基于社交隐式模型的运输类飞机人员疏散轨迹预测  

Evacuation trajectory prediction of passengers in transport aircraft based on social-implicit model

作　　者：陈琨[1] 李放 冯振宇[1] 陈向明[3] 段龙坤 CHEN Kun;LI Fang;FENG Zhen-yu;CHEN Xiang-ming;DUAN Long-kun(Science and Technology Innovation Research Institute,Civil Aviation University of China,Tianjin 300300,China;College of Aeronautical Engineering,Civil Aviation University of China,Tianjin 300300,China;Aircraft Strength Research Institute of China,Xi'an 710065,Shaanxi,China;College of Safety Science and Engineering,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学科技创新研究院,天津300300 [2]中国民航大学航空工程学院,天津300300 [3]中国飞机强度研究所,陕西西安710065 [4]中国民航大学安全科学与工程学院,天津300300

出　　处：《交通运输工程学报》2024年第5期270-284,共15页Journal of Traffic and Transportation Engineering

基　　金：国家重点研发计划(2022YFB4301000);中央高校基本科研业务费专项资金项目(3122017085)。

摘　　要：为精准预测运输类飞机在狭窄空间内紧急情况下的乘客疏散轨迹,构建了基于深度学习的社交隐式(Social-Implicit)模型,该模型包括社交区域、社交神经元和社交损失3个模块,通过轨迹聚类和卷积神经网络处理乘客在不同状态下的行为变化和动态交互;基于波音737-800模拟舱应急撤离试验,采集了乘客的运动状态与冲突行为数据,结合社会力模型参数标定生成了训练数据集,利用该数据集对Social-Implicit模型进行训练和验证,并从平均位移误差(ADE)、最终位移误差(FDE)、平均马氏距离(AMD)和平均最大特征值(AME)四方面评估了模型的预测效果。分析结果表明:Social-Implicit模型在构建的应急撤离数据集上表现良好,ADE为0.011 m,FDE为0.02 m,相比苏黎世联邦理工学院行人数据集和塞浦路斯大学行人数据集均降低了97%,AMD和AME精度分别提高了72.4%和94.1%,说明模型在狭窄环境中捕捉乘客轨迹变化方面表现优异;该模型的撤离时间、路径、速度和瓶颈位置预测都与社会力模型结果非常接近,疏散效率分别为1.92和1.93,路径预测能准确捕捉人员卡塞位置,速度误差不超过0.02 m·s^(-1),瓶颈位置误差在每平方米0.41人以内,表明模型能够模拟乘客疏散中拥堵特征;与社会力模型相比,基于深度学习的Social-Implicit模型的运行时间从65.000 s显著缩短至0.021 s,模型内存减小了78.02%,因此,Social-Implicit模型能够为民航应急疏散系统的优化提供高效、准确的轨迹预测与性能评估方法。To accurately predict passenger evacuation trajectories in narrow spaces during emergencies of transport aircraft,a social-implicit model based on deep learning was developed.The model included three modules:social zone,social cell,and social loss.Passenger behavior changes and dynamic interactions under different conditions were processed through trajectory clustering and convolutional neural networks.The passenger movement and conflict behavior data were collected from Boeing 737-800 simulation cabin emergency evacuation experiments.These data combined with social force model parameter calibration were used to generate the training dataset.The social-implicit model was trained and validated by using the dataset,and its predictive performance was evaluated in terms of average displacement error(ADE),final displacement error(FDE),average mahalanobis distance(AMD),and average maximum eigenvalue(AME).Analysis results show that the social-implicit model performs well on the constructed emergency evacuation dataset,with an ADE of 0.011 m and an FDE of 0.02 m,representing a 97% reduction compared to the ETH BIWI walking pedestrians dataset and University of Cyprus pedestrian dataset pedestrian datasets.The accuracies of AMD and AMV improve by 72.4% and 94.1%,respectively,indicating that the model excels at capturing passenger trajectory change in narrow environment.In terms of evacuation time,path,speed,and bottleneck position prediction,the model closely aligns with the results of social force model,with evacuation efficiencies of 1.92 and 1.93,respectively.The path prediction can accurately capture the congestion points,with the speed error no exceeding 0.02 m·s^(-1),and the bottleneck position error within 0.41 persons per square meter,demonstrating that the model can simulate congestion characteristics during the passenger evacuation.Compared to the social force model,the deep learning model's runtime significantly reduces from 65 s to 0.021 s,and the model size reduces by 78.02%.The social-implicit model provide

关 键 词：民用航空 应急撤离 深度学习 轨迹预测 社交隐式模型 

分 类 号：V328[航空宇航科学与技术—人机与环境工程]