Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
scale (lack of representational detail, specifically with regard to traffic queuing) or a microscopic scale (too computationally intensive to model an entire region). DTA MODEL SPECIFICATION The use of dynamic traffic models requires one to con- sider the specification of time much more carefully than is necessary in customary static network models. Time is relevant in a number of contexts in a dynamic model. First, demand is specified as the number of vehicles to load on the network during a certain time period. The Atlanta regional model defined demand in a 4-h period representing trips departing between 6:00 and 10:00 a.m. Analysis periods should be defined over which results of the DTA model will be compiled and compared with observed data. These periods could be a single 1-h period, several 1-h periods, several 15-min periods, and so forth. The analysis period should almost certainly not align with the demand period. If, for example, the demand period were 6:00 to 10:00 a.m. and the period of interest was 6:00 to 7:00 a.m., the DTA model would estimate travel times on the network by simulating vehicles that were entering the network starting at 6:00 a.m. The first vehicles to load on the network would have no other vehicles to contend with, which in most urban areas is unrealistic. If it is necessary to evaluate traffic simulated at 6:00 a.m., then some estimate of demand occurring before 6:00 a.m., say from 5:00 to 6:00 a.m., should be determined and used to allow the DTA model to produce realistic traffic levels at the times the intervals of interest occur. As is discussed later, getting realistic traffic at 6:00 a.m. is easier than getting realistic flows at 8:00 and at 9:00 a.m., as the network is continually loading. The demand period for trips departing their origins before the time interval of interest is often referred to as a warm- up period. The last vehicles to be loaded onto the network at the end of the demand period also have an important effect on other vehicles. For example, some number of vehicles will be loaded on the network just minutes before 10:00 a.m. It might be thought that those vehicles will only contribute to link flows after 10:00 a.m. and therefore that the simulation of vehicles need only occur between 5:00 and 10:00 a.m., giving 5 h of simulation time. How- ever, vehicles entering the network near 10:00 a.m. will have an impact on vehicles that started their travel ear- lier. The vehicles entering at 10:00 a.m. may contribute to congestion on links at 10:00 a.m. or later. Had these vehicles not been simulated past 10:00 a.m., links could be represented as having less congestion than they should have, which could affect route choices made by vehicles that started their travel much earlier than 10:00 a.m., and that will end their travel after 10:00 a.m. The affected route choices of vehicles will then of course influence link flows in periods earlier than 10:00 a.m. The end of the simulation period, the cool- down period, after vehicles are no longer loaded on the network, is therefore necessary as well. In the implementation, a 1-h warm- up period was used. Three 1-h analysis periods (6:00 to 7:00, 7:00 to 8:00, and 8:00 to 9:00 a.m.) were defined, for which flows were tabulated and compared with observed 1-h counts. Finally, a cool- down period sufficient to allow all vehicles to be simulated entirely from their origins to their destinations and therefore to exit the network was used. Depending on how well converged the dynamic user- equilibrium solution was, the cool- down period could have been from 3 to 7 h. The simulation period was therefore defined to start at time zero at 5:00 a.m. and end anywhere from noon to 5:00 p.m., depending on how well vehicles were allocated to routes. The more converged results that were used to compare with observed counts were typically based on 8 h of simula- tion time, with all vehicles exiting the network during those 8 h. Besides the demand and analysis periods, the DTA assignment procedures use assignment intervals and link aggregation intervals. An assignment interval is a length of time when all vehicles traveling between a given origin and destination and departing their origin during this interval experience the same travel time at equilibrium. When this state occurs over all assignment periods, the DTA is in a state of dynamic user- equilibrium. At that point, no vehicle has an incentive to follow a different route, and the DTA solution is stable. Assignment inter- vals were defined with lengths of 15 min. Link aggregation intervals are the length of time over which the simulated vehicle travel times on a link are averaged to yield a single link travel time for that aggre- gation period. These average link travel times by aggre- gation period are used in the time- dependent shortest path (TDSP) calculations that are part of the DTAâs dynamic user- equilibrium solution procedure. INPUT DATA REQUIREMENTS The input data for Vista include the Atlanta regional highway network described as a link table and a node table, much as the network is defined for the regional demand model. The link table contains node IDs at each end of the link, length, free speed, and link capacity information. The node table contains spatial coordinates and a type to distinguish regular intersection nodes from centroid nodes. Vista also uses input tables to define the location and operational characteristics of signalized intersections in the network. Finally, Vista requires an 102 INNOVATIONS IN TRAVEL DEMAND MODELING, VOLUME 2