State estimation for traffic simulations as coarse grained systems

传统的基于计算机的系统仿真主要关注数学上表现良好的系统，这些系统主要由（微分）方程建模，并通过易于理解的数值方法进行模拟。由于这些模型的形式化性质，已经开发出了用于校准和跟踪它们的可靠数学技术。最突出的例子可以说是Kaiman滤波器，它有大量的实际应用，特别是在物理和控制工程领域。最近，人们对数学上不太容易理解的基于规则的粒子系统的建模越来越感兴趣，并且已经开发了一些微观计算机模拟系统。这种模拟范例将单个粒子及其动态表示为软件对象。目前可用的状态估计方法缺乏处理这种粗粒度过程的能力，这些过程通常是不可微的和随机的。由于微观模拟将继续渗透到实际科学中，因此非常需要对粗粒度系统的新状态估计方法进行系统研究。这种多智能体方法的重要实际应用是交通模拟。作为软件代理的旅行者的个体表示允许高度真实的交通表示，尽管再次以数学可追溯性为代价。由于几乎所有可用的交通规划和管理工具都需要当前和/或预测的移动模式作为输入数据，因此粗粒度系统状态估计的新方法有可能在这一领域提供直接的好处。

Traditional Models and State Estimation Traditional computer based system simulations focus on mathematically well behaved systems that are modeled mainly by (differential) equations and are simulated by well understood numerical methods. Due to the formal nature of such models, sound mathematical techniques for their calibration and tracking have been developed. The most prominent example is arguably the Kaiman filter with its vast number of practical applications particularly in the fields of physics and control engineering. Multi-Agent Simulations Recently, an increasing interest in the modeling of mathematically less well understood rule-based particle systems has emerged and several microscopic computer simulation systems have been developed. This paradigm of simulation represents individual particles and their dynamics as software objects. Currently available state estimation methods lack the capability to deal with such coarse grained processes, which generally are non-differentiable and stochastic. Since micro-simulations will continue to pervade practical sciences, there is considerable need for systematic research on new state estimation methodologies for coarse-grained systems. Application to Traffic Systems Important practical applications of this multi-agent methodology are traffic simulations. The individual representation of travelers as software agents allows for a highly realistic representation of traffic, although again at the cost of mathematical tractability. Since almost all available traffic planning and management tools require current and/or predicted mobility patterns as input data, new methods for state estimation of coarse grained systems have the potential to provide immediate benefits in this field.