行人微观仿真技术已经成为地铁站等行人密集场所方案评价和疏散评估的主要研究手段．社会力模型从受力角度进行建模，可得到行人的运动状态和受力状态，不引入复杂的判断规则即能得到丰富的自组织现象．文中针对该模型存在的算法复杂度高和物理粒子运动盲目性的问题，引入分子动力学的Gear 预测校正法和链接列表元胞算法，基于Agent 感知－决策建模方法，构建了改进模型及其面向对象的编程实现框架．采用椭圆形的行人形体描述方法，标定了行人形体、运动及模型参数，设计了密度扫描转向、变作用强度及碰撞预测规避等行人Agent 交互方法及算法，在VC++ 2008 平台实现了改进模型并进行地铁站通道的单、双向仿真．结果显示链接列表元胞法可以明显减少仿真时间，模型可较好地实现自动渠化，与车站调研得到的通道客流密度－流量关系吻合．

The micro-simulation of pedestrians has become a major tool of facility layout assessment and evacuationevaluation for the places with dense pedestrian flow such as a subway station.The social force model,which is formulatedfrom the aspect of force exerted on pedestrians,helps to obtain both the motion and the force states of pedestrians,and demonstrates various self-organization phenomena without setting complex walking rules.In order toovercome the high complexity of algorithm and the blindfold moving of pedestrian particles in the social force model,this paper introduces the Gear's predictor-corrector method and the linked-list cell algorithm in molecular dynamics,and presents an improved model and the corresponding object-oriented programming implementation frameworkbased on the perceiving and decision-making method of Agent.Then,with the help of the ellipse element modelingthe pedestrian body,the parameters including the pedestrian body,the movement and the model itself are calibrated,and the Agent-pedestrian interaction method as well as the corresponding algorithm,which takes into considerationthe density scanning /turning,the varied interaction strength and the collision prediction /avoidance,designed.Finally,the modified model is programmed with VC++ 2008 and the one-direction and bi-direction pedestrian simulationsof subway passageway are performed.The results indicate that the linked-list cell algorithm greatly saves thesimulation time and that the modified model helps to achieve prominent lane formation and reveal the pedestriandensity-flow relationship which agrees well with that derived from field study.