Research On Abnormal Track Detection In Terminal Airspace Based On Multivariate

With the development of the civil aviation industry,the number of civil aviation flights is increasing,the air traffic is becoming more and more congested,and the difficulty for the controllers to deploy flights is increasing.Accurately and effectively excavate the abnormal flight behavior of the aircraft,which can assist the controller to allocate flights and ensure the safety of civil aviation.The airport terminal area is the most complex flight area in the air traffic system.The flight in the terminal area is dense.Due to problems such as controller deployment,abnormal flight paths may be generated.The flight with abnormal behavior may increase the operating costs of the airline,affect the satisfaction of passengers on this trip,and the extremely uncertain landing time will also increase the labor burden of the airport staff.Therefore,in order to improve the efficiency of civil aviation operation,effective methods are needed to detect abnormal tracks.In this thesis,the abnormal track detection algorithm in the terminal area is mainly used to detect flights arriving from a fixed entry point within a period of time.First of all,the composition and structure of the terminal area of the airport are analyzed,and the causes and characteristics of the abnormal tracks in the terminal area are introduced.The Shenzhen airport in the Pearl River Delta terminal area is taken as an example for practical analysis.Secondly,this thesis comprehensively considers the airport factors,airspace factors,meteorological factors,flight factors and other co-variable factors that affect the track of the terminal area,and explains how these factors affect the track.The historical data composed of multiple variables provides a data basis for realizing the track reconstruction.However,in the process of sending,transmitting and receiving flight data,many uncertain factors will affect the data quality,so the data should be cleaned before training the model.Then,the cleaned normal track data is reconstructed through the TCN model,that is,the longitude,latitude and altitude data of the next moment are reconstructed from the multi-dimensional data of the previous several moments in the historical data,and the mean square error of the reconstructed track data and the track data of the original sequence is used to train the XGBoost classifier to achieve the abnormal track classification.The traditional machine learning model can only get the output results according to the input,but can’t explain how the results are generated.In order to enhance the practicability of the model,this thesis adds SHAP(Shapley Additive Expansions)machine learning interpretability model to explain the reason why the track is detected and predicted as abnormal track while using XGBoost model to classify the anomaly.In addition,in order to reflect the advantages of multivariate anomaly detection,the anomaly detection results using only the data of longitude and latitude height are compared with those using multivariate data,and the results show that the accuracy of multivariate anomaly detection is higher.Finally,this thesis uses the real flight data of Shenzhen Airport to verify the effectiveness of the model.