Research On The Key Methods For Machine Analysis Of ATC Unsafe Incident Report

Civil aviation unsafe event report is one of the important ways to record the information of aircraft accidents,signs of accidents and other unsafe events.It is also one of the important basis for discovering and correcting the defects of the current system.However,there are few researches on the information mining of civil aviation unsafe incident report.In order to be able to more in-depth mining safety information from the civil aviation unsafe incident report,this paper took the aircraft runway excursion on the runway/taxiway as an example,and uses civil aviation unsafe incident report as data source.A series of steps such as human error quantification,topic modeling,and word vector calculation were used to conduct research on information mining of civil aviation unsafe incident reports.Firstly,based on the STPA method and the BP neural network method,the human error in the scene was identified by applying the STPA method,and the human error fishbone diagram in the scene was constructed.The human error fishbone diagram was used as the basis for constructing the BP neural network,with nearly five years under the conditions of the ASRS system in the scenario of unsafe incident report as a data source,through to the BP neural network training and testing,the BP neural network model in the scenario was obtained,and the weight of the model was calculated.The mapping relationship between different occurrence stages and different human errors in the scene was obtained.Secondly,based on the LDA theme model and the Word2 Vec word vector model,the application of the LDA topic model was used to analyze the document-topic distribution and topic-word distribution of the civil aviation unsafe incident report in this scenario,and then used the Word2 Vec word vector model to mine the mapping between topic and lable of cause factors.Thus the mapping relationship between the document and the label was constructed,and the function of identifying the cause of the civil unsafe event report in this scenario was realized.The results show that,based on the civil aviation unsafe incident report,by combining STPA with BP neural network method,LDA theme model and Word2 Vec word vector method,the potential information in civil aviation unsafe incident report can be effectively explored,which can provide an idea for research on civil aviation unsafe report.