Adaptive Forecasting of FreewaY Traffic Congestion

Since 198L, the Washington State Department of Transportation (WSDOT) has used integrated traffic-responsive onramp control to cope with recurring traffic congestion on the Seattle region's portion of Interstate 5 (I-5). The algorithm used to calculate the on-ramp entry rates has two basic routines, Iocal control and bottleneck control. In the local control routine, the algorithm calculates metering rates on the basis of the measured lane occupancy at the main-line station immediately upstream of the on-ramp merge point' This portion of the algorithm is effective only as long as the demand for use of a section offreeway does not greatly exceed the capacity of the section. As is well known, both from traffic flow theory and from practical experience, when demand for access to a section of freeway is excessive, the operating speed of that section is reduced and vehicles queue at the points of congestion (bottlenecks). The bottleneck control routine of the WSDOT algorithm eliminates these speed reductions by restricting access to the freeway at one or more on-ramps upstream from the bottleneck point. A detailed description of the WSDOT control algorithm was provided by Jacobson et al. (1). One limitation of the bottleneck control algorithm is that it is reactive rather than anticipatory. It does not take action until a bottleneck has formed, so speed reduction and instability atready exist in the traffic stream. The bottleneck algorithm is oriented toward cleaning up messes rather than preventing them. If bottleneck formation could be forecast and bottleneck control could be used to prevent bottleneck formation, then at least in theory overall traffic volumes should increase and delay of the traveling public should decrease. The goal of recent research conducted at the University of Washington is to develop an algorithm that can reliably forecast bottleneck formation 1 min or more in advance of occurrence and to incorporate this new algorithm into the WSDOT ramp control algorithm.