Q1. Demand Modelling.

In this question, we will study origin-destination (O-D) demand models of the proposed
city ‘The Line’ in Saudi Arabia. We suppose that the city is organised around a straight
double-track metro line of length L = 170 km, running from x = 0 to x = L, with N
uniformly spaced metro stations at positions xi = (i − 1)L/(N − 1), i = 1, 2, . . . , N .
We will suppose that the line has a working population of P = 5, 000, 000 individuals,
and that their residences and jobs are uniformly distributed along the line so that each
station provides a source of Oi = P/N workers and a sink of Di = P/N jobs. The
distance between two metro stations is clearly |i − j|L/(N − 1), however we will assume
a commute distance of dij = (1+|i−j|)L/(N −1), to account for an additional commute
distance at each end of the metro ride. This formula with j = i, so that dii = L/(N −1),
will also be used to model a short commute distance for journeys entirely within a single
metro station locality. Throughout the question, you may assume that L and P are
fixed. However, N is a parameter that you should experiment with. For example,
N = 10 would correspond to a sparse provision of metro stations where an additional
‘last-mile’ transit solution would be needed, whereas N = 100 would ensure that nearly
all residences and jobs are within walking distance of the nearest metro station.
(a) (Hand calculation.) Consider the application of the classical gravity model where
O-D flows fij from locality i to locality j are prescribed in the form
Oi Dj
fij = c .
d2ij
Show that no choice of c will simultaneously satisfy all of the 2N constraints
X X
Oi = fij and Dj = fij .
j i

(4 marks)
(b) We will now consider the doubly-constrained gravity model where
(O) (D) Oi Dj
fij = ci cj .
d2ij
(O) (D)
involving 2N constants ci and cj .
(i) (Hand calculation.) By using the formulae for Oi and Dj in part (a), derive
a system of 2N nonlinear equations to solve for the constants.
(2 marks)
(ii) Hence design, develop, explain, and exhibit code to solve for the constants
(O) (D)
ci and cj and hence find and exhibit the flows fij . Hint: you may either
use a general nonlinear system solver such as fsolve, or implement your own
numerical method — for example, search up the iterative balancing method
(also called the Furness method ).
(8 marks)
(c) Design, develop, explain, and exhibit code that computes the resulting flow on each
link of the metro line. Without producing full calculations — explain how flows
will work out differently if instead the stations (and localities) are equally spaced
along a circular double track.

(6 marks)

2
Q2. Network modelling.

(a) This question considers the metro line design mooted in Q1(c) where N stations
are equally spaced along a circular double track of length L. Again N = 10 and
N = 100 may be considered representative choices for your experiments.
Suppose that the city in question has a budget to build a length βL of additional
double track, in the form of short-cuts, that is straight-line chords that cut directly
across the circle between stations. Here β is a parameter to be experimented with.
Clearly a very large value of β will enable a direct link between every pair of
stations — not at all a realistic situation — and a small value of β may enable
only one very short link to be built between near-neighbour stations — not at all
a useful intervention. The interesting regimes are in between.
Design, develop, and explain code to experiment with different designs for intro-
ducing the short cuts as β is increased, with implementation in the Matlab graph
toolbox.
Hint: a good answer will either propose a couple of different heuristic principles,
or will propose a systematic procedure, for introducing the short cuts.
For your designs, develop, explain, and exhibit code that computes the shortest
path between pairs of nodes and the average shortest path length over all N (N −
1)/2 distinct (ordered) pairs of nodes. Note that because all of the track is double,
you may work with an undirected graph, or a directed graph with a symmetric
edge matrix, as you choose.
Hence evaluate the efficacy of your designs as β is increased.
(13 marks)
(b) This question concerns the proposed South Bristol Liveable Neighbourhoods — a
form of LTN (low traffic neighbourhood). See
https://www.bristol.gov.uk/residents/people-and-communities/
liveable-neighbourhoods/south-bristol-liveable-neighbourhood
See also the popular media, e.g.
https://www.bristolpost.co.uk/news/bristol-news/
parking-bans-one-way-streets-10477427
Choose one of the proposed sub-zones — for example, Southville, Bedminster
East, Perretts Park etc. — and using OSM data, build and exhibit a Matlab
graph representation of the chosen zone.
Describe briefly the sorts of computations you might perform and principles you
might invoke to evaluate the impact of the proposed scheme. (Full workings /
computer code not required.)
(7 marks)

3 turn over/. . .
Q3. Assignment modelling.
In this question, we will consider the classical two-link parallel network as introduced
in the lecture videos. The (total) flows of traffic on each link will be x1 , x2 ≥ 0 and
the (per user) link cost functions will be f1 (x1 ) := a + x1 and f2 (x2 ) := 1 + bx2 ,
with 0 ≤ a, b ≤ 1. The flows must sum to a total prescribed demand d ≥ 0, so that
x1 + x2 = d.
In contrast to the usual set-up, we will suppose that the demand is split into two
components dA := αd and dL := (1 − α)d, where 0 ≤ α ≤ 1 is a parameter that you
will experiment with. The component dA corresponds to autonomous vehicles that will
be programmed to act altruistically (to try to minimise total system cost for all users)
and the component dL corresponds to legacy vehicles that are driven by humans who
are assumed to be self-optimising in the usual user equilibrium (UE) way. Each link
(A) (S)
flow will thus consist of two components so that xi = xi + xi and we will suppose
that the per user cost experienced by each vehicle on the link is a function fi (xi ) of
the total (autonomous and legacy) vehicle flow on that link. We will explore solutions
of this mixed-flow system. Note that α = 0 corresponds to the standard UE problem,
whereas α = 1 corresponds to the standard system optimal (SO) problem.
Consider the following Stackelberg game.
• STEP 0. Compute the usual system optimal SO assignment to determine the total
flow on each link and assign flows so that each link has the same proportions of
autonomous and legacy traffic. Go to STEP n PART 1.
• STEP n PART 1. All the autonomous vehicles maintain link choices from the
previous step. The legacy vehicles are re-assigned between the two links so that
they are in equilibrium. Go to STEP n PART 2.
• STEP n PART 2. All the legacy vehicles maintain link choices from the previous
step. The autonomous vehicles are re-assigned between the two links so as to
minimise the total system cost (i.e., total cost of autonomous and legacy vehicles).
Go to STEP n PART 1.
The supplementary materials provided on Blackboard show how this game plays out for
the classical choices a = b = 0 and d = 1, with an equal split α = 1/2 of autonomous
and legacy traffic.
(a) Design, develop, explain, and exhibit code to simulate the Stackelberg game and
explore its converged assignments. Examine how the total system cost of the
converged assignment reduces as α is swept from 0 to 1. Your calculations should
explore the a, b, d parameter space, and note any significant material differences,
trends etc as these parameters are varied. Discuss potential practical implications
of your results.
(14 marks)
(b) In the real-world, autonomous vehicles may be able to collaborate to drive partic-
ularly close together, but only if no legacy vehicles are present. You should model
this by supposing that if a link has only autonomous vehicles, the congestibility
parameter (i.e., the coefficient of x in the link cost function) is halved. Adapt your
part (b) solution to explore this effect and comment on any relevant real-world
implications.
(6 marks)

4
Q4. Microscopic modelling.
There is a demonstration taking place with an unknown number of participants, N .
Based on the following observations:

• The crowd completely fills the whole of Westminster Bridge (250m long; 26m
wide), three times over (i.e., only one third of the crowd fits onto the bridge at
the same time).
• It takes around one hour for the crowd to pass through a cross-section of the
bridge.

Assume that the walking speed of the crowd can be described by v(ρ) = 1.3(1−ρ/ρmax )2 ,
ρmax = 6m−2 .
(a) Estimate the average density of the crowd ρ.
(3 marks)
(b) Estimate the total crowd size, N .
(1 mark)
Assume that γ=5% of the crowd participants are wheelchair users, who occupy a cir-
cular space with a 0.75m radius. Calculate the maximum crowd flow for the following
two design choices:
(c) • Wheelchair users blend with the crowd and use the whole area (assume that
wheelchairs will move at the same speed as the pedestrians).
• 2m out of the 26m width of the bridge is desginated to wheelchairs (who
move in a single lane). Assume that wheelchairs move at a constant speed of
0.7 m/s.
(4 marks)
(d) What is the critical value of γ where it is a net-positive (for overall flow) to have
designated wheelchair space?
(4 marks)
(e) In order to do more detailed analysis of the problem, specify the equations of a
modified social force model (any published specification of the social force model
is acceptable) with the following changes:
• Simulate γ=5% wheelchair users; blended within the crowd; occupying a
circlar space with a 0.75m radius each; and adhere to the following kinematics
(can move forward, backward, and rotate, but can not move laterally).
• A fraction α of the participants are particularly loud. When anyone comes
within a 5m distance of those individuals, they will either speed up by 20% (if
they are ahead) or slow down (if they are behind) the noisy individual, until
they are >5m away from them.
(8 marks)

5 turn over/. . .
Q5. Data-driven modelling

For all parts of the question, provide your discussion in bullet points.
(a) Consider the plot shown in figure 1 below. It shows the observed vehicle traffic count
(solid line, arbitrary units, a.u.) and the traffic count forecast by two different
time-series models (red dashed and blue dotted). Both models are fit to the same
data and neither model is a black-box machine learning model.
vehicle count (a.u.)
2 3 4 5 6 7

0 50 100 150 200 250 300

0 50 100 150 200 250 300

Figure 1: Red and blue model fit to the same data.

(i) Use the plot to list as many properties of the red and blue models as you
can identify. Clearly justify which aspects of the plot are indicative for each
feature and why and write down indicative equations for each model (there is
no expectation for exactly correct model terms and estimated parameters).
(8 marks)
(ii) Discuss the usefulness of fitting these two models to the same data for ex-
plaining the dynamics of vehicle counts, clearly justifying the arguments you
make.
(4 marks)
(b) This questions extends the worked example on machine-learning a microscopic
model covered in the lab sheet for data-driven modelling (section 3 on the lab
sheet for data-driven modelling).
(i) Develop and implement in Matlab or Python an improved version of the
worked example in the lab sheet that considers the exit location and per-
forms closed loop predictions instead of the open loop predictions shown in

6 Qu. continues . . .
 the lab sheet. Provide a clear explanation and rationale for the changes you
make.
(4 marks)
(ii) Evaluate the performance of your model by additionally using another data
set of your choice obtained from https://ped.fz-juelich.de/db/doku.php.
Justify your choice of data and evaluation approach (note that marking places
a higher emphasis on the quality of the evaluation than on the prediction
accuracy).
(4 marks)

7 turn over/. . .
Q6. Practitioner studies.
For all parts of the question, please provide your answers as bullet points rather than
continuous text. Your answer to this question should not be longer than 3 pages A4
(less than approx. 1,500 words).
The Green Guide by the Sports Grounds Safety Authority is a UK Government-funded
guidance book on spectator safety at sports grounds. It provides figures for the egress
flow rates of pedestrians moving on level and stepped surfaces.
(a) Find the egress figures mentioned below (they are available online), state them,
and provide your source. Critically discuss these egress flow rates.
(5 marks)
(b) Consider you are a consultant who has been given the following brief: “The city
of Clowntown has decided to update their multi-purpose events hall that is used
for sports, concerts, and other events. The main change will be a second level of
tiered visitor seating above the first level that is also tiered. Currently, the hall
has a capacity of 2,000 spectators, the extension will double this capacity with the
second level seating arrangement mirroring the first level arrangement. Due to the
age of the building, regulatory requirements state that it must be possible to empty
the hall in 2 minutes. There is ample walkway provision around the entire hall.
Your task is to make suggestions for the number, design, and locations of exits.
Please provide justifications for your suggestions. Your customer is concerned with
safety, construction costs, and user experience in this order of importance. You
can find a simplified site plan below (figure 2), the hall currently has four identical
exits.”

Figure 2: Site plan of Clowntown’s multi-purpose hall.

Use the information from the Green Guide, what you have learned on the course,
and any additional relevant information from beyond the course to address the
brief above, focusing on elements relevant to spectator foot traffic and considering
the relevant stakeholders. Explain and justify your approach.
(8 marks)
(c) If you were to be given access to one other model covered in the lecture slides on
microscopic modelling to make your design recommendation, which model would

8 Qu. continues . . .
 it be, and why? Stick to the same brief as in part (b) and state clearly where in the
lecture notes the model was covered. We expect a sufficiently detailed explanation
as to why you choose this model including an explanation on how it would be used.
(7 marks)
Guideline to allocation of marks for part:

• These are open-ended questions, there is not one correct answer. You are be-
ing marked on the level of knowledge, application, and critical appraisal of the
methodological background.
• Excellent answers will go beyond what is explicitly taught on the course.

9 .../ End of Coursework