A GIS Approach to Dynamic Network Routing

System Characteristics
Unlike conventionalGIS applications,only points and linear objectsare consideredin real-time routing problem. Additionally,someof the attributes associatedwith point and linear objectsarea fiction of time; hence,the data model in GIS must be able to handle both spatial and temporal information. For example, as depicted in Figure 1, linear objects (e.g., street networks) and point objects (e.g., fixed objects such as detectors, and moving objects such as probe vehicles) must be relatable for spatial and temporal analysis. Dynamic attributes of dynamic point objects (e.g., vehicle speed) and linear objects (e.g., traffic volume) can be collected in real-time and telemetered to TOC. These dynamic attributes must be integrated with static attributes of linear objects such as number of lanes, directional constraints, and link capacity. That data then serves as input to models derived tlom traffic flow theory and dynamic network models to produce new information such as travel time or travel cost, which are subsequently assigned to the linear objects as attributes. The methodology for integration of this data and models needs to be carefully developed so that predicted travel times reflect the reality of tic conditions.

Figure 1: Trafiic Data Required for Real-Time Routing
Detectors are installed along the highway, major arterial, and intersections to collect real-time data, or so-called “dynamic” data, such as flow rate, travel speed, and trailic volume. The ADVANCE (Advanced Driver and Vehicle Advisory Navigation Concept) project had used probe vehicles mounted with expensive sensing and telemetering systems to report traft3c data, especially congestion situations, back to TOCS so that overall tratlic conditions in study areas can be analyzed simultaneously (Bowcott et al. 1993). Besides real-time tratlic data Kaysi et al. (1993) suggest that historical data or statistical data can also be used to improve the prediction accuracy of future traffic conditions which, thereafter, will assist drivers in providing more realistic route choices.

Procedures for integrating data from various sources are critical to the solution of congestion prediction and dynamic network problems. Such procedures result in “data fusion” -- one of the most important data sources as input to the dynamic traffic assignment model. Because of sparse distribution of probe vehicles, loop detectors and closed-circuit television, real-time data collected by these means are not sticient to give accurate predictions of traffic flow over five- to ten-minute time horizons. Historical origin-destination data and infrastructure data such as signal timing at intersections, number of lanes, and speed limits, are also important for real-time transportation applications. Currently, approaches proposed for data fhsion are by means of fuzzy logic and neural networks (Ivan et al. 1995).