乘客行为系统在地铁仿真培训中的研究与应用

发布时间：2018-09-07 16:03

【摘要】：人群仿真是利用人工智能、计算机图形处理等技术,建立人群行为模型,在虚拟环境中模拟人群行为,现已广泛用于公共交通仿真、大型场所人流分析、影视游戏作品等领域。乘客智能行为仿真是利用人群仿真技术,模拟乘客正常状况下进出站、乘离车等过程及紧急状况下疏散逃生过程。现有的乘客行为仿真大多集中于应急情况下公共交通场所的通过能力仿真,多采取离线、非实时计算方式进行,并不能很好地应用到对实时性要求很高的地铁仿真培训系统中。本文针地铁仿真培训系统对于乘客行为仿真在真实性、沉浸感、实时计算能力等方面的要求,提出一种基于地铁站台场景下的乘客智能行为模拟方案,建立乘客行为模型,实现单线路多车运行及多线路同时发车状况下小客流、中客流、大客流的乘客行为模拟,并应用到地铁仿真培训系统中。本文首先对人群感知模型进行研究。个体的感知由听觉、视觉、区域感知三个部分组成,为了表现个体在现实环境中的感知过程,本文建立由这三种感知方式共同组成的人群感知模型,对不同的感知信息分级处理,体现人群在不同感知效果下做出反应的差异性,较为真实地反应人类感知的特点。其次,本文研究了全局最优静态路径搜索算法和动态障碍避碰算法。采用基于导航网格的A*算法,实现个体最优静态路径搜索功能。在地铁站台场景中动态障碍物主要是其他乘客,根据该情况,提出一种基于权重因子的RVO避碰算法,实现乘客在运动过程对动态障碍物避碰的功能,以及遇到队列时的穿插和绕行两种行为。第三,本文研究了各种人群行为模型。对于基于全局和基于个体的人群模型,进行分析比较,其中基于全局的模型通过组、群将人群分类,可以模拟大规模的人群,但对个体的差异性表现不足。该模型广泛应用于公共场所大规模人群疏散及影视游戏作品中数万级别的人群效果演示。基于个体的人群模型中,个体具有特定的属性,每个个体的行为都单独计算,着重体现个体在状况处理和行为选择上的差异性,可用于复杂环境及多样化行为模拟。根据需求,提出基于行为特征、运动规律的乘客仿真模型,可以实现在不同目的及不同客流量下的人群行为模拟。第四,为了更真实地反应现实场景中乘客的行为,通过实地调查,对不同站台、不同时段、不同客流量下的乘客行为进行分析研究,总结乘客在行为选择过程中的规律,对影响乘客行为的因子进行量化分析,得到由随机因素、个体特征、环境信息组成的乘客行为选择公式,并提出基于该类规则和行为特征的乘客智能行为模拟方案。最后,本文研究了乘客紧急状况下的疏散行为。根据恐慌状态下的人群疏散特征,进行疏散模型验证,并在避碰计算中添加停止规则,解决了由于人群拥挤导致的模型抖动和身体穿插的问题。
[Abstract]:Crowd simulation is to use artificial intelligence, computer graphics processing technology, to establish crowd behavior model, in the virtual environment to simulate crowd behavior, has been widely used in public transport simulation, large-scale place flow analysis, video games and other fields. Passenger intelligent behavior simulation is to use crowd simulation technology to simulate the process of passengers entering and leaving station under normal condition and evacuation process in emergency. Most of the existing passenger behavior simulation is focused on the transit capacity simulation of public transport places in emergency situations, mostly using off-line, non-real-time computing method, which can not be well applied to the subway simulation training system with high real-time requirements. This paper presents a passenger intelligent behavior simulation scheme based on subway platform scene, which is based on the requirement of authenticity, immersion and real-time computing ability of passenger behavior simulation system in subway simulation training system, and establishes passenger behavior model. The passenger behavior simulation of small, medium and large passenger flow is realized under the condition of single line multi-vehicle operation and multi-line simultaneous departure, and it is applied to the subway simulation training system. This paper first studies the crowd perception model. Individual perception consists of three parts: hearing, vision and region perception. In order to express the perception process of individual in the real environment, this paper establishes a crowd perception model which is composed of these three ways of perception. The classification of different perceptual information reflects the difference of the response of the crowd under different perceptual effects, and reflects the characteristics of human perception more truthfully. Secondly, the global optimal static path search algorithm and dynamic obstacle avoidance algorithm are studied. The algorithm A * based on navigation grid is used to realize the individual optimal static path search function. In the scene of subway platform, dynamic obstacles are mainly other passengers. According to this situation, a RVO collision avoidance algorithm based on weight factor is proposed to realize the function of passengers avoiding dynamic obstacles in the course of motion. As well as encounter queue interlude and detour two kinds of behavior. Thirdly, this paper studies various crowd behavior models. For the global and individual-based population models, the analysis and comparison, which based on the global model through groups, groups will be classified, can simulate large populations, but the differences of individual performance is not enough. The model is widely used in mass evacuation of public places and tens of thousands of crowd effect demonstrations in video games. In the individual-based population model, the individual has a specific attribute, and each individual's behavior is calculated separately, which focuses on the difference of the individual's situation treatment and behavior choice, and can be used to simulate the complex environment and diversified behavior. According to the demand, a passenger simulation model based on behavior characteristics and motion rules is proposed, which can be used to simulate crowd behavior under different objectives and different passenger flow. Fourthly, in order to reflect the behavior of the passengers in the real scene more realistically, through the field investigation, the passenger behavior under different platforms, different time periods and different passenger flow is analyzed and studied, and the rules of passengers' behavior selection are summarized. Based on the quantitative analysis of the factors affecting passenger behavior, a passenger behavior selection formula composed of random factors, individual characteristics and environmental information is obtained, and a passenger intelligent behavior simulation scheme based on this kind of rules and behavior characteristics is proposed. Finally, the evacuation behavior of passengers in an emergency is studied. According to the characteristics of crowd evacuation in panic state, the evacuation model is verified and the stopping rule is added to the calculation of collision avoidance, which solves the problem of model jitter and body insertion caused by crowd congestion.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.9

【参考文献】