近年来群体性事件频发，尤其是大规模高度暴力的群体性事件，严重威胁了公共安全，甚至可能造成政权动荡。因此，关于群体性事件中暴力行为的研究具有重要意义。基于此，本文从群体暴力行为的影响要素出发，分别针对事件的发生期、发展期和爆发期开展暴力威胁评估和暴力行为防控仿真研究。本文的主要研究内容和成果如下：第一，分析群体性事件的时空分布规律、主要诱发因素以及暴力程度的分布和变化，研究暴力程度与各类背景要素、组织要素以及警务处置要素之间的相关性；基于新闻检索数据收集群体性事件案例数据，并对案例进行编码，展示我国群体性事件的图景和全貌。第二，针对发生期的群体性事件，建立基于贝叶斯网络的威胁评估模型；利用贝叶斯网络推理，构建未来情景，评估正在发生的群体性事件的威胁，对可能的警务策略进行情景分析，研究可以降低事件威胁的最佳警务措施；对比模型建议的警务策略与历史案例中记录的警务策略，评估贝叶斯网络模型支持群体性事件警务决策的可行性。第三，针对发展期的群体性事件，建立基于Agent的情绪感染（ABEC）模型，分析群体暴力爆发的条件；考虑抗议者不满情绪的感染过程以及抗议者与警察之间的行为交互，将抗议者分为冷静、活跃和暴力三种状态，以空旷广场中抗议者和警察随机分布为仿真场景，探索群体性事件从和平集会到暴力爆发的动态演化；通过调整仿真模型中与情境要素相关的参数，开展对照实验，识别影响暴力爆发的主要因素以及遏制暴力爆发的方式。第四，针对爆发期的群体性事件，建立基于Agent的人群控制（ABCC）模型，分析真实的事件视频，总结常用的人群控制策略及其对应的人群特征，估计模型参数；考虑警察防御和进攻两种状态，引入力学模型开展多目标和自组织的运动特征研究，以街头抗议中警察试图阻止抗议者进入某一区域为仿真场景，分析警察采取不同人群控制策略时人群的运动典型特征，比较不同策略对暴力人群的围堵性能和威慑效果，研究最佳的控制策略。本文对群体性事件暴力行为的关键要素及其相互影响进行了系统的梳理和分析，通过贝叶斯网络模型对群体性事件的威胁进行评估，基于仿真模型分析了群体暴力行为的特征及其控制策略。研究成果将显著改进群体性事件应急处置的威胁评估和预测能力，为群体性事件的应急决策提供支持。

In recent years, the frequent occurrence of mass incidents, especially the large-scale mass incidents with group violent behaviors, has seriously threatened public security and may even cause social unrest. Therefore, the study of group violence in mass incidents is of great significance. This paper analyzes the influence factors of group violence, and then respectively focus on the occurrence period, development period, and outbreak period of mass incidents, assessing the threats of mass incidents and simulating the prevention and control strategies of group violence. The main contents and achievements of this paper are as follows:Firstly, based on news retrieval, this paper constructs a database of mass incidents and code it with several attributes. Then, the paper analyses the temporal and spatial distribution of mass incidents, the main inducing factors and the distribution and change of violence degree, showing the panorama of mass incidents in China. Using correlation analysis, the paper studies the correlation between the violence degree and various background elements, organizational elements and policing disposal elements. Secondly, aiming at the mass incidents in the occurrence period, a threat assessment model based on Bayesian network is established. Using Bayesian network reasoning, the paper constructs future scenarios, assesses the threat of ongoing mass incidents, makes what-if analysis to compare the effects of possible policing strategies, and analyzes the best policing strategy that can reduce the threat of mass incidents and provides reasonable suggestions. To evaluate the feasibility of the Bayesian network model in supporting policing decision-making in mass incidents, the policing strategies proposed by the model are compared with those recorded in historical cases. Thirdly, aiming at the mass incidents in the development period, the paper establishes an Agent-Based Emotional Contagion (ABEC) model to analyze the conditions of the outbreak of group violence. The ABEC model considers the infection process of protestors’ grievance and the interaction between protestors and police and assumes that protestors have three states: quiet, active and violent. Taking the random distribution of protestors and police in the open square as the simulation scenario, this paper explores the dynamic evolution of mass incidents from peaceful gatherings to the outbreak of group violence. By adjusting the parameters related to situational factors in the ABEC model, the comparative experiments are conducted to identify the main factors affecting the outbreak of violence and the ways to curb it. Fourthly, aiming at the mass incidents in the outbreak period, this paper analyzes the real incident video, summarizes crowd control strategies commonly used by police and their corresponding crowd characteristics, estimates crowd parameters, and then establishes an Agent-Based Crowd Control (ABCC) model. The model considers the two states of police, defensive and offensive, and introduces the force-based model to realize the multi-objective and self-organizing movement characteristics of agents. Taking the police trying to prevent the protestors from entering a certain area in street protests as a simulation scenario, this paper analyzes the typical characteristics of the crowd when the police adopt different crowd control strategies and compares the blocking performance and deterrence performance of different strategies to the violent crowd.This paper systematically combs and analyzes the key elements of group violence in mass incidents, evaluates the threat of mass incidents through Bayesian network model, and analyzes the characteristics and control strategies of group violence based on simulation models. The results of this study can improve the ability of threat assessment and prediction in the emergency response process of mass incidents, and provide support for the decision-making.