基于计算智能的群体行为控制模型及路径生成研究

Crowd motion is a very interesting and popular phenomena in nature. This research proposes a control model of crowd behaviors and the approaches for path generation, aimed at the problems of current control model lack of flexibility for expressing individual and extremely calculate spending of multi-agent local control model, based on the deep analysis for the mechanism of crowd motion.This approach employs the dynamic niche technology to classify the large size crowd into three levels of crowd, group and agent, and selects an optimal agent in the group as the leader. It establishes the cognitive model, sesentive and cooperative reinforcement learning mechanism for leader's fitness function being adjusted dynamically according to environmental feedback information, and the adaptive ability of the leader is increased while the adaptive ability of the group is increased via cooperative learning and information sharing. The path programming is carried out at the level of group. It combines the multi-object optimizing property of artificial bee algorithm and the group intelligence property of particle swarm optimization algorithm for increasing convergence and optimizing speed, and solving the problem of individual obstacle avoidance and collision in group, and realizes collection, separation and leader-following of crowd effectively.

群体运动是自然界中非常有趣并且普遍的现象。本研究拟通过对群体运动的机理做深入分析，针对现有的全局控制模型缺少表征多样个体的灵活性及多Agent局部控制模型计算开销太大的问题，提出一种群体行为的控制模型及路径生成方法。该方法采用动态小生境技术对大规模群体进行群体、组及Agent三个层次的分类，并在每个组中选择一个最优Agent作为leader; 建立Leader的认知模型、感知及协同强化学习机制，使Leader能依据环境的反馈信息，动态调整自适应函数,提高自身的自适应能力，并通过协同学习及信息共享，提高群体的适应能力；路径规划在组的层面进行，结合人工蜂群算法的多目标寻优特性和微粒群算法的群体智能特性，提高算法收敛及寻优速度，解决组中的个体的避障及碰撞问题，并有效地实现群体的聚集、分离及跟随Leader。

群体运动是自然界中非常有趣并且普遍的现象。本研究通过对群体运动的机理做深入分析，针对现有的全局控制模型缺少表征多样个体的灵活性及多Agent局部控制模型计算开销太大的问题，提出一种群体行为的控制模型及路径生成方法。该方法采用基于网格-密度-关系的算法对大规模群体进行群体、组及Agent三个层次的分类，并在每个组中选择一个最优Agent作为leader; 建立Leader的认知模型、感知及协同强化学习机制，使Leader能依据环境的反馈信息，动态调整自适应函数,提高自身的自适应能力，并通过协同学习及信息共享，提高群体的自适应能力；路径规划在组的层面进行，结合人工蜂群算法的多目标寻优特性和微粒群算法的群体智能特性，提高算法收敛及寻优速度，解决组中的个体的避障及碰撞问题，并有效的实现聚集、分离及跟随Leader。本项目以大规模人群疏散仿真作为应用背景，试图为紧急情况下群体运动及人群疏散路径规划的仿真和决策提供辅助的支持。该项目已经在在国内外期刊上发表论文22篇，其中SCI期刊发表论文15篇，EI期刊发表论文4篇，中文核心期刊发表论文3篇。授权发明专利4项，获山东省科技进步二等奖1项。该项目还培养了博士研究生4名，硕士研究生10名。