Harnessing Connected Vehicle Data

By Zachary Jerome, Xingmin Wang, Zihao Wang, Henry Liu, Contributing Authors

Traffic signal optimization is known to be a cost-effective method for reducing congestion and energy consumption in urban areas without changing physical road infrastructure.

However, the high installation and maintenance costs of road-side detection systems have prevented the widespread implementation of traffic responsive systems. As a result, a large proportion of the signalized intersections in the United States do not have detection capabilities and are still controlled by fixed-time traffic signals.

Most signal retiming projects at these intersections are only executed every three to five years and rely on temporally limited manual traffic count studies. These “one-shot” optimizations can become outdated quickly as traffic demand undergoes natural changes or growth, which can increase congestion and energy costs. 

Data collection remains the chief bottleneck for more iterative retiming. A cheaper, more efficient method for continuously monitoring traffic and diagnosing signal timing issues would potentially shorten the time between each iteration, allowing for a more responsive traffic signal system that could recover lost opportunities as traffic demand changes over time (Figure 1).

In response to this opportunity, our team at the University of Michigan developed a large-scale traffic signal re-timing system that uses GPS data from a small percentage of connected vehicles (CV) as the only input without reliance on any infrastructure-based detectioni.

Connected Vehicle Data

In recent years, vehicle trajectory data from GPS points has become increasingly available from various connected vehicle services such as en-route navigation, roadside assistance and ride-hailing services. Monitoring traffic through vehicle trajectory data offers many advantages over fixed-location detectors and sensorsii iiiiv.

It has a much larger coverage area than detector data because it is available at almost every intersection, especially those with high traffic volumes. While detector data can only provide traffic counts and estimated speeds at certain locations, vehicle trajectory data spans the entire spatial-temporal space and provides more enriched information such as delay, number of stops, and travel path (Figure 2).

This presents an unprecedented opportunity for traffic signal optimization that can reduce traffic congestion without additional sensor instrumentation on physical road infrastructure.

Traffic State Estimation

Traffic signal retiming relies on models that can estimate the current traffic state and predict future states based on different signal timing parameters such as cycle lengths, offsets and green splits.

Once you have a calibrated model, you can identify the traffic plan that has the lowest predicted performance measure you are trying to optimize for (e.g., delay) in your model. Since vehicle trajectory data provides different performance measurements (delay, stop locations and arrival time) than detector data (speeds and volumes), we developed a new traffic flow model that could be calibrated from this new data set (Figure 3).

Another challenge with vehicle trajectory data is the sparse and incomplete observation of the overall traffic state based on low penetration rates (i.e., the proportion of observed vehicles to the overall number of vehicles).

Many existing studies that investigate traffic signal control with connected and automated vehicles assume a high penetration rate, which is not realistic in the current practice. Our goal was to retime traffic signals utilizing vehicle trajectories at the currently available market penetration rate.

To address these challenges, we developed a stochastic traffic flow model under our proposed Newellian coordinates, which is established based on Newell’s car following modelv. This coordinate system allows us to harness the spatial-temporal advantages of vehicle trajectory data.

Even at a low penetration rate, recurrent traffic states can be accurately reconstructed by aggregating sufficient historical data to build a probabilistic time-space (PTS) diagram. This enables us to develop a traffic signal optimization method that can transform the state-of-the-practice at scale. 

Optimizing Traffic Signals as a Service (OSaaS)

With the proposed model, we presented a large-scale traffic signal optimization system — Optimizing Signals as a Service (OSaaS) — based on vehicle trajectory data collected by connected vehicle service providers.

OSaaS is a closed-loop signal optimization system that includes monitoring, modeling, diagnosis and optimization (Figure 4).

In each retiming iteration, delay and stop measurements are first calculated from the collected trajectories to evaluate traffic performance. Traffic flow parameters such as the penetration rate and arrival rate are then estimated based on the proposed traffic flow model.

Based on the calibrated model, the diagnosis module finds the traffic performance optimality gap with respect to different signal timing parameters, which indicate different traffic signal re-timing opportunities.

Optimization algorithms are developed to update signal timing parameters for intersections that show potential for improvement. In this way, the OSaaS system can dynamically optimize traffic signal periodically every few weeks, compared to the ~3-5 years in the current practice. 

Field Implementation

The OSaaS system was tested in Birmingham, Mich., which has 34 signalized intersections, including three main corridors and some isolated intersections.

More than three quarters of these intersections had not been retimed for more than two years. In addition, most of these signalized intersections are not equipped with any connected detection systems, so the proposed system provided previously unavailable opportunities. Two of the corridors were diagnosed with coordination opportunities.

New optimized signal timing plans were deployed in April 2022, and three weeks’ data before and after the implementation were used to evaluate before-and-after performance. We estimated that the penetration rate was around 7%. 

Figure 5 illustrates the results of the field implementation for one of the optimized corridors (Adams Road) during the morning peak period (northbound).

Left and right figures show the aggregated time-space diagram of the corridor using vehicle trajectories before and after the offsets were adjusted, respectively. The aggregated time-space diagram is directly generated from the observed vehicle trajectory by aggregating trajectories within the same time of day interval into one cycle.

It clearly shows how vehicles pass the whole corridor across multiple signalized intersections. The dashed outlined box highlights how the stopping behavior improved for two specific intersections, and the blue lines are hypothetical trajectories that traverse the whole corridor and demonstrate how these trajectories have less delay and stops after the optimization.

In the northbound direction, the number of stops and average control delay decreased by over 40% and 20%, respectively. The other direction of this corridor which is not shown also performed better than before by roughly 5%. 

OSaaS is a large-scale traffic signal optimization system based on low penetration rate vehicle trajectory data. This system is cost-effective because it does not require installation and maintenance of road-side detectors.

Without being restricted to installed locations, vehicle trajectory data is more scalable and is available for the whole road network, particularly for intersections with high traffic volumes. Collective observation is also more robust to equipment failure, as it will not be affected if one vehicle loses its connectivity.

As a closed-loop system, OSaaS continuously monitors urban traffic and can generate new signal timing plans whenever sufficient historical data is accumulated. It significantly shortens each re-timing iteration, so a more responsive traffic signal retiming is feasible.

OSaaS provides a scalable, sustainable, resilient, and efficient traffic signal retiming method that can potentially upgrade every fixed-time signalized intersection in the world from static systems with irregular retiming to dynamic systems with frequent updates. RB