Research On Flight Mode Recognition Method In Non Cruise Phase

In the context of energy saving and emission reduction in civil aviation industry,ICAO documents clearly indicate the need to review and evaluate the effectiveness of fuel saving in the climb and descent phases,so the effective identification of flight pattern data is of great importance for fuel analysis.However,traditional flight pattern identification only uses threshold processing,which is prone to large errors and poor generalizability,and the flight parameters of different aircraft vary.Based on this,this paper combined the feature parameter selection with the identification method,and used the feature parameter selection as the basis of the identification method.The main work of the thesis is summarized as follows.Firstly,a filter-encapsulation based RFE(recursive elimination)feature selection method was proposed to determine the appropriate input parameters for the recognition method.The information gain rate was calculated as a filter,and then the encapsulation method feature selection idea of recursive feature elimination was combined to effectively select the optimal feature combination.Besides,a DE-RF-RFE flight mode recognition method based on DE-RF-RFE was proposed.A differential evolutionary algorithm was used to automatically find the optimum for the random forest dual parameters.Then,a recursive feature elimination idea based flight mode recognition method was constructed using random forest as both the base classifier and feature importance ranking algorithm of the wrapper method.Next,a test sample set and 200 flight QAR data were used to validate the feasibility of the recognition method,and the problems of the recognition results were analyzed.Finally,for the DE-RF-RFE recognition method can only identify the status category of feature parameters from the numerical perspective,ignoring the time series characteristics of QAR parameters,the RF-RFE-LSTM based flight mode recognition method was proposed.The applicability of the LSTM algorithm was analyzed and elaborated.Then the RF-RFE-LSTM flight mode recognition method was constructed by directly drawing on the findings of feature parameter importance scoring in the DE-RF-RFE recognition method,using LSTM as the base classifier and RF as the feature importance ranking algorithm in the encapsulation method.Comparing and analyzing the experimental results of this method with DE-RF-RFE,LSTM and PCA-LSTM methods for flight mode recognition,the results show that this method has the best recognition effect among the above methods and improves the defects of DE-RF-RFE recognition method.