In the previous blogpost we identified the most important locations in a road network with the aim of finding the best locations for sensor placement. In this post we continue our search for interesting locations in a traffic network. However, in contrast to the previous post, where we were after the most informative spots in order to optimize a robust sensor placement strategy, here we focus on identifying the sensitive and problematic locations in the system in terms of traffic.
As the medium that allows traffic demands to be met, the road network topology of a city is vital in order to ensure smooth conditions for all commuters. As such, its design and incremental alteration are complex and challenging tasks. One common misconception is that building more roads will remove congestion. On the contrary, the Fundamental Law of Road Congestion (Duranton and Turner 2011) states that widening of major roads not only does not lead to an improvement in the traffic system but also increases the level of congestion. This is simply due to an increase in willingness of traffic participants to travel. On the other hand, simply removing roads is not a practical solution as the system obviously experiences high congestion due to bottlenecks (Koshi et al., 1992) that propagate like a wave throughout the network (Laval and Lecrercq, 2010). In other words, in order to avoid the aforementioned negative effects of over or under supply of traffic media, the road-designers have to take several factors into account to match the needs of the population.
In a perfect static world, these problems can be addressed easily using the huge amounts of data generated by road sensors. However, traffic demand and the road network, the two factors that determine traffic, both evolve over time and at different paces. Even if much planning and effort is put in to designing and building new roads (see this cool video showing the evolution of the London road network), the changes in traffic demand is much more dynamic and frequent. Therefore, the road network is bound to always fall behind in its adaptation attempts. This leads to mismatched intersections, lack of optimal roads, existence of obsolete roads and irregular traffic patterns that ultimately all cause high congestion levels. In this article we present some findings that can help identify intrinsic sources of congestion.
In Figure 1 we can see a map of Singapore that depicts as hot spots, intersections in the network; however, there is a mismatch between the traffic demand and the provided infrastructure. The measure that we use here is the total number of lanes that should be redistributed within an intersection in order for it to optimally meet the traffic demand. For example, let’s assume that there is an intersection with two options; left turn and right turn. Let’s say that 70 percent of the drivers turn right, however two lanes are allocated for the cars that turn left and only one for the cars that turn right. In this case there is a mismatch between the demand and the road network design. We measure this mismatch by saying that one lane must be redistributed from the left turn to the right, in order for the demand to be met by the network. The analysis indicates that even in Singapore, that has a rather good reputation for its traffic network, there are severely mismatched intersections with as many as six lanes that need to be redistributed within a single intersection. These are examples of situations where fast changes in the traffic demands have resulted in bad allocation of road resources.
Even if we assume that the problematic locations identified above can be resolved, there are still other severe issues for the traffic system. The demand on a daily scale can vary a lot more than the long term variations. For example, a left turn at a certain intersection might be favored during the morning rush hours, while in the evening the right turn may be more common. Even if adaptive traffic control systems are put in place, the number of lanes allocated for every road (which ultimately determines it's capacity), is static and cannot be changed on an hourly (or even monthly) basis. In Figure 2 we have shown the intersections that exhibit such extremely dynamic behavior. To clarify, we are not examining intersections that have a high degree of variation in the volume of vehicles that pass through them; rather these are the intersections that experience QUALITATIVELY different demands throughout the day.
We call the measure of dynamicity the dynamic factor of an intersection. The maximum value of this factor is 1 and corresponds to a 100% change of the turn preferences of the drivers at an intersection every 15 minute interval. This scenario sounds highly improbable; however, the biggest dynamic factor that we measured in Singapore is not that far off from 1! As a matter of fact, there are 4 intersections that have a dynamic factor over 0.7 with the highest being 0.8. Another interesting observation can be made from Figure 2. While in most central zones like the downtown area (south central part), the residential areas (east, north and central parts), there are plenty of intersections with high dynamic factors, the situation in the east seems rather static and apparently homogeneous throughout the day. One possible reason for this is that, the two biggest universities are located there and the population in those regions is, to a large part, students. Due to the flexible hours and qualitatively different lifestyle we can observe that the traffic there is a lot more homogeneous and therefore the problem with highly dynamic intersections is almost non-existent.
We hope that the presented measures can be used by city planners as an analytical tool that directly points out to sensitive and problematic locations, that when resolved will lead to serious improvements in the traffic conditions overall. Using those measures as a heuristics for the global minimisation of any desired factor such as average travel time or average travel distance, can speed up the whole process since it serves to direct the attention of any optimisation algorithm directly to the places of interest rather than blindly searching for them in the first place. The work presented in this post covers two papers that have been accepted to the ITSC'2015. The work will be presented on site and consequentially published.