is easy for pilots to regard "non-over-the-limit data" as their flight criteria and ignore the original intention of flight quality monitoring.This practice has potential safety risks and is not conducive to objective evaluation by airlines.Therefore,from the perspective of data mining,combined with the actual flight operation data,pilot safety performance can be quantitatively evaluated.First,determine the pilot safety performance evaluation index system.Based on the perspective of data mining,combined with the overlimit monitoring standard and the standard operation manual,this paper summarizes the initial index system of 18 flight parameters which are focused on during the approach landing stage,and analyzes the characteristics of these parameters in the approach landing process from the perspective of pilot technical skills.Then the initial index system is reduced by correlation-principal component analysis to reduce the computational complexity.Finally,the pilot safety performance evaluation index is obtained,which consists of eight parameters: radial acceleration,pitch rate,heading,radar high,pitch angle,glide angle,throttle angle and glide deviation.Secondly,the deviation degree of pilot’s distance standard under the performance index system is calculated.Using the mean value of approach landing data fitting of 179 flights as the standard,the similarity of the multivariate time series composed of the 8 parameters was calculated.Using the method of pattern representation,the dimension of the multivariate time series was reduced to the unitary time series represented by the norm.QAR data of 10 pilots were randomly selected.The Euclidean distance,DTW distance and their weighted distance were used to calculate the similar distance between different pilots and the standard sequence,and the rank of the deviation degree between each pilot and the standard was obtained.Finally,a pilot safety performance evaluation model based on time series deviation degree and deviation frequency is established.LOF algorithm is used to determine the pilot point set similar distance discrete points,with corresponding local outlier factor of discrete points as modified index of deviation degree,get the pilot risk index,statistical distribution as the pilot safety performance divided interval,get the level of safety performance evaluation of the pilot,realizes the quantitative evaluation of the pilot safety performance level.