Classification Of Behavioral Characteristics In Aircraft Emergency Evacuation Process Based On Deeplearning

Emergency evacuation of civil aircraft is an important means for passengers and the crew to evacuate from the aircraft after an accident.It can prevent passengers suffering from secondary damage if the aircraft is burned or exploded again.For airplanes having a seating capacity of more than 44 passengers,the airworthiness regulation clearly requires the use of a full-scale demonstration test to demonstrate the conformity of the aircraft's emergency evacuation capability.However,there are many shortcomings in the demonstration test,such as the cost is very high,volunteers are vulnerable,and the test can't simulate real disaster scenes.In recent years,with the rapid development of computer technology,using computer simulation models has become an important means to study the characteristics of passengers' behavior in the emergency evacuation process.At present,the research on behavioral characteristics mainly focuses on large buildings such as stadiums,transportation hubs,shopping malls,etc.The scene of emergency evacuation of civil aircraft has the characteristics of narrow space,numerous obstacles and dense personnel,different from large buildings.The behavioral characteristics under the scene of large buildings are difficult to reflect in the process of emergency evacuation of the civil aircraft.At present,the evacuation time is an important evaluation index for the effectiveness of the civil aircraft emergency evacuation simulation model,but it can not indicate the effectiveness of the simulation model.Therefore,how to extract the behavior characteristics of the civil aircraft emergency evacuation process is the key to evaluating the effectiveness of the simulation model.In this thesis,an emergency evacuation simulation test,referring to the requirements of airworthiness regulations and the cabin configuration of the Boeing 737-200 plane front section,is designed to obtain test data under the cooperation mode and the competition mode.Two different reward methods are designed to make the testers to exhibit different behavioral characteristics.Through analyzing the test process,it was found that the testers had different behavioral characteristics in the two modes.The test data is converted into image information form by referring to the data processing method in the video classification task.Convolutional neural networks and long-term and short-term memory neural networks were used to construct a deep learning classification network,used to extract and classify behavioralfeatures under two modes.The classification results show that the testers have different behavioral characteristics in different mode,and have same behavior characteristics in the different aisle width scenes of the same mode.The simulation model was validated and analyzed by using evacuation time,velocity-density relationship and deep learning.The results show that the evacuation time and velocity-density can only explain the effectiveness of the simulation model to some extent.The effectiveness of the simulation model needs to be verified by more behavioral features.Deep learning can provide a more reliable method for verifying the validity of the simulation model.