Journal of Maps
ISSN: (Print) 1744-5647 (Online) Journal homepage: https://www.tandfonline.com/loi/tjom20
Transport Planning with Microsimulation
Pete Sykes
To cite this article: Pete Sykes (2007) Transport Planning with Microsimulation, Journal of Maps,
3:1, 122-134, DOI: 10.1080/jom.2007.9710833
To link to this article: https://doi.org/10.1080/jom.2007.9710833
Copyright Taylor and Francis Group, LLC
Published online: 18 Jan 2012.
Submit your article to this journal
Article views: 1002
View related articles
Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=tjom20
Journal of Maps, 2007, 122-134
Transport Planning with Microsimulation
PETE SYKES
SIAS Ltd, 37 Manor Place, Edinburgh EH3 7EB, UK; Pete.Sykes@sias.com
(Received 3rd August 2006; Accepted 12th March 2007)
Abstract: Microsimulation is an emerging technique used to investigate transport
planning options through simulation modelling. Its outputs serve two purposes; to
present a visual image of the predicted traffic suitable for public presentation and to
provide quantitative answers to differentiate between the levels of service offered by road
design options. This paper discusses the input data requirements the simulation process,
and these outputs.
122
ISSN 1744-5647
http://www.journalofmaps.com
Journal of Maps, 2007, 122-134 Sykes, P.
1. Introduction
Traffic flow software has been available in almost its current form since the
1960s. Its function is quite simple: build a computerised model of a road
system and load it with vehicles representing the demand on that system.
The software then enables the user to adjust the road network description
or travel demand within the model so that it represents what is observed on
the road. The model is then used to predict what will happen in the future
if the road network or travel demand changes.
For 40 years traffic flow software has evolved slowly, but now it is
undergoing a major change. Older methods would draw out vehicles into
strands along their entire journeys, accumulating these strands to represent
traffic flows on each link in the road network. Speeds would be inferred
from road type and vehicle (strand) density and the software would iterate
to converge on a solution which ensured that all journeys were made by the
quickest route. In effect traffic was considered as a river of constant flowing
metal over units of time, typically 1 hour. Geographic Information Systems
proved ideal for displaying this kind of spatial and static data.
Most travellers do not perceive road traffic to be smooth flowing,
particularly in congested road networks. In the real world, traffic signals
divide traffic into clusters, platoons form behind slow vehicles, buses stop at
bus stops and all manner of incidents occur. Some drivers are more aware
and more aggressive than others and while some dawdle, others speed.
Finding a gap in traffic to move out of a side road is not determined by an
average wait, it is determined by the available gaps, and these depend on
the prevailing circumstances. Smooth flow assignment methodologies
struggle to successfully simulate traffic in congested situations, which
effectively means almost everywhere.
Microsimulation is a relatively new technique for representing road traffic
flow, in which the actions of every individual vehicle are evaluated at sub
second intervals. Their journeys through the road network are derived by
analysing their interactions with the road network, the traffic control
systems, and with the other vehicles on the network. The clustering of
vehicles, the random wait for a gap in traffic and the variance in the length
of time required to make a journey are all automatically assessed and
assimilated.
123
Journal of Maps, 2007, 122-134 Sykes, P.
Microsimulation is an inherently dynamic process, so displaying its results
in conventional GIS form presents new demands, especially in view of the
variety of audiences for the output. At a public consultation meeting the
video gaming generation expect to see high quality animated graphics to
show the current and predicted scenarios as a video of moving vehicles. At
the engineering decision making level the requirement is for measurable
service indicators to differentiate between road network design alternatives.
2. Simulation
A microsimulation model is fundamentally quite simple. Each individual
vehicle follows the road alignment and constantly makes decisions
concerning speed and lane choice. At the end of the road it must decide
which way to turn, and also it must wait for a suitable gap in the traffic
before moving out. Such simulation can be implemented using a simple car
model and elementary physics involving equations of motion. However, the
devil is in the detail and there is a devil of a lot of detail.
S-Paramics (2006) is one example of a number of similar microsimulation
software packages. All are designed to perform a similar task. S-Paramics,
Aimsum (2006) and Vissim (2006) are the most well known. S-Paramics is
based on research from Fritzsche (1994) with modifications derived from
the experience of using it on many transport planning projects (Duncan
and McArthur, 1997). Recent research from the USA into driver behaviour
is also contributing to internal behaviour models (NGSim, 2006).
S-Paramics is a discrete time step based simulation where the timestep is
commensurate with a driver’s reaction time. At each time step drivers
make a set of decisions about what they will do in the future. It is the
interaction between individuals that allows microsimulation to represent
the flow of traffic on a road system.
One behavioural decision is how fast should a vehicle travel? Is it to be
constrained by the actions of others around it, by the rules of the road and
the speed limit, by the need to make a turn soon, by the caffeine level in
the driver’s bloodstream, the kinematics of the vehicle or by the
instructions from roadside gantries and their associated speed cameras. Not
surprisingly the answer is that all such factors influence the speed, and are
124
Journal of Maps, 2007, 122-134 Sykes, P.
subject to prioritisation. In a microsimulation model, each individual will
only be looking at the small number of vehicles ahead and alongside it to
determine its speed. “Conventional” traffic models simply assume that the
flow of vehicles is X per hour on a section of road then the vehicle speed
must be Y kmh at all points along it. This is known as the “speed/flow
curve”, and is an input to conventional models. It is however an output of
the microsimulation process.
Similarly, lane choice is a decision to be made by the individual based on
the attractiveness of the lanes available to that vehicle. This is determined
by the range of lanes, the sub range if we are in the vicinity of a junction,
the relative speed of vehicles in adjacent lanes and of course any
restrictions. It is important in the simulation that these choices are the
controlling factors in determining which lane is to be used. Some simulation
systems allow a modeller to specify lane use from observations, but these
will fail when used to predict traffic flow behaviour beyond the locus of the
observed input data.
These examples lead to one of the major principles of microsimulation:
Model the causes from first principles of driver behaviour, not the effects
observed. The effects will manifest themselves in the simulation if the
behaviour is correct. Representing the infrastructure of the road network is
the key to efficient simulation. Each lane is a thread along which a vehicle
will travel. This gives a one dimensional simulation, with lane changing
implemented as a dynamic trajectory enabling a vehicle to cross to the
parallel dimension. The threads connect and cross at junctions, and are
linearly segmented by features such as width, curvature, gradient and speed
limit.
Traffic demand is represented by a matrix of journeys between origin and
destination zones, which act as sources and sinks for vehicles. Zones
represent geographical areas of the model, typically a housing area or
industrial estate, or external connections at the periphery of the modelled
area.
As is the case for real drivers, modelled vehicles travelling from one zone to
another will vary their routes according to the prevailing network
circumstances. S-Paramics does not pre-ordain fixed routes between zones,
but allows vehicles to vary their routes during simulation depending on the
prevailing conditions. Each vehicle attaches a relative cost to distance, time
and tolls, and at each junction chooses the route that minimises the
125
Journal of Maps, 2007, 122-134 Sykes, P.
aggregate cost to the destination. If there is a choice of routes of similar
cost, then a degree of randomisation is first applied before the choice is
made. However, not all drivers have detailed knowledge of the road system,
so drivers are categorised into “familiar” and “unfamiliar”, the latter
perceiving non signposted roads as being a more costly alternative to
signposted main roads. A taxi driver or a commuter would be considered a
“familiar” driver, while a tourist is considered “unfamiliar”.
Familiar drivers benefit from local knowledge of congestion which, in
S-Paramics, is derived from the current state of the simulation. At regular
intervals the extra delays due to congestion are incorporated in the routeing
system, and these are treated as a proxy for the process of a driver learning
about congestion by trying different routes over a period of time. Since the
detailed level of route knowledge is shared by many drivers, and decisions
are made as needed rather than in advance, the dynamic route following
process is remarkably efficient.
Long journeys can be subdivided into segments through the addition of
“waypoints” to the model (Cragg, 2007). These are the key steps in the
overall journey and a driver may be thought of as “micro routeing” between
the waypoints that form the macro route. As an example, a trip from an
address in Central Edinburgh to a town in Fife may be described as “First
go to Blackhall Junction, then out of town through the Barnton
roundabout, over the Forth Road Bridge” etc. Each component is “micro
routed” independently and the overall journey put together as a macro
route. The key benefit of using waypoints to subdivide a journey occurs
when a long journey starts in an area of rich route choice. In the example
above, there may be several routes to Blackhall and the time difference
between them, expressed as a fraction of the overall journey time, is small
and hence different routes are likely to be chosen. However a driver whose
destination is Blackhall would find the time difference more significant
when expressed in terms of the overall journey and would therefore be more
likely to choose the shortest. Using waypoints ensures that the route choice
between them is similar whether the vehicle is on a short or a long journey.
Another benefit of the waypoint scheme is to provide labelled route choice
points which are used by intelligent transport systems through variable
message signs or broadcast media to provide advice on delays or diversions
to drivers and allow them to re-route accordingly.
As a consequence of change, every day on the road network is different, and
126
Journal of Maps, 2007, 122-134 Sykes, P.
drivers behave differently too. S-Paramics models this by assigning every
vehicle driver a measure of “awareness” and “aggression” which together
control how they react to road conditions, the headway they adopt, the
speed they choose and how early or late they make their lane choice. Daily
variation is taken care of by use of a random number generator to control
the time a vehicle is generated, the assignation of driver attributes, the
randomisation of route knowledge, overtaking, gap acceptance and other
parameters. This allows investigation of the variability of road network
performance. For example, if a network experiences free flow conditions
then journey times will be consistent and reliable. If it is congested then
large queues may develop as a result of minor events, and journey times
will vary much more. Modelling this variation is essential for the proper
analysis of network performance and is achieved by multiple model runs
with different random number seeds.
Some practitioners question the application of microsimulation models,
partly because official guidelines are yet to be updated with respect to
microsimulation methodology. But there are many situations where only
microsimulation can replicate contemporary traffic flow and congestion.
Consider an example of a roundabout on a dual carriageway, where
vehicles’ average journey speeds are reduced on joining the approach queue
and negotiating the roundabout. A proposal is made for a roundabout
flyover, but this must be justified by an economic evaluation which includes
an assessment of vehicle operating costs with and without the upgrade. It
is hypothesised that the overall journey speed rises from 50 mph to 70 mph
for the segment of dual carriageway that is proposed to bypass the
roundabout. A conventional assessment approved by the Department for
Transport (WebTAG, 2006) will calculate higher fuel consumption for the
design scheme, since optimum fuel efficiency is considered to be achieved at
55 mph, a factor which will act against its implementation. An assessment
based on a microsimulation model will take into account the difference
between fuel consumption in the decelerations and accelerations at all
stages of negotiating the queue and the roundabout, and compare this with
that of an uninterrupted vehicle using the flyover. The interrupted journey
incurs greater fuel use, and a microsimulation model will measure this as a
benefit of the flyover scheme, and not a disbenefit. This is intuitively
”correct” and represents one example of how less accurate traffic evaluation
procedures can lead to the wrong solution.
127
Journal of Maps, 2007, 122-134 Sykes, P.
3. Data Input and Results Output
Accuracy in representing the road system is vital for accurate simulation.
Typically a model will be constructed using a generally available electronic
map underlay of the road system. Maps or aerial photography can also be
used, but the level of detail may be lacking. Figure 1 shows two
representations of the same junction - in the same orientation - taken from
an Autocad drawing (Figure 1a) and from a roads “route planner” map
(Figure 1b).
In Figure 1a the location of the stop lines, the shape of the splitter islands
and an accurate size and shape of the overall geometry is clearly visible.
This detail is missing in the simpler map (Figure 1b). Since accurate
vehicle paths are needed to model the detailed interactions between
vehicles, the simple digital map is inadequate as a starting point.
Travel demand data - the number of vehicles travelling between zones - is
derived from a combination of land use data, census data, roadside
interviews and manual or automated traffic counts. From this information
Origin/Destination matrices are created in preparation for their contents
being released onto the modelled road network. Adjustments are made to
this matrix to rationalise it with the observed traffic counts as a part of the
model calibration procedure
Figure 1 Comparison of mapping detail.
128
Journal of Maps, 2007, 122-134 Sykes, P.
Once the basic simulation model is complete, the zone scheme and traffic
demand added and model calibrated to agree with observed data, then a
completed basic model is produced, as shown in Figure 2. This allows full
network simulation, including the production of network performance
statistics such as queue lengths and journey times and for comparing these
statistics when proposed changes to the road scheme are tested by applying
them to the model.
However, Figure 2 is simply a basic level of presentation, and for
consultation purposes more cosmetic work needs to be performed to
contextualise the model for a general audience.
Figure 2 Basic model display (click right image to play animation).
However strong the desire for presentation, there is rarely a matching
budget and so adding a visual layer to the model must be cost effective.
Figure 3 is an example of the same junction with three options:
• basic - no enhancement
• an aerial photograph and vehicle shapes
• 3D objects
In this case, the aerial photo is generic and its texture is solely adding
depth to the scene. More usually the photo will be of the actual area, and
relevant to placing the model in the correct geographic context.
129
Journal of Maps, 2007, 122-134 Sykes, P.
Figure 3 3D Presentation: Different levels of sophistication.
Figure 4 presents a simple animation of a junction with images of building
facades stood up alongside the road. The second animation shows three
options of quality of presentation which depend on the quality of the source
material to be included in the model.
Figure 4 Urban models with 3D annotation (click each image to play animation).
In this example, Haymarket in Edinburgh, the streetscape images match
the location of the simulation model. However, this particular model is
widely available and these images have been taken and incorporated in
models of other areas. The use of such presentation artefacts in such models
exists at two levels - an accurate map, or a set of visual clues to aid context.
Traffic models do not usually exist in isolation, and often there are other
resources available to enhance the basic simulation. 3D models prepared by
the project architect may be available, and these can be included included
130
Journal of Maps, 2007, 122-134 Sykes, P.
as graphical annotations in the simulation. Two examples are shown in
Figure 5. The first is a model in which landscape issues were a concern, and
a terrain model was created to include the proposed new road. The
microsimulation model was then prepared using the outline of the road
within the terrain model as the template. The second is of a small traffic
impact assessment model prepared to study the effect of adding
supermarket traffic to an existing junction. Budget was a major concern,
but the architects 3D model, in an industry standard file format, was
readily included in the simulation.
Figure 5 Simulation model with 3D terrain (left) and buildings (right; click each image to play
animation).
4. Results
Animations and smart images are the microsimulation output that users
expect, but while these are excellent for making side by side comparisons,
engineers require direct measures such as journey times, queue lengths and
the changes in vehicle distribution attributed to a design scheme. However
as microsimulation modelling includes the effect of natural variability of
road traffic, interpreting the results must take this into account.
Figure 6a shows the accumulation of queued vehicles on a route. Each line
on the chart represents a different run of the simulation using different
random seeds to represent daily variation within the model. The individual
runs can quickly be compared and aggregated in a number of ways. Figure
131
Journal of Maps, 2007, 122-134 Sykes, P.
6b shows the same data, but aggregated to produce an average of the 5
runs and with 95% confidence limits plotted. If, when an analysis is
performed for a design scheme, the new values for the number of vehicles
queued falls clearly outside this range, there is a definite result. If not, then
more statistical analysis is needed.
Figure 6 Vehicles queued: Individual runs (left) and mean with confidence intervals (right).
This mimics the behaviour of the traffic on the road - no two days are the
same and a queue may vary in length for no apparent reason. In order to
compare the queue before and after the changes to the road, average queue
length is taken. Using multiple simulations and analysing the variance
ensures that the significance of the difference in average length can be
determined; whether this is a real change or just due to random sampling.
Wide area impacts can also be determined. In larger models with route
choice, local changes resulting from the implementation of a scheme in one
part of a town can have an effect across the entire modelled area. Figure 7
shows the re-routeing effects created by upgrading the main orbital route in
an attempt to discourage through-trips in town. The traffic flow increases
are shown as green bars, and decreases as red bars. The width of the bar
shows the scale of the change. The amount of traffic in the model is
constant as there is no proposed development. The figure allows the
modeller to interpret the impact of the carriageway upgrade on the entire
132
Journal of Maps, 2007, 122-134 Sykes, P.
town; in this case it can be concluded that trips through the centre of town
have been displaced to the main orbital.
Link Flow Diff: -178 to 0
Link Flow Diff: 0 to 114
Showing 08:30:00 to 08:45:00
Figure 7 Analysis of driver routeing changes.
Conclusions
Microsimulation is now a widely accepted tool for assessing the benefits of
road scheme improvements. The output can be in the form of an animation
which can be feature-rich with 3D representation, or be functional and
sparse according to the needs of the intended audience. Analysis of
microsimulation model data can identify the empirical effects of changes to
the road network or travel demand, and hence the viability of the scheme
proposals. The accurate representation of road network layout and
geometry in microsimulation models is essential for the integrity of their
outputs, and mapping technology is an important ingredient for obtaining
reliable results.
133
Journal of Maps, 2007, 122-134 Sykes, P.
References
AIMSUN (2006) AIMSUN NG - The Integrated Traffic Environment
[online]. Available from: http://www.aimsun.com [Accessed: 6th June
2006].
CRAGG, S. (2007) How microsimulation modelling can address the real
world problems of navigation in congested networks, Traffic Engineering
and Control, 123-126.
DUNCAN, G. and McARTHUR, D. (1997) Paramics technical report,
EPCC-SIAS.
FRITZSCHE, H-T. (1994) A Model for Traffic Simulation, Traffic
Engineering and Control, 317-321.
NGSIM (2007) Home of the Next Generation SIMulation community
[online]. Available from: http://www.ngsim.fhwa.dot.gov [Accessed: 6th
June 2006].
S-PARAMICS (2007) S-Paramics microsimulation [online]. Available from:
http://www.paramics.co.uk [Accessed: 6th June 2006].
VISSIM (2007) PTV simulation [online]. Available from
http://www.vissim.de [Accessed: 6th June 2006].
WEBTAG (2007) Transport Analysis Guidelines [online]. Available from
http://www.webtag.org.uk [Accessed: 6th June 2006].
134
