Overview of the literature

Sonja Predin
August 9, 2018

1 Introduction
The problem of the understanding of daily mobility is studied by several groups
all over the world. I think that these studies are the most important Refs. [1–7].
Until now, studies are based on large-scale mobile phone data, with detailed
Geographic Information Systems (GIS) data, and GPRS data. From these data,
they got positions of starting and end points. This is the main difference to our
work, because we do not have exact coordinates. Also, these studies considered
urban areas, i.e. very big cities as Boston, San Francisco, Rio de Janeiro, etc.
While we study the small town, Hof and mainly rural areas of districts of Hof
and Wunsiedel (Landkreis Hof and Landkreis Wunsiedel). The main aim of
studies Refs. [2–4] is to develop algorithms suitable for calculation of travel
trajectories. Furthermore, in the conditions of trafic in urban area congested
travel is considered. Authors found also that an actual time is a function of a
free travel time and a ratio between the number of cars actually using a road
and it’s maximum flow capacity. In the recent studies were taken into account
that individuals sometimes break they routines Ref. [6]. This is done by using
physical models: Markov model and mobility entropy. To conclude, we also
consider time dependence of traveling, reason for traveling. Thus, our study
will be a significant contribution to the exacting literature.

2 Four-step model
The four-step model is a classical urban transport model which consists of four
steps:

1. trip generation - the goal of trip generation (production) is to estimate

the number of trips that are produced or originate in each Traffic Analysis
Zone. A set of equations is used to estimate the number of trips produced
by and attracted to each zone based on its residential and employment
characteristics. For example; the more employment a zone has, the more
work trips it attracts. The more retail employees in a zone, the more
shopping trips are assumed to be attracted there.

1
 2. trip distribution - this is the second component, where matches between
origins and destinations are developed. Trip ends are linked to create
complete trips. This can occur within the same zone, or between adjacent
zones with varying distance.
3. mode choice - mode choice predicts the choices that individuals or groups
make in selecting their transportation modes. An important objective is
to predict the share of trips attracted to public transportation. Other
factors considered for mode choice include: travel time, travel cost and
access to mass transit options.
4. route assignment - the final step is to determine the routes travelers choose
to reach their destinations.
For me the best reference for the four-step model is Ref. [1].

3 Origin-destination matrix
In this section I will give a brief overview of papers [2, 3].

3.1 Aim
The aim of these studies is to develop algorithms to generate road trips through
road networks using mobile big data. Authors find dilation of time in the state
of congested travel.

3.2 Origin-destion matrix

In Refs. [2–4] the trip there is defined when the same mobile phone user is
observed in two district zones within one hour. In Ref. [2] origin-destination
matrices are defined as matrices that display the number of trips from each
origin to each destination. For an example:

F rom/T o → 1 2 3 4 5
1 − 30 35 40 15
2 10 − 15 12 10
[t(i, j)] = (1)
3 50 40 − 35 20
4 25 30 35 − 40
5 45 30 35 40 −

In particular, they calculated the origin-destination matrix as:

Fijvehicle
ODij = W · PA (2)
all
ij Fij

where A is the number of zones. W is the one-hour total trip production in

the studied urban area. Directly from the mobile phone data they calculate

2
Tij (n), which is the total number of trips that user n made between zone i and j
during time of study. Via calibrating Tij (n) for the total population is obtained:
Pall
Fijall = n=1 Tij (n) · M (k), where Nk is the number of users in each zone k.
The ratio M scales the trips generated by mobile phone users in each zone to
the trips generated by mobile phone users in each zone to the trips generated
by the total population living there: M (k) = Npop (k)/Nuser (k), where Npop (k)
and Nuser (k) are the population and the number of mobile phone users in zone
k. Furthermore, to assign only the fraction of the trips attributed to vehicles,
we correct Fijall by the vehicle usage rate, which is a given constant for each
zone and therefore obtain Fijvehicle .

3.3 Estimating OD matrices

• Problems in mobile phone data
– Imperfect measurement of geographic position
– Incomplete contextual information and individuals
– Differences in daily usage of mobile phones among population
• Solution
– Given a user’s trajectory of spatiotemporal points

P = {p1 (x1 , y1 , t1 ), ...., pn (xn , yn , tn )} (3)

– A candidate set ∆ri,i+1 < δ.

S = {ps (xs , ys , ts ), ...., pt (xt , yt , tt )} (4)

– A candidate stay ∆tm,n < τ .

– Final stay point: centroid of points in S.

3.4 System architecture

• Parsing, standardizing and filtering
– Google’s open source Protocol Buffer library
– user data protocol buffer message
• Creating and storing

– rasterize polygons into a small pixel grid

– node and way elements in OpenStreetMaps
– cleaning steps: irrelevant residential roads

3
 Figure 1: Sheme of system architecture

3.5 Trafic jam

The ratio between the number of cars actually using a road (volume) and it’s
maximum flow capacity (volume over capacity or V /C). At low V/C, drivers
enjoy large spaces between cars and can safely free-flow speeds tf ree . As roads
become congested and V/C increases, drivers are forces to slow down to insure
they have adequate time to react. The actual time tactual in a road segment is
assumed to follow the Bureau of Public Roads function
 
b
tactual = tf ree 1 + a (V /C) (5)

where a = 0.15 and b = 4 are selected.

3.6 Weaknesses of the work

• Not enough background information about previous work. Especially
about four-step model and connection of this study to four-step model.
• Authors ignore consideration other types of vehicles and it is not clear did
they consider that few people can travel together in a car.

4
 Figure 2: Traffic jam in Belgrade

4 Understanding congested travel in urban ar-

eas
In this section I will give a brief overview of papers [4].

4.1 Aim
They analyze how efficiently people move across cities. They parse Open-
StreetMap data, use origin-destination matrices mined from call detail records
data and route trips in the road networks. They estimate morning peak ve-
hicular volumes from origins to destinations and compare the inferred travel
times based on demand with estimates of an online map provider in the respec-
tive routes and hour of the day. Then, they explore the relationship between
distance and travel time. They studied the inefficiency of selfish routing by
comparing obtained travel times to those that would be observed if the routes
were selected to attain the social optimum.

4.2 Approach
The cost function is defined as
X
C= xe te (xe ) (6)
e∈p

where e represents the road segment e along the chosen route p. te (xe ) represents
the travel time t observed on road e for vehicle flow xe . The typical traffic

5
problem is formalized as follows:
XZ xe
minimizexe ∀e ∈E te (xe )dx (7)
e∈E 0
X
subject to fpst = f st . (8)
p

where fpst is the flow between the source s and target t on route p.
XXX
xe = fpst δ st (p, e) (9)
s t p

where xe ≥ 0 and fpst ≥ 0. Here δ st (p, e) = 1 when road e lies on the route p.

4.3 Algorithm B(N)

Algorithm B(N)
Initialize B as the shortest path tree rooted at the origin. Assign all flows to
links to B.

while rg > 0.001 (10)



 forall origins o
 Add to B0 edges e with negative reduced costs.

do (11)
 do
 Solve the Restricted Master Problem for B0
Simplify B0 by removing {e | xe = 0} .
 

Restricted Master Problem(Bush B, )

• Uptade costs on all links on B.
• Calculate the longest route tree with paths Pi and costs Ui .

• Calculate the shortest route tree with paths pi and costs ui .

if max {Ui − ui , 6} ∀i , stop.

else continue.
for all j


 set of links in pi not in Pi : Si = pi \ Pi ,



 set of links in Pi not in pi : Li = Pi \ pi ,


 difference in costs to j : g = (uj − ui ) − (Uj − Ui ),
0

 total marginal cost of sets S and L h = P
j j e∈Sj ∪Lj ce ,
do

 flow to be shifted : dx = min {g/h, min {xe \ e ∈ Lj }} ,
add flow to shorter path: xe = xe + dx, e ∈ Sj ,




remove flow to shorter path: xe = xe − dx, e ∈ Lj ,




update travel times: te , e ∈ Sj ∪ Lj .


6
4.4 Demand to supply ratio
Γ as a single dimensionless parameter, captures the load on the road infras-
tructure by bringing together trip distances, trip magnitudes, road capacities,
and the distances they span. It also helps explain navigation speed in cities for
varying distances, with α essentially describing the sensitivity of the city to the
stress imposed by travel demand on its roads.
P
le x e
Γ = P e∈E (12)
xe >0,e∈E le Ce

where le and Ce are the length (km) and the flow capacity (vehicles per hours)
of a road segment e.
Here they investigate a typical relationship to test the common conception that
cities with higher population densities ρ tend to be more congested. Their
findings show that Γ is a better predictor than ρ as it lacks the outlier problem
and provides a reasonable trend relating to overall congestion.

4.5 Weight of social good.

Next, they want to explore the potential of routing solutions on congestion
alleviation. To this end, they use a model that represents cost as a linear
combination of the actual cost and the marginal cost of a selected path.
d [xe te (xe )]
cλe (xe ) =(1 − λ)te (xe ) + λ (13)
dxe
dte (xe )
=te (xe ) + λxe (14)
dxe
They refer to λ, the combination parameter in [0, 1], as the weight of social
good. A driver with λ = 1 chooses routes with respect to the marginal costs,
thus moving the system closer to the system optimum, and λ = 0 refer to
opposite situation.

4.6 Weaknesses of the work

• Authors ignore consideration of other types of vehicles and it is not clear
did they consider that few people can travel together in a car.

5 Understanding individual routing behavior

In this section, I will give a brief overview of papers [5].

5.1 Aim
Here they study 92,419 anonymized GPS trajectories describing the movement
of personal cars over an 18-month period. Their group user trips by origin-
destination and they find that most drivers use a small number of routes for

7
routine journeys, and tend to have a preferred route for frequent trips. In con-
trast with the cost minimization assumption, they also find that a significant
fraction of drivers’ routes is not optimal. These basic discoveries can inform re-
alistic route choice models that are not based on optimization, having an impact
on several applications, such as infrastructure planning, routing recommenda-
tion systems, and new mobility solutions.

5.2 Aproach
The same as in the previous paper Ref. [4].

6 Data-driven generation of spatio-temporal rou-

tines in human mobility
In this section, I will give a brief overview of papers [6].

6.1 Aim
Previous studies did not take in account for the possibility that individuals
may break the routine during periods of variable duration. In this article, they
present DITRAS, (DIary-based TRAjectory Simulator), a framework to simu-
late the spatio-temporal patterns of human mobility, available (https://github.com/jonpappalord/DITRAS).
Ditras is composed of two main steps. During the first step, the diary generator
builds a mobility diary D of N time slots, each of duration t. During the second
step, Ditras uses the trajectory generator and a given spatial tessellation L to
transform the mobility diary into a sampled mobility trajectory, see Figure 3.
We use a Markov model to describe the probability that an individual follows
her routine and visits a typical location at the usual time, or she breaks the
routine and visits another location, see Fig. 4.

6.2 The mobility entropy

The mobility entropy S unc of an individual u is defined as the Shannon entropy
of her visited locations:
∈ L(u) pi log(pi )
P
unc
S (u) = i (15)
log | L(u) |

where pi is the probability that individual u visits location i during the period
of observation and | L(u) | is a normalization factor. The mobility entropy
of an individual quantifies the possibility to predict individual’s future where-
abouts. Individuals having a very regular movement pattern possess a mobility
entropy close to zero and their whereabouts are rather predictable. Conversely,
individuals with a high mobility entropy are less predictable.

8
References
[1] Michael G. McNally (2007), The Four-Step Model,
http://www.its.uci.edu/its/publications/papers/CASA/UCI-ITS-AS-WP-
07-2.pdf
[2] Pu Wang, Timothy Hunter, Alexandre M. Bayen, Katja Schechtner, and
Marta C. Gonzalez, Understanding Road Usage Patterns in Urban Areas,
Scientific Reports 2:1001 (2012).
[3] Jameson L. Toole, Serdar Colak, Bradley Sturt, Lauren P. Alexander,
Alexandre Evsukoff, and Marta C. Gonzalez, The path most traveled: Travel
demand estimation using big data resources, Transportation Research Part
C: Emerging Technologies 58, 162 (2015).

[4] Serdar Çolak, Antonio Lima, and Marta C. Gonzalez, Understanding con-
gested travel in urban areas, Nature Communications 7, 10793 (2016).
[5] Antonio Lima, Rade Stanojevic, Dina Papagiannaki, Pablo Rodriguez,
Marta C. González, Understanding individual routing behaviour, J. R. Soc.
Interface 13: 20160021.

[6] Luca Pappalardo, and Filippo Simini, Data-driven generation of spatio-

temporal routines in human mobility, Data Mining and Knowledge Discovery
32, 787 (2015).
[7] Minjin Lee, Hugo Barbosa, Hyejin Youn, Petter Holme, Gourab Ghoshal,
Morphology of travel routes and the organization of cities, Nature Commu-
nications 8, 2229 (2017).

9
Figure 3: Sheme of DITRAS algorithm

10
Figure 4: Sheme of Markov model

11