Course: CT4801 Name: Transportation & Spatial Modeling

Date: Time:

1 November 2007 9h00 12h00

For this exam you are allowed to use a pen, paper, a calculator, and a formula sheet. Books and lecture notes cannot be used during the exam. The exam consists of 7 questions on 7 pages for which in total 100 points can be gained. It is allowed to answer the questions in Dutch.

Question 1: [20 points]

A study area is typically divided into zones. There are two types of zones, internal zones that are of main interest, and external zones to connect the study area to surrounding areas. Suppose we are interested in determining the trip generation of commuters in the morning peak. For the internal zones a regression model is used to determine the trip productions and attractions. (a) Name two zonal attributes or variables that are important to include in the regression model for computing the trip production for the internal zones. [4] Examples of zonal attributes important for computing trip production are number of households, (average) level of income, number of car owners, value of land, etc. (b) Name two zonal attributes or variables that are important to include in the regression model for computing the trip attraction for the internal zones. [4] Examples of zonal attributes important for computing trip attractions are number of offices, number of retail shops, number of schools/universities, etc. (c) How can the trip production and attraction be computed for the external zones? [2] By traffic/traveler counts. (d) Besides the trip purpose commuting, name three other trip purposes for which separate regression models for trip generation have to be estimated. [3] Other trip purposes can include shopping, recreational, social, educational, business, etc. The Dutch national transportation model LMS uses tours instead of trips in their generation model. (e) What is the difference between trips and tours? [3] A trip has a single origin and single destination, which are often unalike. A tour is a cyclic chain of trips. For instance, a trip can be made from home to work,

1 - 12

whereas a tour may consist of three consecutive trips from home to work, work to shop, and from shop to home. (f) What type of trip generation model does the LMS use? [2] The LMS model applies a binary logit model (stop/repeat) to compute the trip generation. (g) Name one other type of trip generation model that could be applied. [2] Other models may be a regression model or cross-classification model.

Question 2: [12 points]

Consider cities A and B that are connected by a local road. Furthermore, between these cities is a train service available with a frequency of 3 trains per hour (departing each 20 minutes). The travel time by car is 60 minutes, and by train is 40 minutes. The total number of travelers between A and B is 1000. car
A train 3x per hour B

In order to determine the modal split, a multinomial logit model is used with the following estimated utility functions: V car = 1.0 0.09TC V train = 0.12 A 0.10TT 0.16 F where TC is the in-vehicle travel time by car [min.], TT is the in-vehicle travel time by train [min.], A is the average access waiting time [min.] (waiting time to board the first train) F is the average transfer waiting time [min.] (waiting time between two trains) (a) Determine the number of travelers taking the train. [4] The train service departs every 20 minutes. Given that the arrival at the station and departure of the train are unsynchronized (note that only a frequency schedule is given, not a time schedule), the average access waiting time is half the headway and thus equal to 10 minutes. The utility for each alternative can be computed using the given utility functions: V car = 1.0 0.09 60 = 4.4 V train = 0.12 10 0.10 40 0.16 0 = 5.2 The share of travelers taking the train is computed using the logit model:

2 - 12

e 5.2 = 0.31 car train e 4.4 + e 5.2 eV + eV The number of travelers taking the train then equals 0.31 1000 = 310 travelers eV
train

Suppose that a high-speed rail connection is opened such that travelers from city A to city B can travel by a new train service. This new service is not a direct connection, but a transfer in city C is required, as illustrated in the next figure. The travel times (A,C) and (C,B) are both 10 minutes. The high-speed train service from A to C has a frequency of 2 trains per hour (departing each 30 minutes) and the high-speed train service from C to B has a frequency of 4 trains per hour (departing each 15 minutes).

car train A B 3x per hour 2x train n i r pe t r a hou rh r e ou xp r 4 C (b) Using the same utility functions and again applying a logit model, what percentage of train users will use the high-speed train service? [5] The average access waiting time now becomes 15 minutes. Similarly, the average transfer waiting time (given that the services are not synchronized) is half the headway and thus equal to 7.5 minutes. The utility to use the highspeed train service is:
V train = 0.12 15 0.10 20 0.16 7.5 = 5.0 The share of train travelers taking the high-speed train is: hs train eV e 5.0 = = 0.55 train hs train e 5.2 + e 5.0 eV + eV Since the train alternatives cannot be seen as independent alternatives, a simple multinomial logit model is not the correct model to apply and may make incorrect forecasts about the modal split between car and train. (c) How can the model be improved such that the modal split rates between car and train are more accurate? [3] The two train services can be combined to a new combined (virtual) train service, where the costs of this new train service can be computed using, for instance, the minimum costs, or logit analogy. Then the share of travelers choosing the car alternative and (virtual) train alternative (i.e. the modal split) can be calculated as before, after which the share of train travelers can be distributed over the two train services using a logit model. This corresponds to applying a nested logit model.

3 - 12

Question 3: [16 points]

Three zones, A, B, and C, are connected by roads and rails as indicated below. Each (uni-directional) link represents both a road segment as well as a rail segment, and the distances are indicated (for simplicity it is assumed that for each link the length of road and rail are equal).

15

15

30

C The trip production and attraction for each zone is given in the table below.
Zone A B C Production 2800 3600 1550 Attraction 2450 4260 1240

Furthermore, the trip distribution functions for each mode (car and train), which described the accessibilities Fij from zone i to zone j for that mode as a function of the travel distance cij [km], are indicated in the next figure. Assume that all intrazonal distances are equal 5.
F ij 10 8 6 4 2 0 0 10 20 30 40 50 c ij train

car

(a) Determine the skim matrix c = [cij ] consisting of travel distances. [3] Determine the travel distances from the 1st figure. Note that all intrazonal distances are assumed equal to 5.

4 - 12

[cij] =
A B C A 5 45 15 B 15 5 30 C 45 30 5

(b) Compute the simultaneous trip distribution / modal split based on the doubly constrained gravity model for obtaining the trip matrices for the car and train (perform only 1 iteration). [8] Use the 2nd figure to convert the skim matrix from (a) into the accessibility matrix. Note that we now distinguish car and train, since the accessibility function is mode specific. [Fijv] =
A A B C car 10 1 6 train 5 2 4 car 6 10 3 B train 4 5 3 car 1 3 10 C train 2 3 5

Then calculate the (mode specific) origin-destination (OD) matrix based on the accessibility matrix, and perform one iteration. (Note that the productions equal the attractions, such that trip balancing is not needed.) [Tijv] =
A A B C Attraction car 10 1 6 train 5 2 4 2450 car 6 10 3 B train 4 5 3 4260 car 1 3 10 C train 2 3 5 1240 factor 2800/28 3600/24 1550/31 Production 2800 3600 1550

A B C Attraction

A car train 1000 500 150 300 300 200 2450

B car train 600 400 1500 750 150 150 4260

C car train 100 200 450 450 500 250 1240

factor 2800/28 3600/24 1550/31

Production 2800 3600 1550

A B C factor Attraction

A car train 1000 500 150 300 300 200 2450/2450 2450

B car train 720 480 1800 900 180 180 4260/3550 4260

C car train Production 2800 63.6 127.2 3600 286.2 286.2 1550 317.9 159.0 1240/1950 1240

5 - 12

(c) Compute and sketch the trip length frequency distribution for the car (choose your own length classes, use at least three classes). [5] The skim matrix in (a) shows the travel distance per OD pair, while the OD matrix in (b) shows the number of trips performed per OD pair. By combining these two matrices the trip length frequency distribution can be computed. For example, choose classes of size 20. Then the trip length frequency distribution for the mode car will look as follows:
bin 0 - 20 20 - 40 40 - 60 no. trips 4137.9 466.2 213.6 OD-pairs AA,AB,BB,CA,CC BC,CB AC,BA

trip length frequency distribution

5000 4000 frequency 3000 2000 1000 0 0 - 20 20 - 40 trip length 40 - 60

Question 4: [18 points]

Consider the following transportation network with two origins (A and B) and two destinations (B and C).
link 2 A link 1 B link 3 link 5 C

link 4

Each link in the network is assumed identical and the travel time on each link a is given by ta as a function of the link flow qa ,

ta (qa ) = 1 +

qa , a = 1, 2,3, 4,5. 2000

6 - 12

The travel demand trip matrix is given as follows: from/to A B C

A B 500 C 2000 3500 -

A stochastic assignment can be determined using the logit model or the probit model. (a) Determine the link flows in a stochastic assignment using the logit model. [4] For each OD-pair we can check the possible routes. This leads to the following route choice set:
OD-pair AB AC route 1 1 2 3 1 2 links on route 1 1,2 1,3 4,5 2 3

BC

For OD-pair AB only one route exists. Therefore all travelers will use this route. For OD-pair AC three routes exist, whereas all three routes are equal in length (and thus equal in travel time, since for the stochastic assignment the travel time is independent of the flow). Since all routes are equal, the travelers between origin A and destination C will be equally distributed over these three routes. Finally, for OD-pair BC two routes exist of equal length. Therefore, similarly, the travelers between origin B and destination C will be uniformly distributed over these two routes. The link flows can then be computed from the route flows, by looking at the routes which use a certain link. link 1: link 2: link 3: link 4: link 5: 500 + 1/3 * 2000 + 1/3 * 2000 = 1833.3 1/3 * 2000 + 1/2 * 3500 = 2416.7 1/3 * 2000 + 1/2 * 3500 = 2416.7 1/3 * 2000 = 666.7 1/3 * 2000 = 666.7

(b) Will the flows on links 4 and 5 increase or decrease in case the probit model is used instead of the logit model? Explain your answer. [3] The logit model assumes independent alternatives. If more than two routes are available of which two are largely overlapping (in this case the routes using links 1-2 and links 1-3), all routes are still considered as independent alternatives. This results in an overestimation of traffic on the overlapping

7 - 12

routes. The probit assignment can deal with overlapping route alternatives, since in the probit assignment the error terms are defined on link level (instead of route level). Hence, in case the probit model is used, traffic will shift from the (formerly overestimated) upper routes to the lower route consisting of links 4-5. The flows on these links will therefore increase. (c) Formulate Wardrops first principle for a deterministic user equilibrium assignment. [3] Wardrops first principle for a deterministic user-equilibrium assignment is that the travel time on any used route equals the travel time on any other used route between the same origin and destination and is no greater than the travel time on any unused route. (d) Determine the deterministic user equilibrium assignment in the network above using the given trip matrix. [8] Between OD-pair AB only one route can be chosen, therefore all 500 travelers from A to B will use this route. Between OD-pair BC two routes can be chosen, which have the same travel time function. Therefore, in the deterministic userequilibrium (DUE) assignment, the OD-specific traffic flow on both routes will be equal (1750 travelers). Between OD-pair AC three routes can be chosen. In the DUE assignment the travel time on these routes must be equal. Comparing the travel time on the upper route (link 1 and 2) and lower route (link 4 and 5), we can see that:
q1 q2 q4 q5 1 + + 1 + = 1 + + 1 + 2000 2000 2000 2000 q1 + q2 = q4 + q5

Making the link flows OD-specific and adding the traffic flows from the other OD-pairs, we obtain:

(q

AC 1

+ 500 ) + ( q2AC + 1750 ) = q4AC + q5AC

From the network we can see that q4 = q5 and q1AC = q2AC + q3AC . Thus,
q1AC ( q + 500 ) + 2 + 1750 = 2 q4AC The total travel demand from origin A to destination C is 2000, therefore q4AC = 2000 q1AC and all link flows can now easily be computed, leading to the following traffic volumes:
AC 1

link 1: link 2: link 3: link 4: link 5:

500 + 500 = 1000 vehicles 1750 + 250 = 2000 vehicles 1750 + 250 = 2000 vehicles 1500 vehicles 1500 vehicles

8 - 12

Question 5: [14 points]

Suppose there are 3000 travelers from city A to city B and they can choose between two routes. The first route has a large capacity and is assumed to have a fixed travel time of 8 minutes. The second route is capacity constrained and the travel time t2 on this second route depends on the flow q2 on this second route, and is given by

q2 t2 (q2 ) = 2 + 1 . 2 1000
Furthermore, suppose that the value of time is 0.20 per minute. (a) Determine the total system travel time in case of a deterministic user equilibrium (DUE). [4] In case of a DUE assignment, the travel time on any used route is equal and no greater than the travel time on any unused route. The free flow travel time on the second route is smaller than on the first route (respectively 2 minutes and 8 minutes). Thus, travelers will start using the second route. Due to the increase in traffic flow, the travel time will increase as well. Once the travel time on the second route reaches 8 minutes, additional traffic will choose the first route. Hence, the travel time on the first route can never exceed 8 minutes (i.e. 2 + (q2/1000)2 8) and therefore the traffic flow on the first route can never exceed 3464 vehicles. Since the travel demand is only 3000, all travelers will choose the first route. The travel time on the first route will then be 2 + (3000/1000)2 = 6.5 minutes. The total system travel time equals 6.5 * 3000 = 19500 minutes. (b) Determine the system optimal assignment and show that the total system travel time is lower than in the case of a DUE. [6] In the system optimal (SO) assignment the total system travel time is minimized. The SO assignment can be solved with an equivalent approach to the one used for the DUE assignment in (a) if we include the marginal link travel time (this defines the total additional system travel time due to an extra vehicle. The new link travel time function then becomes:
3 q (q2 )q2 + t2 (q2 ) = 2 + 2 t (q2 ) = t2 2 1000 Following the same line of reasoning as in (a) we can see that
2 2

3 q 2 + 2 8 q2 = 2000 vehicles 2 1000 and the total system travel time is 2 1 2000 + 2000 + 8 ( 3000 2000 ) = 16000 min 2 2 1000 which is indeed lower than in the case of the DUE assignment.

9 - 12

(c) How much toll should be charged on the second route (assuming a DUE) in order to achieve a system optimum? [4] To achieve the system optimum the traffic flow on the first route should equal 2000, resulting in a travel time of 4 minutes. To ensure the SO solution in a DUE assignment, the generalized costs on both routes need to be equal. Therefore the toll level on the second route is the difference in travel time (8 4 = 4 minutes) times the value of time:
toll = 4 0.20 = 0.80 euro

The same answer is obtained when multiplying the marginal costs of route 2 q2 (q2 )q2 = ( t2 q2 ) and the value of time. 10002

Question 6: [10 points]

For each statement below, indicate if the statement is true or false. (a) For land use modeling, knowledge of the functional land use is more important than knowledge of the formal land use. [1] True (b) According to the single commodity model of Von Thnen, the amount of land used for the production of the commodity will increase as the transportation costs per unit of the commodity will decrease. [1] True (c) In the classical theory of Christaller and Lsch the size of the distribution area is determined by the trade-off between the benefits of producing in larger quantities and extra distribution costs. [1] True (d) The classical model of Hansen aims to explain the employment levels in each zone. [1] False (e) The LMS model uses a gravity type model to compute the trip distribution. [1] False (f) The Dutch integrated land use and transportation interaction model TIGRIS XL forecasts location choices of households and firms (jobs) for different policy scenarios. [1]

10 - 12

True (g) In regulated development of land (directed allocation) the spatial developments follow preferences of households and firms and are restricted by the availability of land. [1] False (h) The accessibility of the location of a firm is an important factor in the probability to move to a different location, in other words, accessibility is a push factor. [1] False (i) Accessibility is an important factor in the location choice probability of a firm, in other words, accessibility is a pull factor. [1] True (j) Firms have a strong preference to move to nearby locations, in other words, accessibility is a keep factor. [1] True

Question 7: [10 points]

Consider 4 four cities, A through D, in which residents live and firms offer jobs. The distances [in km] between the cities are indicated in the figure below. A 3 3 4 B 4

3 C

The number of residents living in each city, and the number of jobs being offered by firms in each city are listed in the next table. Assume that all intracity distances are 2 km.
City A B C D Residents 4000 1000 2000 3000 Jobs 2000 500 1000 4000

11 - 12

The accessibility can be computed with the following potential accessibility measure,

i = Pj / cij
j

[origin-based definition]

or j = Pi / cij
i

[destination-based definition]

where i is the accessibility of city i, Pi is the potential of city i, and cij is the distance from city i to city j. (a) From the perspective of residents (who would like to be close to jobs), which city is most accessible? [5] From the perspective of residents the potential of a city is given by the number of jobs. Given the number of jobs (table) and the distances between cities (figure) the accessibility of each city can be computed.
A 2000 1/ 2 B 500 1/ 3 city = C 1000 1/ 3 D 4000 1/ 7

1/ 3 1/ 2 1/ 3 1/ 4

1/ 3 1/ 3 1/ 2 1/ 4

1/ 7 2071.4 1/ 4 = 2250.0 1/ 4 2333.3 1/ 2 2660.7

City D then proves to be the most accessible. (b) From the perspective of a shop (that likes to be easily reachable by residents), which city is most accessible? [5] Similar to (a), the accessibility of each city can be computed from the perspective of a shop (thus considering the potential as to the number of residents). A 4000 1/ 2 1/ 3 1/ 3 1/ 7 3428.6 B 1000 1/ 3 1/ 2 1/ 3 1/ 4 3250.0 = city = C 2000 1/ 3 1/ 3 1/ 2 1/ 4 3416.7 D 3000 1/ 7 1/ 4 1/ 4 1/ 2 2821.4 City A then proves to be the most accessible.

12 - 12