U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590

Skip to content

Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations

This report is an archived publication and may contain dated technical, contact, and link information
Publication Number: FHWA-HRT-10-038
Date: October 2010

Balancing Safety and Capacity in an Adaptive Signal Control System — Phase 1



Over the past 20 or more years, several adaptive traffic control algorithms have been developed and used for traffic signal control around the world, including Split Cycle Offset Optimization Technique (SCOOT), Sydney Coordinated Adaptive Traffic System (SCATS®), Federal Highway Administration (FHWA) adaptive control systems (ACS) such as Optimization Policies for Adaptive Control (OPAC), Real Time Hierarchical Optimized Distributed Effective System (RHODES), and, most recently, ACS Lite. While these systems differ in operation, they share the common objective of increasing throughput capacity and optimizing traffic flow by responding to current traffic demands rather than assuming predetermined traffic volumes.

The SCOOT algorithm simultaneously optimizes three parameters: (1) splits, (2) offsets, and (3) cycle. At every intersection and for every phase, the split optimizer determines whether to make the change earlier, later, or as currently planned. The split optimizer implements the decision, which adjusts the phase time by a few seconds to minimize the degree of saturation on the approaches to the intersection. Similarly, the offset optimizer determines whether to alter all the offsets by a fixed amount every cycle. The offset optimizer uses information stored in cyclic flow profiles to compare the sum of the performance measures on all adjacent links of an arterial system of intersections for the current offsets and possible changed offsets to determine if the changed offsets will improve traffic flow efficiency. The cycle optimizer varies the cycle time of all the intersections in a traffic section or arterial in small intervals each cycle in an attempt to ensure that the most heavily loaded intersection in the system is operating at less than 90 percent saturation. SCOOT includes a variety of other optimization and congestion management features, all of which are focused on improving efficiency with no assessment of the effect on traffic safety.

SCATS® uses different techniques with a similar goal of improving efficiency of traffic flow and minimizing total system delay. It is based on a split plan selection, matching traffic patterns to a library of signal timing plans, rather than incremental tuning. By measuring real-time traffic flows at the intersection (notably using stop-bar detectors), SCATS® determines the degree of saturation of each traffic phase and selects the signal plan that best minimizes the total degree of saturation at each intersection. Offset and cycle time adjustment algorithms work in a similar fashion to SCOOT but, again, have no assessment of the impact of system decisions on safety. Both SCATS® and SCOOT make decisions at an intersection no more than once per cycle.

FHWA ACS algorithms OPAC and RHODES were developed to provide real-time adaptive traffic control at intersections on a second-by-second basis. Similar to SCOOT and SCATS®, RHODES and OPAC estimate the effect of signal timing changes on efficiency of traffic performance with no assessment of traffic safety effects. RHODES and OPAC make frequent decisions about phase timing (green durations) with consideration of parameters such as cycle time, offsets, and splits, but they do not explicitly consider these parameters in their optimization. OPAC and RHODES focus on modeling the traffic state (length of queues and approach demands) in the system and minimizing the total delay to vehicles in a rolling-horizon optimization fashion.

The FHWA ACS Lite optimization algorithm was developed to address some of the shortcomings of the ACS algorithms. It is based on a simple traffic model with few tunable parameters requiring little calibration. ACS Lite uses three levels of optimization to refine and update traffic signal timing: (1) a time-of-day tuner, (2) a run-time refiner, and (3) a transition manager. The time-of-day tuner is intended to keep baseline timing plans updated by learning from past run-time refiner actions. At the time this report was written, the time-of-day tuner had not yet been implemented in the ACS Lite system. The run-time refiner makes adjustments to signal timing parameters every few cycles based on cyclic flow profiles and stop-bar occupancy data similar to the information collected by SCOOT and SCATS®. The transition manager is intended to select the best method of transitioning between two traffic signal timing plans when the run-time refiner makes an adjustment to the timings. ACS Lite may be slow to respond to rapid changes in traffic flows, but it requires fewer detectors and much less infrastructure, configuration, and calibration than other real-time adaptive traffic control systems.

Each of these systems has been developed to optimize traffic flow, minimize traffic delay, and improve efficiency of traffic signal operations. It is unknown, however, if maximizing traffic efficiency in real time compromises safety in any way. For example, an adaptive system might tend to generate relatively short cycle lengths because short cycles tend to minimize delay. An adaptive system might also generate network-wide solutions that will have many stops but with an average stop of a short duration. Are such solutions less safe than a throughput-maximizing solution that would have fewer stops in the network but higher average delay? Evaluating such issues and developing an adaptive control strategy that balances maximizing traffic flow efficiency and safety is the primary purpose of this project.


The body of literature analyzing the effects of signal timing settings on intersection safety is very limited. The analysis of safety from a real-time context is an emerging topic that is primarily addressed from a vehicle-centric approach. In the authors' opinion, the best example of research on the effects of signal timing on safety is the development of the Detection Control System (DCS) by Texas Transportation Institute.(1) This system detects the type of vehicle (large truck or regular passenger vehicle) approaching an intersection and modifies the clearance time of the service phase appropriately to reduce the occurrence of large trucks caught in the dilemma zone. Bonneson's studies have shown that this type of real-time adaptive control is effective at improving the relative safety of an intersection (e.g., 80 percent reduction in red-light running by heavy trucks) with limited negative effect on the intersection's efficiency and, in some cases, even improvements in total delay and stops.(1) These results were found with the application of the DCS system in rural areas that have high-speed main line approaches carrying significant truck traffic and relatively low-volume side street flows. This real-time adaptive control approach was found to be effective because it focuses on one particular signal timing parameter (i.e., yellow and red clearance). Urban and suburban applications of the DCS approach are not likely to see similar results due to differences in the operational environment.

Research on real-time predictions of when a crash is going to occur is very limited. Some research has been pursued to develop relationships with aggregate variables such as average speed, occupancy, or volume data in a freeway context. These studies showed that these relationships provided poor predictive power (R2 < 0.5), as observed in the regression model approach.(2,3) Beyond these studies, current research in real-time crash prediction is focused on vehicle-centric approaches such as the Cooperative Intersection Collision Avoidance System (CICAS). CICAS warns drivers when they are about to violate a red light by a high-speed communications link between the traffic signal and individual vehicles. Future extensions of this type of interaction between vehicles and the traffic signal, under the broad umbrella of IntelliDrive℠ technology, provide additional potential for changing signal timing parameters in response to the observation of impending crash conditions. These effects, however, are localized and provide only microscale changes, such as extensions to clearance times. This approach is not really applicable to making changes to aggregate-level traffic control parameters such as cycle time, splits, and offsets. When there is a significant penetration of IntelliDrive℠ equipment in the vehicle fleet of the traveling public, this technology may offer more promise for providing a balance of safety and efficiency in traffic signal operation.

For these reasons, the analysis of traffic conflicts has emerged as the best possible surrogate measurement technique for predicting unsafe situations when making changes to signal timing parameters. It is hypothesized that locations that experience a higher rate of traffic conflicts will typically have a higher rate of crashes. Several field studies have shown that conflict rates do have a relationship with an increased rate of crashes, although this relationship is still debated by safety analysis statisticians.(4) Although the correlation is relatively weak (as is the correlation of crash prediction models to real crash rates), it is positive. Although the conflict analysis technique is based on manual observation, it is still the best tool available for safety analysis in lieu of collecting crash records.

The recent FHWA research on surrogate measures of safety from simulation models concluded from literature review that conflict analysis represents the most appropriate and intuitive approach for assessing the safety performance of new and innovative intersection designs.(5,6) Since a new geometric design, signal system operational setting, or control method has no crash history to draw from, there is essentially no way to predict safety performance except for some combinations of existing regression-based relationships and crash modification factors (CMFs) using engineering judgment. Thus, FHWA's Surrogate Safety Analysis Model (SSAM) project focused on providing a safety analysis tool for simulation-based comparative studies.

The SSAM software tool processes trajectory information from a microscopic simulation model to compute the frequency of various types of conflicts (crossing, lane changing, and rear end) and severity indicators. Time to collision (TTC), postencroachment time (PET), speed differential, and several other measures can be used to compare one traffic facility design to another. The results of the FHWA studies showed that the frequency measures were more reliable than the severity indicators, but many interesting effects were observed, such as a change in statistical significance at varying levels of traffic volume.(5, 6) That is, the conclusion that one design is more safe than another may change when the real-time traffic volume is higher or lower. For example, it was found that single-point urban interchanges (SPUI) had lower conflict rates than traditional interchanges at low volumes but higher rates at high volumes.

The validation study as part of the SSAM development project showed a reasonably weak correlation of conflict data with crash records for a battery of statistical tests, including rank-correlation and deriving regression models for crashes from conflict data.(6) This validation analysis did not, however, study any effects of the differences in signal timings at the 83 intersections used in the study.

The SSAM research did evaluate differences in safety performance of several signal timing-related parameters as follows:(5,6)

  • Leading versus lagging left turns.
  • Three-phase versus four-phase diamond intersection control.
  • Protected versus permitted left turns.

The remainder of the test cases focused on comparing geometric design features of intersections. As expected, leading and lagging left turns were found to have no appreciable difference in conflict rates for a test case performed at an isolated intersection. Results for three-phase and four-phase control at a diamond interchange were found to be inconclusive because low volumes indicated superiority of three-phase control while four-phase control was found to produce lower rates of conflicts at higher volumes. For the current project, however, it is actually a positive result that the three-phase and four-phase approaches had different safety dominance at different traffic volume levels, since this means that in a real-time setting, an algorithm might switch the interchange operation from a three- to four-phase operation to improve safety performance. Finally, it was found that protected left turns reduced the rate of crossing conflicts over permissive lefts, which was also expected.


The overall objective of this research was to develop real-time adaptive signal timing methodologies and algorithms that balance safety and efficiency. This research consists of two phases, and this report summarizes only the findings of phase 1. The first step of phase 1 was to identify the relationships between traffic signal parameters (cycle, spilt, offset, detector extension time, change and clearance intervals, left-turn phase sequence, and left-turn phase protection alternatives) and safety (rear-end, angle, and lane-change conflicts). FHWA's SSAM was used to evaluate various simulated scenarios to test the relationships between the signal timing parameters and the occurrence of traffic conflicts in the simulation model.

Since there is a limited amount of research on relationships between signal timing and safety (considering both surrogate measures and crash records), these initial test cases were necessary to identify evidence of a relationship between a particular signal timing parameter and a statistically significant change in safety performance. After identifying parameters that hold the promise of an effect on safety and efficiency, the next objective was to identify an approach for combining these relationships into a multivariate performance function that can be used to predict the safety implications, or the change in conflict rates, of modifications to signal timing.

This performance function could be used in tandem with efficiency assessment functions to provide a computational engine that can be used to assess the performance of a given set of signal timing parameters with respect to the current traffic conditions. A multiobjective adaptive control algorithm was developed to balance the two objectives, although it is important to note that the two objectives may not necessarily be competing in all areas of the parameter space. Next, the algorithms are to be implemented in an offline and/or online solution with modern controller hardware and software and tested for proof of the concept. Finally, validation studies are needed to verify that the proposed algorithms have a positive effect on safety in real-world deployments.

This research project achieved the first three objectives of the effort: (1) to identify a short list of signal timing parameters that are positively correlated with safety, (2) to develop a plan for combining the results into a safety performance function, and (3) to develop a multiobjective algorithm that provides a balanced solution between safety and efficiency in the context of real-time adaptive traffic control. Depending on the outcome of this initial effort, development of the performance function, algorithm implementation, field testing, and validation studies will be performed in future phases of the research.


This report is organized into the following six sections:

  • Section 1: provides an overview of the project and a statement of the research objectives.

  • Section 2: explores the relationships between signal timing parameters and safety.

  • Section 3: discusses the signal timing and traffic simulation tools used in the analysis.

  • Section 4: discusses the various case scenarios considered in the project.

  • Section 5: summarizes the findings of the analysis and simulations.

  • Section 6: discusses the concept of algorithms for developing a multiobjective optimization and safety performance function. It outlines a detailed experimental design approach and methodology, and it summarizes the various methodologies for developing both offline and real-time multiobjective optimization and safety performance algorithms.

Federal Highway Administration | 1200 New Jersey Avenue, SE | Washington, DC 20590 | 202-366-4000
Turner-Fairbank Highway Research Center | 6300 Georgetown Pike | McLean, VA | 22101