Over the last few months City Science has been developing tools to convert between traditional four-stage transport models and microsimulation models. In this blog we set out our approach and the key benefits these techniques can deliver to planners and modellers.
The Conversion Tool
As part of our Innovate UK “Multi-Horizon” project we have developed tools to automatically convert traditional transport models to microsimulation and agent-based models. The Microsimulation converter makes use of the detailed junction data already coded within the four-stage model – this includes lane turn information, flares, merges, signal timings and phases, bus lanes and routes. In addition, from the majority of models we can capture how signals and bus routes change over the course of the day.
Example of a SATURN junction with bus priority
The set of SATURN matrices are merged into a single route file (a method of converting flows to vehicles which rounds the flows to integers while preserving the overall totals). Our system also enables us to cordon demand automatically in order to provide models for individual locations – this is useful if there is a particular area of focus for the microsimulation. The cordoning process is straightforward; the users provides a boundary polygon of the area of interest, and the model is cut to only those links and the vehicles whose routes have at least one link within that sub-model. The result is a “cookie cutter” approach where one large strategic model can be split into many models of the individual towns and city areas where the runtimes are fast enough to allow operational decisions as well to run signal optimisations which may require 100s of iterations. Since this is an automated process, we can enable data between the parent (4-stage) and child (microsimulation) model to be continually updated.
The MicroSimulation Model
The resultant models can be run at macro, meso or micro simulation levels of granularity. As the runtime varies as the model size increases, routes can be derived from a macro assignment before being cordoned and a Dynamic User Assignment (DUA) applied. The mesoscopic approach handles some signal and junction effects, yet by bucketing lengths of roads and queues into 100 meter segments, is able to run 100s of times faster than a full microscopic model.
Figure 1. An example of model interactions. Source: TfL Modelling Guidelines
The Benefits of Automated Conversion
The following are five reasons why such a conversion can be worthwhile and add value to your modelling processes.
1. Debugging and comparison
First, during the prototyping phase we quickly discovered that a converging macrosim model doesn’t necessarily runs well in microsimulation. By modelling individual vehicles it is easy to produce congestion effects that do not appear in the strategic model. Investigating where the new congestion occurs has often lead to the identification of coding errors in the strategic model which might otherwise have been overlooked.
Excessively large queues are an indication of coding error, by following the queue to source or rerunning the model it is possible to identify the source of any bottleneck. SATURN models are much more forgiving of bottlenecks and queuing than microsimulation techniques, microsim can hence be used to identify these errors in signal timings or turning movements.
2. Exploring Signal strategies
Traditional strategic models are limited in their treatment of signals. If signals are modelled at all they are modelled with fixed timings and often do not consider offsets or green waves. Microsimulation provides a much richer treatment of signals allowing for vehicle actuation, scheduled changes or the application of real-world control strategies such as SCOOT or MOVA.
Demonstration of a live phase diagram
It is a well-known challenge in highway engineering that maximising capacity and minimising average waiting times are two separate goals; what is optimal for one is not optimal for the other.
Working in collaboration with the University of Exeter we have also developed a multi-objective algorithm able to explore the optimal signal strategies that improve overall network performance. This approach is able to deliver optimal solutions for each separate goal but also identify solutions that are best at meeting any combination of goals.
Inspection of a traffic light program
This approach is transformative as no longer do mutually exclusive goals have to be decided prior to optimisation. Instead the exact trade-offs and compromises can presented alongside a set of strategies.
Using this approach around a model of an example UK city, we were able to reduce average junction waiting times by 10% while also reducing fuel and emissions by 2% compared to the base model. When applied to optimising travel time the approach identified improvements of up to 25%.
3. From Models to Digital twins: replicating live conditions
Based on our research, local authorities often don’t have access to tools to forecast near-term conditions on their network. For example, a common challenge is understanding how quickly a queue is likely to build up in the aftermath of an incident, and how traffic might reroute.
By connecting the microsimulation model to live data feeds it is possible to create a digital twin – a virtual representation of the current transport network that can be used for testing strategies and training. The models created by our process run 10-100 times faster than real time and can therefore be used to provide early warning or forecast the impacts of operational changes.
Example of how a virtual control centre could appear
It is also possible to create a virtual control centre that can then be used to trial new traffic management strategies; test UTC/UTMC hardware or software; and to train new staff on a sandboxed version of the real system.
4. Extending the model with new features – for example Parking
It can be challenging for traditional models to simulate certain features common to transport networks – for example the interaction between vehicles, public transport, pedestrians and cyclists as well as modelling parking capacity and vehicles searching for spaces.
To demonstrate the potential added value from model extensions, we enhanced the Microsimulation to include both parking and parking search behaviours for on and off-street car parks. The potential for Microsimulation, when paired with a strategic model is considerable. Our automated conversion system therefore provides the ideal platform upon which to model:
Pedestrian crossings
Cyclists
Park and ride sites
Bus and taxi ranks
Driver information systems such as SATNAVs and VMS
Variable Speed Signs
Electric vehicle charging
Emergency vehicles
Logistics
And even a series of use-cases related to Connected and Autonomous Vehicles.
Example of a modelled off-street car park, when a parking areas fills vehicles will reroute to another parking areas.
5. Visualisation and engagement
Finally, we’ve found considerable engagement from users with the Microsimulation since the outputs are easy-to-understand and highly accessible to non-modellers. While the engagement of microsimulation has been known for some time, our approach enables authorities to clearly link strategic models to the microsimulation. By easily converting between the two systems it is possible to have the best of both worlds while saving time and cost.
Demonstration of the user closing a lane
Conclusion
This project demonstrated that a model developed with strategic scope and detail is suitable for conversion to a microsimulation model. Such practices could also enable closer links between those involved in strategic planning and those with an operational focus. An automated conversion process removes the barriers between professional silos as well as the barriers between technical and non-technical users.
By enabling rich and varied modelling approaches to work seamlessly together we can enhance our understanding of the differences between techniques and apply a larger toolbox of approaches to understand and improve our transport networks.