Retime, or more precisely Retime:Vissim is an online platform for optimization of traffic signal timings. It requires that a user uploads a Vissim (network) file (preferably properly calibrated and validated, to ensure high-quality of the results) and relevant signal controller files, which contain original signal timings (e.g. from field or any others). Once the files are uploaded, the Retime:Vissim uses a Genetic Algorithm (GA), as a smart search mechanism, to test hundreds/thousands of signal timings to find the set of best signal timings for the given Vissim network. A user can select which performance measure from Vissim is used as an objective function (to be minimized or maximized) during the optimization process. Once the optimization is finished all of the tested signal timings are saved in the database that a user can either download or utilize Retime’s analytics engine to visualize and analyze.

On one hand, Retime:Vissim is similar to these conventional signal optimization tools as it also optimizes signal timings to find the best set that improves traffic flows in a network. On the other hand, Retime:Vissim is very different from two major perspectives: 1. Unlike these other tools, Retime:Vissim utilizes very powerful high-fidelity Vissim model to accurately represent field traffic conditions (without limitations of deterministic traffic models utilized by the other tools), and 2. Retime’s online analytics allow users to analyze and visualize all of the optimization results, long after the optimization is finished. This powerful analytics gives users power to search and customize optimization results in any way it fits their interests.

These are only two major differences; there are many small differences that give Retime:Vissim advantage over the other conventional tools.

Retime:Vissim enables users to get more realistic signal timings that will require less fine-tuning in the field and will reliably bring benefits of new signal retiming projects. Many times, when signals are retimed with the traditional tools (based on deterministic traffic models), retiming does not bring any significant benefits as the deterministic tools are not able to account for true traffic demand and dynamics observed in the field. Such retiming projects, although this is often not publicized, end up with the field signal timings being reverted to the old signal timings ( which were in effect before retiming). This practice depletes infrastructure improvement and maintenance funds of the public agencies and contributes to an opinion that signal retiming projects are not cost effective.

Also, most of the conventional tools do not offer users options to optimize/retime their signals based on a variety of objective functions. For example, signal retiming cannot be done, with the conventional tools, to reduce person-delay for all (multimodal) users in a transportation network. To be more precise, short of few traditional objective functions supported by most of such deterministic tools, users have no choice to select which performance measure should be minimized or maximized. On the other hand, Retime:Vissim offers users almost unlimited possibilities regarding which parameters (or even their combinations, in future) could be minimized/maximized during the optimization/retiming process.

Retime: Vissim uses Genetic Algorithm to find an optimal solution in an optimization process that mimics evolution of living organisms. The GA is a metaheuristic approach inspired by the process of natural selection that belongs to the larger class of evolutionary algorithms (EA). Genetic algorithms are commonly used to generate high-quality solutions to optimization and search problems by relying on biologically inspired operators such as mutation, crossover, and selection.[1]

Whether GAs fit into a family of Artificial Intelligence is a matter of many scientific debates. One school of thoughts states that one of the characteristics of any intelligence is ability to solve problems. Another characteristic of intelligence is ability to learn, which means improving its performance based on past experiences. Thus, since GAs are computational problem-solving techniques that find and improve solutions (they learn), they can be considered a branch of AI.

[1] Mitchell, Melanie (1996). An Introduction to Genetic Algorithms. Cambridge, MA: MIT Press. ISBN 9780585030944.

This depends on the user inputs, or number and quality of signal timing solutions that a user wants to achieve from the optimization (and populate the database with enough solutions for post-optimization analysis). The total optimization time mainly depends on four major factors: 1. Size of the Vissim network (especially number of signals) to be optimized; 2. Duration of a single simulation run (to extract Vissim results); 3. Number of signal timing solutions to be evaluated and recombined through many generations during the optimization process; and 4. Number of generations in the optimization process.

1. Size of the network (e.g., number of controllers to be optimized) will impact Vissim loading time (each controller takes some time to load). So, Vissim files/networks with more controllers will take more time to load and this will impact the total optimization time as evaluation of each signal timing solution requires a corresponding Vissim file to be loaded and run.

2. Simulation run time of a single Vissim simulation will have a significant impact on the overall optimization time. If users want to base their retiming analysis on 15-minute simulation runs, the optimization time will be about 4 times shorter than if the analysis is done based on 1-hour simulation runs.

3. Each signal timing solution, that is being evaluated in a single Vissim run during the optimization process, represents a group of signal timings (where size of the group is equal to number of the signals in Vissim network that are being optimized) of individual signals. More signal timing solutions means that the user will start with a wider reach when trying to find combination of signal parameters that work for the Vissim network. If an analogy is made to a fishing net – more signal timing solutions (e.g., 50 or 100 instead of 10 or 20) means ‘a larger fishing net and thus a higher chance to catch fish’.

4. A generation in GA optimizations, represents one recombination of signal timing solutions. Similarly to how humans had evolved through generations of human populations, GA in Retime takes all of the signal timing solutions from one generation, recombines them to try to find even better solutions, and runs all solutions to evaluate their quality (regarding particular objective function that is sought to be minimized/maximized), before recombining them again. The major difference between human evolution and GAs is that human population increases with every new generation (so far, under normal circumstances) whereas populations in GA are usually kept fix: while each generation will have a slightly different population of the solutions, the size of population (or number of solutions) remains constant.

For example, if a user wants to run an optimization of 10-20 signals, based on 1-hour analysis, with a population of 20 signal timing solutions, and through 100 generations – this would take, approximately, 30-40 hours of optimization run time. Obviously, with all other things being equal, if the optimization is done for 15-min analysis (with 5 minutes of warm-up time), which may be long enough for many undersaturated networks, this would reduce optimization run time to 10-15 hours, or practically the optimization would be done overnight.

Hypothetically, both consulting firms and government agencies could be users of Retime:Vissim. Practically, consulting firms could be slightly ‘better equipped’ with experience and knowledge to handle Vissim files and therefore may be able to adopt and deploy new technologies in a faster way. While Retime LLC team could help in beginning with execution of some Retime:Vissim retiming projects, our long-term vision is to stay away from consulting jobs on retiming of traffic signals and simply provide (and maintain) tools which will help others in the industry to routinely perform signal retiming jobs.

This artificial competition between ‘modeling’ and ‘measuring’ does not reflect reality. We measure one realization of the events that have happened in the field, but we cannot measure hundreds (if not thousands) of realizations that have not occurred in the field. To provide a very simple signal-timings example: if green time at an upstream signal is restricted to 30 seconds every cycle, an observer at a downstream signal will never be able to count/measure traffic flow higher than what can ‘fit’ in those 30 seconds of upstream signal’s green time. Thus, one could get a wrong impression that providing similar duration of green at downstream signal is enough, while not understanding that the upstream signal is heavily metering traffic that would flow downstream in higher quantities if only more green time were given at the upstream signal. Only if this information is fed into a software, which can model traffic properly, one will be able to try various signal timing combinations and will be able to achieve an optimal solution. Some people call this approach applying multiple “what-if scenarios”. However, we can call it any way we want – such an approach requires ability to model traffic conditions and various signal timing combinations and without modeling there is no way to test which signal timing combination is the best.

That being written, many software tools for optimization of traffic signals are based on low-fidelity models which cannot realistically represent traffic in the field. Even highly sophisticated tools, such as Vissim, require extra efforts to properly calibrate and validate the models to realistically mimic the field conditions. Retime LLC team strongly advocates use of properly calibrated and validated Vissim models to achieve high quality of optimal signal timings.

The ATSPM tools and data can greatly aid the processes of: 1. increasing the quality of our traffic simulation models, 2. utilizing novel objective functions, and 3. analyzing signal performances. Several plans are being considered to integrate ATSPM elements in the Retime:Vissim platform.

Currently, if users want to test Retime:Vissim and they are not sure if their Vissim files are compatible with Retime:Vissim, they can send us a note and we will work with them to test compatibility of their Vissim files. Such a compatibility includes two components: whether Vissim version used in Retime:Vissim can open given files, and/or whether the files (otherwise compatible with the relevant Vissim version) have all of the elements needed for full utilization of Retime:Vissim platform. For example, related to latter component, analytics part of the Retime platform offers certain features (e.g., time-distance diagrams) which can be utilized only if the relevant travel time sections are defined in the Vissim file.

Very soon Retime:Vissim will offer a testing feature where a user will be able to test compatibility of its files on its own, which will significantly simplify the process. Retime:Vissim is always updated to work with the latest version of the relevant Vissim software, which ensures backward compatibility in a similar way that such compatibility is offered by Vissim.

Retime:Vissim currently supports only RBC controllers in Vissim. Our team is working on making Retime:Vissim capable of working with Econolite ASC/3 controllers (this will probably be the first next signal controller available in Retime:Vissim) and others. Also, there is a plan to make Retime:Vissim available for fixed-time and Vissig signal timing options, in near future.

We agree – proper calibrations and validations take time. So does extra time spent in the field to tweak controllers to adjust timings to accommodate shortcomings of the poor deterministic traffic models used by other HCS-based optimization tools.

Retime:Vissim also offers that all of the signal timings are stored in a database connected with powerful search engines and visual analytics. By using those Retime’s features users can explore various signal timing solutions and pick and choose ones that they like, based on several customized criteria. So, chances that none of the solutions will be very valuable are very slim.

No matter that common signal controllers (unless there are special signals for bicyclists, transit signal priority, or special pedestrians) do not contain many parameters that can be tweaked during the optimization process, a user can always (in Retime:Vissim) decide to minimize or maximize an objective function based on multimodal performance of its users. For example, one can minimize the total person delay (combined delay of pedestrians, bicyclists, transit riders, and drivers) by adjusting regular signal timing parameters such as cycle length, offsets, phase sequence, etc.

Theoretically Retime:Vissim can work with as many random seeds as someone wants to use. However, we intentionally do not support this option, currently. The reason for this is quite simple: considering that each random seed introduces a different random condition, such practice throws the searching algorithm of the ‘right’ path to find a good solution quickly. Another reason is that random process in Vissim should not be used to model traffic variations (as there is no coherent structural relationship between how traffic flows at various parts of the network are changed), which is why introducing various random seeds introduces much more harm than brings benefits.

However, a user can always take several of the best signal timing solutions (this is where a database with all of the solutions becomes pretty handy) and put those through multi-random-seed evaluation before making decision which one to adopt as the best. Currently, such multi-random-seed evaluation needs to be done outside of the Retime platform, but we will work on integrating this feature in the Retime, in future.

Let us first clarify here difference between signal timing (for single intersection), signal timing solution (for all intersections optimized in the same Vissim file/network) and signal timing pattern (within RBC or another controller, of a single intersection).

Signal timing for an intersection (e.g. RBC file) can contain one or more signal timing patterns (e.g., for AM peak, Midday, PM peak). While Retime:Vissim will be able to work with multiple signal timing patterns (within RBC or similar file) in future, right now it works only with one such a pattern. This is similar to other signal optimization tools, which also separately optimize signal timings for each Time-of-Day (TOD) pattern.

Again, signal timing solution is just a group of individual signal timings for all of the signals optimized within the same Vissim network/file.

Yes, any Retime optimization can be interrupted at any time. If a user wants to continue with optimizations, the process will need to start from scratch, but it can be resumed from the generation when interrupted.

Both options are available. If users only need to perform an optimization once, and do not need to use of any of Retime analytics, they can log in, upload files, finish the optimization, pay a fee, and download all of the results in a variety of data formats. On the other hand, if users want Retime to store their data and provide access to those data through visualization and analytics tools, it is required to pay monthly or annual subscription fee.

Yes, a user can do that. When starting optimization process, users are asked to select all of the signals that need to be optimized (e.g., a network can have 10 signals, but the user can decide to optimize only 7). This is simply done by unchecking the boxes next to those signals which are not supposed to be part of the optimization process.

Yes, this option is available too. When starting optimization, a user can simply select which signal timing parameters should be modified during the optimization process and the program will modify only those types of timings, while keeping all others as originally provided.

The best assurance, in addition to the fact that Retime:Vissim consistently outperforms deterministic optimization tools when optimizations are done with Vissim’s high-fidelity models, is the fact that all of the tested signal timings are stored in the database and available for further analysis and processing. In our experience, this guarantees that users will always find some signal timing solutions that satisfy their requirements.

Retime:Vissim simply takes what is specified in users’ signal controllers, within Vissim network. If those controllers are already set to work in coordinated fashion, Retime:Vissim will assume that this is what a user wants and it will continue to optimize these signals to work as a coordinated group. If some of the signals, from the initial Vissim file, are isolated – Retime:Vissim will continue to handle/optimize those signals as isolated signals. If there are several coordinated groups within the same Vissim network, Retime:Vissim will optimize signal timings separately for each group making sure that the same coordinated groups are preserved.

If users want to group signals in coordinated groups differently from how they are grouped in the original Vissim network (or not grouped at all), this is possible to achieve too but through a workaround that requires at least two optimizations. In future we will enable a feature in Retime:Vissim where users will be able to specify, at the start of optimization, how they want to group the signals, regardless of how they are grouped in the original Vissim network.

Right now, Retime:Vissim enables various objective functions, mostly all based on network evaluation outputs from Vissim. This includes an array of common performance measures, such as delays (stopped, total, latent, etc.), stops, throughput, travel time, speed and others. All of these are objective functions consisting of individual performance measures with the exception of Performance Index (PI), which is defined as a linear combination of delays and stops where each stop is given a delay penalty of 10 seconds.

In future, users will be able to combine various performance measures into objective functions of their own, by giving each performance measure a specific weight. Until then, users can use Pareto-Front charts in Retime analytics to manually identify (from the entire pool of solutions) those signal timing solutions which satisfy their criteria – e.g., a solution with minimal delay and maximum throughput.

Yes, Retime:Vissim can better process oversaturated traffic conditions than deterministic signal optimization tools. Major reason for this is that traffic models of deterministic tools do not support concept of building longitudinal queues and thus cannot be aware (reliably) of situations when queues grow backwards so much to affect upstream traffic. On the other hand, Vissim (if properly calibrated and validated) can handle such situations. Thus, Retime:Vissim can produce signal solutions which can remedy these situations (as much as it may be possible, considering that sometimes no signal solutions can fix oversaturation problems).

Right now, the Retime:Vissim cannot use Purdue Coordination Diagrams (PCDs), or any other of the ATSPM signal performance measures, as objective functions during its optimization processes. While the interface between Retime and ATSPM will be developed sometimes in future, users should be aware that many of PCDs features are available in Retime’s time-distance diagrams (based on Vissim files and signal timing solutions generated by Retime:Vissim). For example, Percent (of Arrivals) on Green, Platoon Ratio, and a couple of similar Retime’s proprietary performance measures are available in such time-distance diagrams. And such measures are calculated more accurately than the PCD’s measures, by taking in consideration queue growth and discharging processes, as recorded by relevant trajectories from vehicles from the Vissim models.

So, while such “PCD-like” measures are not used as objective functions, a user can create a time-distance diagram (for each signal timing solution) from Retime’s database and observe how well that particular signal timing solution works to provide progression in the optimized transportation network.