Académique Documents
Professionnel Documents
Culture Documents
EXAM
Please return this exam with your answers!
1a b c
2a b c
3a b c
4a b c
10
10
10
20
4 2 4
2 4 4
2 4 4
d 5a b c
10
4 6 4 6 1 3 6
6a b c
7a b c d 8a b c
9a b c
10
20
10
4 2 4
15
4 4 6 6 3 6 6
3 3 4
Please start each question on a new page and keep your answers brief (and readable!). Question 9 is a BONUS question, with it you can compensate
for lost points in other questions, but you dont loose points if you miss it (the maximum score still is 100%)
a) The news spreads fast via social media, and around 16:00 a large amount of people decides to leave
TU Delft. Argue which are the bottlenecks (and/or bottleneck seeds) for this case
b) Propose traffic management solutions to prevent (mitigate) these bottlenecks from oversaturating.
Hint 1: consider which roads youre going to use for which (traffic management) purpose; Hint 2:
consider the four traffic management solution directions;
c) At 16:30 the crisis team decides that the entire TU Delft area needs to be evacuated immediately for
safety purposes. Which bottleneck (seeds) do you expect now to become active?
d) Sketch a traffic management strategy from 16:30 onwards for this scenario. In this strategy list
which roads/routes are used for what (traffic management) purpose; which routes are prioritized and
which traffic management solutions (on which routes/locations) you will deploy.
TRAFFIC STATE ESTIMATION
5) The figure below shows a control loop, and
the different variables (state variables x,
observations / measurements y, disturbances
d, control signals u) that play a role in it.
a) What are these so-called state variables?
(give a definition)
State variables are those variables that
uniquely and completely describe the
time evolution of the process that is controlled.
b) Consider a simple network with two alternative routes between A and B. Suppose our objective is to
distribute traffic coming from A with direction B over these two alternative routes, such that the total
delays for all drivers are minimized. To achieve this objective we provide all drivers with the
expected delay on the two alternative routes (e.g. using a dynamic route information panel).
Motivate which of the following variables could be used as state variables in this controller:
Consider what process is controlled: the distribution of traffic over (i.e. the amount of vehicles on)
the two routes. So we need state variables that uniquely describe the time evolution of that process
given our control signal (queue length information)
i)
No, this is typically a disturbance (although you can argue that the quality of the route advice
may in time affect the sensitivity of users and therefore the effectiveness of the controller)
c) The adaptive smoothing method (ASM) is an example of a traffic state estimation method. Explain
in your own words how this method works (use a figure to explain if needed)
See lecture slides
TRAFFIC CONTROL APPROACHES
6) One could classify different control approaches along two dimensions (see figure below). The first
dimension relates to the degree of adaptiveness (i.e. whether the control approach uses feedback and/or
adapts its inputs/parameters to the situation at hand). The second dimension relates to whether the
approach is reactive (based on past information only) or proactive (based on predictions). Classify the
following control applications along these two dimensions and motivate briefly (motivation is as
important as the answer):
a) The SPECIALIST algorithm (dynamic speed limits to solve wide moving jams)
Mid-slightly above bottom: There is a bit of short-term prediction involved (shockwave theory to
compute the speed limit and length of the speed limited area), and you must make sure that the used
fundamental diagram fits with the location specific circumstances. But in terms of control its still
open-loop (no feedback)
b) Fixed-time intersection control
Left-bottom: its reactive and open loop. You may argue (but than the argument needs to be precise)
that it is predictive (topleft) in case there are
different phase plans for
different times of the day
a
c
c) The coordinated ramp
metering / intersection
b
control approach used in
the PPA (large scale field
operational test
Amsterdam)
Also in the middle, but a
bit more to the right (not entirely though!). There is a bit of prediction (which bottleneck is about to
saturate) and there is (local) feedback
7) Isolated traffic management measures have been effectively deployed to solve (local) traffic problems.
However, effectiveness of isolated controllers is limited due to a number of reasons. One of the reasons
is that a measure (e.g. ramp metering) cannot be deployed for a considerable time due to storage
limitations (metering needs to stop once the on-ramp is full of traffic).
a) Name and discuss at least two other reasons that limit the effectiveness of isolated control measures.
Effect local measures may simply not be sufficient enough; Local effects may be counteracted by
problems in other parts of the network (e.g. oversaturation intersections, spillback or activation of
bottlenecks downstream); Travelers response may counteract the measure (rat-running)
b) Describe how the lack of buffer space could be improved using the coordinated deployment of
measures. You can give an example to illustrate your description.
Essentially coordination gives you additional storage space: on alternative routes; on the buffer
space at other ramps or intersection controllers.
Suppose the objective is to develop a Model Predictive Controller to compute the control signals for the
measures that are to be jointly deployed.
c) Sketch the system (the MPC traffic control loop) and its constituent elements and indicate how
policy objectives can be catered with an MPC approach.
See sheet 26 of lecture slides 3 for the scheme. Policy objectives can be realized via the objective
function of an MPC controller (e.g. minimize total time spent, or any other objective that you can
formulate, e.g. maximizing travel time reliability, minimizing total amount of emissions, equalizing
distribution of queues, etc)
d) Describe some pros and the cons of an MPC approach.
Pros: MPC is automated process and uses a traffic flow model to find the optimal set of control
measures and parameters in an entire network over a longer time period (this is virtually impossible
to do for a human); MPC combines feedback (about its own control computation) with prediction (it
optimizes control for a long time period instead of just the here and now);
Cons: MPC works well for nice linear processes, but traffic is a highly nonlinear and complex
process; MPC is computationally very complex, huge nr degrees of freedom; MPC provides no
motivation / explanation for a computed control scenario (i.e. difficult to explain)
8) The SPECIALIST algorithm uses dynamic speed limits to solve wide moving jams on freeways. A wide
moving jam is a short traffic jam with a high density/low speed that propagates upstream for many
kilometres.
a) Explain in your own words why it is beneficial to remove such wide moving jams
These WMJ reduce capacity by up to 30%. Removing them thus leads to large efficiency gains (and
also environmental gains: avoiding steep acc/dec)
b) Explain in your own words how SPECIALIST resolves these wide moving jams (No formulas are
asked - just the principle idea(s) behind this algorithm). Be brief (use a figure if needed)!
The basic principle is very simple: since q=ku: lowering u without changing k will reduce the inflow
into a WMJ. If inflow < outflow the WMJ will resolve (this is 75% of the answer). The additional
25% you earn with logical statements discussing e.g. the different steps / phases or an insightful
picture
c) Motivate whether the following statement is TRUE or FALSE: Mainline metering (i.e. using traffic
lights on the freeway) can achieve exactly the same gains as the SPECIALIST algorithm
FALSE - The key here is that SPECIALIST removes the capacity drop (i.e. traffic operations are
resolved to the freeflow capacity). Mainline metering at best resolves operations to the discharge
capacity, i.e. it could trap the WMJ to a fixed location (a buffer). You can easily verify with
shockwave theory what could happen under different scenarios of inflow.
The setup is as follows: we need data (e.g. from loops, or manual observations, etc) to construct
realistic scenarios in which we will compare the different cases (feedforward RM vs feedback
RM & no coordination vs coordination). For that we need realistic traffic demand patterns
upstream and at the ramps; and along with these realistic initial traffic conditions along A13. We
also need the current settings of the RM algorithms.
ii)
You could think of a micro-simulation model of the freeway stretch (FOSIM or VISSIM) in
which we can implement ramp metering. A macroscopic simulation would also work (e.g. a first
order traffic flow model). Of course, we need to make sure this model is predictively valid, so
we need to calibrate the model with the data. Since the main measure of effectiveness relates is
Total Vehicle Loss hours, my strong preference would be to use a macroscopic traffic flow
model (we dont really need detailed vehicle trajectories to compute TVLH) But theres no
fundamental argument against using a microscopic model (in that case you could do more
scenarios, e.g. related to vehicle composition)
iii)
Given we proper data to calibrate its parameters (fundamental diagrams capacity, critical
speeds, etc) I am quite positive about the validity of a study with a macroscopic model.
iv)
I am quite positive about the validity effectiveness of coordination, it stands to reason that if you
have more buffer space (at more ramps) and if you use these in coordination, the metering time
will be much longer