Application Framework

As stated above, the basic input to the dynamic assignment process is a list of trips, defining what vehicles depart at what departure time from which origin to what destination. For all considerations presented in this paper, the list of trips is regarded as given and fixed. The main goal lies in assigning actual route plans to these trips which fulfill a certain optimization criterion, e.g. minimizing each individual's travel-time based upon the actual time-dependent link travel-times which would be generated by executing the route-plans. In TRANSIMS a microsimulation is used to provide (and verify) the link travel-times generated by the route-plan. The goal is to generate route plans which are (within the limitation of the stochasticity of the process) optimal.

Now, initially, the algorithm cannot predict which route will be the fastest, since that depends on congestion, and no information about congestion is available initially. This problem is solved via an iterative relaxation approach, that is, one generates an initial set of routes, runs it through the micro-simulation, re-plans a fraction f_r of the trips, runs the micro-simulation again, etc., until some convergence criterion is fulfilled. Choosing f_r too large can result in oscillations (e.g. []); more sophisticated and computationally more costly route choice models can prevent them again (e.g. []). Figure 3 depicts the data flow during an iteration series. (An iteration series refers to a collection of iterations. An iteration consists of a run through the route planner, and another run through the micro-simulation.) Figure 4 shows the accumulated travel-time as a function of the iteration number for different iteration strategies and re-planning fractions of 1% and 5% (see []). With 5% re-planning, approximately 40 to 60 iterations are required before the travel-time has reached a sufficiently relaxed state.

  

Figure 3: Iterative assignment with simulation feedback. For iteration i+1the fraction f_r of the previous route-set is re-planned by the router using link travel-times of iteration i. The remaining fraction 1-f_r is simply reused. Before the route-set is fed into the simulation it is clipped to the boundaries of the study-area.

  

Figure: Relaxation of the accumulated travel-time. The series marked ``run 5'' uses a replanning fraction of f_r = 0.01, all other series use variants of f_r = 0.05. See [].