研究表明减少人的不安全行为是降低建设项目安全事故频发的主要途径，但现有研究大多狭隘地将"人"理解为个体，忽略了群体作用，更未见研究建设行业中普遍存在的非正式群体（如：同乡）及其对个体不安全行为影响的成果。因此以建设项目群体不安全行为为研究对象，首先分析农民工、管理者等个体不安全行为特征、行为参数，研究正式和非正式群体对个体不安全行为的作用机理，分析行为交互规则和群体行为规律，构建行为概率模型，完善了建设项目群体不安全行为理论框架；其次，引入人工社会方法建立多层次agent模型，采用swarm平台进行了重复实验，实现群体不安全行为涌现，揭示建设项目人的不安全行为的群体本质；最后，对建设项目安全行为管理制度的控制过程和效果进行了多次实验，提出优化制度建议，并以重庆为例进行了实证研究，为矫正建设项目群体不安全行为、改进现有行为管理措施提供了"清晰"的思路，也为行为矫正理论提供了新的视角。

It has been determined in previous literature that minimization of human beings' unsafe behavior in construction projects is a primary manner for reducing the frequent occurrence of safety accidents. Nevertheless, the research object associated with unsafe behavior in existing studies is merely limited to individuals, neglecting the important role of groups. Also, some informal groups which are very common in the Chinese construction industry, such as fellow-townsman, are not given significant attention. Thus, this study aims to examine the group unsafe behavior in the construction sector. The participants' individual unsafe behavior will be first investigated, including the factors affecting the unsafe behavior and the quantitative model for measuring the occurrence of unsafe behavior. Then a multi-level agent-based model will be developed to simulate the emergent of unsafe group bahavior by using the artificial societies approach. A series of experiments will also be conducted with the model to reveal the fundamental causes leading to the group unsafe behavior. Finally the model will be applied empirically in Chongqing with an intention to improve the safety behavior in construction activities. The findings can not only enrich the existing theory in construction safety management, but can be very useful in construction safety management practices.