發(fā)布時間：2018-07-05 18:28

  本文選題：人群行為仿真 + 多智能體系統(tǒng)�。� 參考：《國防科學技術(shù)大學》2014年博士論文

【摘要】：隨著社會不安定因素的增多,突發(fā)安全事件特別是國內(nèi)群體性事件的頻繁發(fā)生給社會造成了巨大破壞和人員傷亡。在此背景之下,提高應對突發(fā)事件的能力對于保障公共安全具有重要意義。利用計算機仿真技術(shù)對人群行為進行仿真研究,從而為具有軍事應用背景的人群管理控制決策提供技術(shù)支持,已經(jīng)成為近年來的研究熱點。論文以群體性事件作為代表性的軍事應用背景,針對當前虛擬人群行為仿真中面臨的行為模型真實性差、單一模型不能夠真實模擬人群行為現(xiàn)象、人群行為仿真框架適用性差等問題,兼顧群體性事件對人群行為模型的實際需求。分別從人群行為仿真框架設計、獨立個體的導航行為、各種類型的組行為生成、群體性事件中的組控制行為等方面,開展虛擬人群行為仿真關鍵技術(shù)研究,提出了相應的控制策略和模型,并對傳統(tǒng)的算法進行了改進,主要研究內(nèi)容概括如下：一、論文首先對虛擬人群行為建模相關技術(shù)及應用的國內(nèi)外研究現(xiàn)狀進行了概述,對不同方法的優(yōu)缺點進行了分析、比較,提出以群體性事件為應用背景開展人群行為仿真研究,將群體性事件中的人群分為平民和士兵,其中平民又由獨立個體和組構(gòu)成。從而提出分別對人群中的不同構(gòu)成成分進行行為建模,以期建立逼真的人群行為模型。二、對于人群行為仿真框架設計問題,在多智能體體系結(jié)構(gòu)的基礎上,將基于行為的智能體體系結(jié)構(gòu)和慎思式智能體體系結(jié)構(gòu)相結(jié)合,提出一種行為／認知混合式體系結(jié)構(gòu)。此體系結(jié)構(gòu)不僅能設計明顯不同于反應式規(guī)則的、復雜的單個行為,還能夠通過在行為模塊中加入一定程度的規(guī)劃從而生成更復雜的行為模塊,因此能對非結(jié)構(gòu)化的環(huán)境進行很好的處理,實現(xiàn)復雜的高層規(guī)劃和對環(huán)境變化的快速反應二者優(yōu)勢互補。行為/認知混合式控制結(jié)構(gòu)由于建立在基于行為的系統(tǒng)設計框架之上,因此具有比慎思/反應式混合體系結(jié)構(gòu)或單純采用基于行為的結(jié)構(gòu)更大的優(yōu)勢,也更適合于對人群行為的仿真建模。同時在我們設計的體系結(jié)構(gòu)中,實現(xiàn)了全局規(guī)劃和局部行為的結(jié)合、宏觀行為和微觀行為的結(jié)合、數(shù)據(jù)驅(qū)動方法和模型驅(qū)動方法的結(jié)合,從而能夠支撐產(chǎn)生更為真實的人群行為模型,也極大地提高了人群行為仿真框架的適應性和可擴展性。三、對于獨立個體的導航行為生成問題,為提高虛擬人運動軌跡的逼真自然性,必須使虛擬人的軌跡滿足平滑性、多樣性以及具有最大速度和加速度約束。首先對PRM (probabilistic roadmap)算法的采樣策略進行了適應性修改,提出將非均勻采樣和bridge-test采樣策略相結(jié)合作為虛擬人的路徑點采樣策略。利用解耦式軌跡規(guī)劃方法,提出一種具有最大速度和加速度約束的多樣性軌跡設計策略。同時,為了提高避碰行為的真實性,提出利用虛擬人的社會學角色參數(shù)和個性參數(shù)改進避碰行為模型,從而根據(jù)不同的虛擬人內(nèi)在屬性生成多樣性的避碰行為。四、針對組行為生成問題,分別對平民組行為和士兵組行為進行研究。提出將可視-Voronoi圖用于組運動路徑的生成,進而利用平滑方法、多樣性軌跡生成策略以及修改評價函數(shù)三種方式產(chǎn)生組的運動軌跡,使組運動軌跡滿足四個方面的需求,即組的軌跡與障礙間應具有所設定的間隙、組的路徑應盡量短、具有平滑性和多樣性。針對非一致性平民組,基于分布式控制算法,對基本組行為進行了分析,進而提出了一種具有多個leader和全局軌跡約束的組行為生成方法。設計了一個兩步驟控制策略：第一步通過在支撐軌跡上設置一個可移動的“假想領導者”使組中所有領導智能體的中心滿足全局軌跡的約束；第二步利用所有領導智能體帶領所有跟隨智能體,從而將組中所有跟隨智能體的中心滿足全局軌跡的約束。將這兩個步驟相結(jié)合,最終實現(xiàn)組中心滿足全局軌跡的約束。另外,通過在障礙的邊界引入一類“假想的β智能體”,增加組對環(huán)境中任意障礙的避碰能力。針對群體性事件應用背景,對一致性平民組和士兵組行為進行了研究,提出利用組外形保證組的一致性,并且使得組外形根據(jù)環(huán)境狀態(tài)和全局軌跡的曲率變化而產(chǎn)生靈活的修正。提出利用組外形結(jié)合組隊形對具有嚴格隊形約束的士兵組行為進行模擬,在此過程中,利用組外形和期望隊形在初始隊形和目標隊形間構(gòu)建一個隊形關鍵幀序列,使得組中的所有士兵智能體沿全局軌跡從位于初始區(qū)域的初始隊形運行至位于目標區(qū)域處的目標隊形。一方面既保證了個體間不發(fā)生碰撞,另一方面又保證了組隊形盡量與期望隊形接近以及組隊形的靈活性。提出的組行為生成方法能夠生成人群行為仿真中的各種組行為,滿足了日常生活和群體性事件中對組行為的需求。五、為了實現(xiàn)日常人群管理和對騷亂人群的控制,針對組控制問題,提出了一種改進的shepherding行為。根據(jù)已有shepherding行為的不足,引入具有最大速度和最大加速度約束的時間最優(yōu)軌跡規(guī)劃方法,生成控制智能體的運動軌跡,使得控制智能體在滿足最大速度和最大加速度約束的前提下,盡快運動至目標位置,從而對人群進行更有效的管理和控制。六、最后進行了數(shù)據(jù)和模型混合驅(qū)動的虛擬人群行為仿真系統(tǒng)開發(fā)及其應用研究。以現(xiàn)有的虛擬人應用開發(fā)軟件為基礎,利用數(shù)據(jù)和模型混合驅(qū)動的行為建模思想,我們將基于速度場的行為建模方法和基于智能體的行為建模方法相結(jié)合,實現(xiàn)二者的優(yōu)勢互補。以廣泛應用的商業(yè)軟件DI_Guy和Vega Prime軟件平臺作為可視化渲染輸出平臺,通過調(diào)用底層的SDK動作庫函數(shù)對虛擬人群進行控制,并以群體性事件想定為例,完成相關應用系統(tǒng)開發(fā),進而檢驗了本文所做的研究工作。仿真結(jié)果表明,本文提出的方法真實地模擬了群體性事件中人群的各種行為,從而為群體性事件中人群的管理、人群行為的分析提供有價值的參考。
[Abstract]:With the increase of social unstability factors, the frequent occurrence of sudden security events, especially domestic group events, has caused great damage and casualties to the society. In this context, it is of great significance to improve the ability of coping with emergencies to ensure public safety. In order to provide technical support for the crowd management and control decision making with military application background, it has become a research hotspot in recent years. This paper takes group events as a representative military application background. In view of the poor behavior model in the current virtual crowd behavior simulation, the single model can not simulate the real behavior of the crowd. Phenomenon, the problem of poor applicability of the crowd behavior simulation framework, and the practical needs of the crowd behavior model. The key technology of virtual crowd behavior simulation is carried out from the design of the crowd behavior simulation framework, the independent individual's navigation behavior, the various types of group behavior generation and the group control line in the group events. In this paper, the corresponding control strategies and models are proposed and the traditional algorithms are improved. The main contents are summarized as follows: first, the thesis first summarizes the research status of the related technologies and applications of virtual crowd behavior modeling at home and abroad. The application background carries out the crowd behavior simulation research, divides the crowd in the group events into the civilians and the soldiers, among which the civilians are made up of independent individuals and groups. Thus, the behavior modeling of the different components in the crowd is proposed in order to establish a realistic model of the crowd behavior. Two, the design of the crowd behavior simulation framework is a problem. On the basis of multi agent architecture, a behavior / cognitive hybrid architecture is proposed based on the behavior based agent architecture and the prudent agent architecture. This architecture can not only design obviously different from the reaction rules, complex individual behavior, but also can be added to the behavior module. Degree planning thus generates more complex behavior modules, so it can handle unstructured environments well, achieve complex high-level planning and rapid response to environmental changes two complementary advantages. The behavior / cognitive hybrid control structure is based on the behavior based system design framework, so it is more prudent / than prudent. The combination of reactive hybrid architectures or simple behavior based structures is more advantageous and more suitable for simulation modeling of crowd behavior. In our architecture, the combination of global planning and local behavior, the combination of macro behavior and microscopic behavior, and the combination of data driven methods and model driven methods are realized. It can support the production of a more real crowd behavior model and greatly improve the adaptability and extensibility of the crowd behavior simulation framework. Three, to create the problem of the navigation behavior of the independent individual, to improve the realistic nature of the virtual human movement, the trajectory of the virtual human must be satisfied, the diversity and the most important. Large speed and acceleration constraints. Firstly, the adaptive modification of the sampling strategy of the PRM (probabilistic roadmap) algorithm is adapted, and a path point sampling strategy is proposed, which combines the non uniform sampling and the bridge-test sampling strategy to the virtual human, and uses the decoupling trajectory planning method to bring up a maximum speed and acceleration constraints. At the same time, in order to improve the authenticity of the collision avoidance behavior, this paper proposes to improve the collision avoidance behavior model by using the sociological role parameters and individual parameters of the virtual human, and thus generates diversity avoidance behavior according to the intrinsic attributes of different virtual people. Four, the behavior generation problem of the group and the group behavior of the civilian group and the soldier group are respectively discussed. In order to study, the visual -Voronoi graph is applied to the formation of the group motion path, and then the path of the group is produced by using the smoothing method, the diversity path generation strategy and the modified evaluation function to produce the movement track of the group, which makes the group motion track meet the needs of the four aspects, that is, the path between the group and the obstacle should have the set gap and the path of the group. It should be as short as possible, with smoothness and diversity. Based on the distributed control algorithm, the basic group behavior is analyzed based on the distributed control algorithm, and a group behavior generation method with multiple leader and global trajectory constraints is proposed. A two step control strategy is designed: the first step is set by the support trajectory. A mobile "imaginary leader" makes the center of all the leading agents in the group meet the constraints of the global trajectory; the second step uses all leading agents to lead all the following agents to meet the constraints of the global trajectory of all the following agents in the group. The two steps are combined to finally achieve the group center satisfaction. In addition, a class of "imaginary beta agents" is introduced into the boundary of obstacles to increase the collision avoidance ability of any obstacle in the environment. According to the application background of the group events, the behavior of the conformance group and the soldier group is studied, and the consistency of the group shape is proposed and the group shape is rooted. According to the curvature of the environment and the global trajectory, a flexible correction is produced. A group shape is used to simulate the behavior of a group of soldiers with strict formation constraints. In this process, a series of key frame sequences are constructed between the group shape and the expected formation between the initial formation and the target formation, so that all of the groups are in the group. The soldier's agent runs along the global trajectory from the initial formation in the initial area to the target formation located at the target area. On the one hand, it not only ensures no collisions among the individuals, but also guarantees the formation of the team shape to be close to the desired formation and the flexibility of the formation. For the various groups of behavior in the simulation, the demand for group behavior in daily life and group events is met. Five, in order to realize the daily crowd management and control of the riot population, an improved shepherding behavior is proposed for the group control problem. According to the shortcomings of the existing shepherding line, the introduction has the maximum speed and maximum acceleration. The time optimal trajectory planning method of degree constraint generates the motion trajectory of the control agent, making the control agent moving to the target position as quickly as possible to the maximum speed and maximum acceleration constraint, so as to manage and control the crowd more effectively. Six, the virtual crowd driven by data and model is carried out most later. The behavior simulation system development and its application research. Based on the existing virtual human application development software, using the behavior modeling thought mixed with data and model, we combine the behavior modeling method based on the speed field and the behavior modeling method based on the agent to realize the complementary advantages of the two. DI_Guy and Vega Prime software platform as the visual rendering output platform, by calling the underlying SDK action library function to control the virtual population, and take the group events as an example, complete the development of the related application system, and then test the research work done in this paper. The simulation results show that the method proposed in this paper is truly simulated. The various behaviors of the crowd in mass incidents provide valuable reference for the management of crowd events and the analysis of crowd behavior.
【學位授予單位】：國防科學技術(shù)大學
【學位級別】：博士
【學位授予年份】：2014
【分類號】：D631.43;TP391.9

【參考文獻】

相關期刊論文 前1條

1 林履端;匹配與最優(yōu)分配問題(連載四)[J];福州師專學報;1999年03期

，

本文編號：2101294