A simulation heuristic for traveler- and vehicle-discrete dynamic traffic assignment
针对出行者和车辆均为离散个体的动态交通分配问题，提出一种基于Nikaido-Isoda函数的仿真启发式方法，在保证计算效率的同时逼近最优解，并在斯德哥尔摩、奥斯陆和柏林案例中验证了优越性能。
• Formal modeling of an agent-based traffic assignment problem using the Nikaido-Isoda (NI) function. Construction of an upper bound on this function and recovery of the sorting method of Sbayti et al. (2007). • Reformulation of the NI function bound into a new assignment formulation that transparently trades rigor against heuristic efficiency. Thorough analysis and illustration. • Approximation of expected utilities and expected dynamic network flows given a stochastic DNL for discrete vehicles. Development of an efficient recursive computation scheme. • Experimental comparison to alternative heuristics within a mainstream agent-based DTA simulation. Method parameter exploration in different case studies and congestion regimes. A dynamic traffic assignment problem is considered where travelers are modeled as integral decision makers and network flow is composed of integral vehicles. As travel behavior affects network conditions and network conditions affect travel behavior, a complex model system results. The versatility of the considered model class has led to increasing practical interest (“agent-based simulation”) but also complicates the development of solvers for mutually consistent travel behavior and network conditions that represent possible long-term states of a transport system. Continuum flow assignment techniques are not applicable to this model class. This work starts out from a Nikaido-Isoda gap function for the traveler- and vehicle-discrete dynamic traffic assignment problem. A tractable but rather uninformative upper bound on this gap function is derived. A reformulation is presented that violates this bound as little as possible while ensuring that the reformulated bound carries relevant information for the subsequently developed new assignment heuristic. The proposed approach is formally related to and experimentally compared with relevant methods from the literature. It is found to exhibit superior performance in nontrivial case studies for Stockholm (Sweden), Oslo (Norway), and Berlin (Germany).