Professional Documents
Culture Documents
Laboratorio
http://www.webs1.uidaho.edu/niatt_labmanual/index.htm
(ya no est disponible en lnea)
Introduction
Travel Demand Forecasting is a key component of the transportation engineers technical
repertoire. It allows the engineer to predict the volume of traffic that will use a given
transportation element in the future, whether that element is an existing highway or a potential
light-rail route.
Like many other predictive sciences, Travel Demand Forecasting is continually evolving.
Special refinements based on experience and research are proposed each year, but the general
ideology behind Travel Demand Forecasting has remained relatively untouched.
The travel demand forecasting process can be confusing. This chapter is designed to introduce
the fundamentals of Travel Demand Forecasting to undergraduate engineering students by
dividing the process into manageable steps.
To begin exploring Travel Demand Forecasting, click on one of the links to the left.
Lab exercises
These exercises will help increase your understanding of Travel Demand Forecasting, by
presenting more complicated problems that require more thorough analysis.
Exercise 1: The Gravity Model
Exercise 2: Cross-Classification
Travel Demand Forecasting can seem like a long and daunting process when viewed as a
whole. It is much easier to approach when broken into small steps. The discussions below
should help you develop a basic understanding of the Travel Demand Forecasting process.
Overview of the TDF Process
Description of the Study Area
Trip Generation Analysis
Cross-Classification
Multiple Regression Analysis
Experience Based Analysis
Trip Distribution Analysis
The Logit Model
The Gravity Model
Modal Choice Analysis
Trip Assignment Analysis
Results
Cross-Classification
The three major techniques used for Trip Generation Analysis are Cross-Classification,
Multiple Regression Analysis, and Experience Based Analysis. Cross-Classification
procedures measure the changes in one variable (trips) when other variables (land use etc.) are
accounted for. Cross-Classification resembles multiple regression techniques. CrossClassification is essentially non-parametric, since no account is taken of the distribution of the
individual values. One problem with the Cross-Classification technique is that the
"independent" variables may not be truly independent, and the resultant relationships and
predictions may well be invalid.
The FHWA Trip Production Model uses Cross-Classification and has the following submodels.
a. Income sub-model: reflects the distribution of households within various income
categories (e.g. high, medium and low).
b. Auto ownership sub-model: relates the household income to auto ownership.
c. Trip production sub-model: establishes the relationship between the trips made by
each household and the independent variables.
d. Trip purpose sub-model: relates the trip purposes to income in such a manner that the
trip productions can be divided among various purposes. These models are developed
using origin-destination travel surveys.
A considerable amount of research and development has focused on the area of disaggregate
models for improved travel demand forecasting. The difference between the aggregate and
disaggregate techniques is mainly in the data efficiency. Aggregate models are usually based
upon home interview origin and destination data that has been aggregated into zones; then the
"average" zonal productions and attractions are derived. The disaggregate approach is based
Household
size
Household
size
Household
size
Household size
on large samples of household types and travel behaviors and uses data directly. There are
savings in the amount of data required and some of the data can be transferred to other
applications. The disaggregate approach expresses non-linear relationships and is more easily
understood. The tables shown below show several steps of a cross-classification analysis.
Auto ownership
0
1
HH
Trips
HH
Trips
1,200 2,520 2,560 6,144
874 2,098 3,456 9,676
421 1,137 2,589 8,026
1
2
3+
Number of trips
per household size
2+
HH
Trips by auto ownership,
54
130 obtained from
5,921 20,165 regional suty
8,642 33,704
1
2
3+
Auto ownership
0
1
2.1
2.4
2.4
2.8
2.7
3.1
1
2
3+
Auto ownership
0
1
25
125
32
175
10
89
1
2
3+
Auto ownership
0
1
52
300
77
490
27
276
Where:
Ti = Total number of trips produced in zone I
Aj = Total number of trips attracted in zone j
P = Total Population for zone I
DU = Total number of dwelling units for zone I
A = Total number of automobiles in zone I
E = Total employment in zone j
Multiple Regression Analysis is relatively simple to understand. First, data regarding the
actual number of productions and attractions is coupled with data about the area that is
thought to impact the production and attraction of trips. For instance, the total population is
believed to impact the number of trips produced. If we know the number of trips produced
and the population for the present and a few time periods in the past, it is possible to develop
a relationship between these parameters using statistical regression. Once we are satisfied
with the relationship that has been developed, we can extrapolate into the future by plugging
the future population into our relationship and solving for the number of productions. The
process is called Multiple Regression, because there are normally several variables that
impact trip production and attraction.
where
Pij= probability of trips from zone i choosing destination j
Vij= Aj- tij+ t2ij where and are parameters to be estimated
Aj=trip attractions estimated for zone j
tij= highway travel time to zone j from zone i
Z= total number of zones
Multiplying the probability of traveling from zone i to zone j by the number of trips produced
by zone i will yield the number of trips produced by zone i that will travel to zone j.
Where:
Tij = trips produced at I and attracted at j
Pi = total trip production at I
Aj = total trip attraction at j
F ij = a calibration term for interchange ij, (friction factor) or travel time factor ( F ij =C/tijn )
C= calibration factor for the friction factor
Kij = a socioeconomic adjustment factor for interchange ij
i = origin zone
n = number of zones
Before the gravity model can be used for prediction of future travel demand, it must be
calibrated. Calibration is accomplished by adjusting the various factors within the gravity
model until the model can duplicate a known base years trip distribution. For example, if you
knew the trip distribution for the current year, you would adjust the gravity model so that it
resulted in the same trip distribution as was measured for the current year.
Where:
Pit = probability of individual t choosing mode i
Uit = utility of mode i to individual t
Ujt = utility of mode j to individual t
For example:
Uauto = 1.0 -0.1 (TTauto) - 0.05(TCauto)
4. The travel time for all links are calculated and revised if changes result.
5. The process of incremental assignments, followed by calculation of revised shortest
travel times, by link, continues until all trips have been assigned.
The capacity restraint model used by FHWA is applied in an iterative manner. The adjusted
link speed and/or its associated travel impedance is computed using the following capacity
restraint function:
T=To[1+0.15(V/C)4]
Where:
T= balance travel time (at which traffic V can travel on a highway segment)
To= free flow travel time: observed travel time (at practical capacity) times 0.87
V= assigned volume
C = practical capacity
Results
Once you have completed the trip assignment analysis, you have a picture of the volume of
traffic that each element of your transportation system can expect to service in the future. This
gives you insight into the ramifications of changing the transportation system. For example,
widening a highway will increase capacity and shift more traffic onto that highway in the
future. Using travel demand forecasting, you can explore the impacts of alternatives before
their construction.
Professional practice
The Transportation Planning Handbook serves as the primary source for information
concerning Travel Demand Forecasting. Published by the Institute of Transportation
Engineers, this manual serves as a general reference for professional engineers. It has been
developed extensively to encompass all aspects of planning and provides essential knowledge
for the transportation engineer.
We referenced the 1992 publication of the Transportation Planning Handbook because it is
generally accepted as "the" authority on planning in the professional realm of engineering.
For more a extensive analysis the Institute of Transportation Engineers also publishes a
Travel Demand Forecasting manual. This manual provides in-depth analyses for various
aspects of travel demand and impact studies and is also frequently referenced in the
professional arena.
Zones and Zoning
Networks and Nodes
Trip Generation Analysis
Trip Distribution
Modal Choice
Trip Assignment
Model Calibration and Validation
Trip Distribution
The following excerpt was taken from the Transportation Planning Handbook published in
1992 by the Institute of Transportation Engineers (pp. 112-114).
Trip distribution models connect the trip origins and destination estimated by the trip
generation models to create estimated trips. Different trip distribution models are developed
for each of the trip purposes for which trip generation has been estimated. The trip
distribution models found most often in practice today are "gravity models," so named
because of their basis in Newtons law. . . .
The measure of separation between zones most commonly used for trip distribution is
roadway travel time, calculated from the computerized transportation networks. Most
transportation planning efforts use peak-period travel times as a measure of zonal separation
for home-based work and home-based school models. . . . Recent studies have tried to
incorporate travel cost and transit travel time into the separation measure. Cost has been
considered in an attempt to estimate effects on trip distribution of parking costs, vehicle
operating costs, and tolls.
Logit Model
Other trip distribution models that have been used include "opportunity" models and logit
models, both of which estimate the probability that travelers will accept various destination
options available. The logit formulation has recently been used for the Portland, Oregon
metropolitan area. As shown in Figure 4.20, the probability of selecting a particular
destination zone is based on the number of trip attractions estimated for that destination zone
relative to the total attractions in all possible destination zones. The probability is applied to
trip productions estimated for the origin zone, making it conceptually similar to the gravity
model.
Gravity model
Those models generally estimate the distribution of trips to be proportional to the number of
trip ends estimated by the trip generation models and inversely proportional to a measure of
separation between the origin and destination zones. The gravity model has achieved virtually
universal use because of its simplicity, its accuracy and due to its support from the U.S.
Department of Transportation. . . .
Developing a gravity model is a trial-and-error process that requires considerable care. This
process, often called calibration, identifies the appropriate decay function or "friction factor",
that represents the reluctance or impedance of persons to make trips of various durations or
distances. . . . The adjustments are made incrementally with successive iterations of the model
until the trip length frequency distribution produced by the model closely matches the
frequency distribution from the travel survey or demonstrates an acceptable shape and
average trip length.
An important consideration in developing trip distribution models is "balancing" productions
and attractions. One aspect of balance is to assure that the total productions equal the total
attractions in the study area for each trip purpose. Deciding whether the productions or
attractions should be the control total depends on whether there is greater confidence in the
production (usually population) growth estimate or the attraction (usually employment)
growth estimate. It is not unreasonable to average the two (production and attraction) trip
estimates. The productions and/or attractions for all zones must then be factored so that their
sum matches the control total. . . .
(p. 114) At each iteration of the gravity model, the total trips attracted to each zone is adjusted
so that the next iteration of the gravity model will send more or fewer trips to that attraction
zone, depending on whether the immediately previous total trips attracted to that zone was
lower or higher, respectively, than the trip attractions estimated by the trip generation model. .
. . Any unacceptable difference between the generation and distribution model estimates after
five iterations of the gravity model usually indicates an inconsistency in the assumptions or
functions of the trip distribution model and the growth allocation model.
One other consideration in developing a trip distribution model is how to handle unexplained
and unacceptable differences between observed and estimated travel patterns. Rather than
conduct extensive research to try to find an explanation for all such phenomena, the accepted
practical approach is to factor the model estimates to match observed patterns. . . . With the
gravity model, and often with other models in this situation, the adjustment factors are called
"K" factors. The "K" factors are developed for individual trip interchanges and are assigned
values that adjust the estimated trips for the interchanges of concern to match the observed
values.
Modal Choice
The following excerpt was taken from the Transportation Planning Handbook published in
1992 by the Institute of Transportation Engineers (pp. 114-115).
Mode choice models are usually the most complex of the sequential model structure.
Typically these models estimate how many persons will ride public transit and how many will
use private vehicles. Further sophistication of these models may include identifying submode
choice among different transit services and estimating the number of car pools or van pools of
various sizes for high occupancy vehicle facilities. . . .
Logit Model
Mode choice models are found in numerous formulations, but the most common are based on
the probabilities estimated by some variation or sophistication of the logit function. . . . The
common logit mode choice relationship states that the probability of choosing a particular
mode for a given trip is based on the relative values of the costs and levels of service on the
competing modes for the trip interchange being considered.
The level of service provided by a particular mode for a specific trip interchange is usually
represented in part by the travel time for that interchange as computed from the transit and
roadway networks. The travel time components used to represent level of service include the
in-vehicle travel time for each mode and the out-of-vehicle time required to use that mode,
such as walking to a transit stop or from a parking lot. The level of service also includes the
waiting time likely to be experienced, either to board transit or to transfer. The delay due to
roadway traffic congestion is included inherently by using attenuated speeds for congested
roadway network links. . . .
The travel time and cost of a trip are usually combined using an estimate of the cost of time to
convert either cost or time to the terms of the other. The cost of time is usually a variable,
based on the economic level of the traveler. Although the mode choice model may be
developed using the economic level of individual travelers, forecasts of mode choice are
prepared for different economic groups, such as high, medium, and low income travelers. The
resulting combination of time and cost is commonly referred to as the "utility" or "generalized
cost.". . . .
The logit formulation is not a complex mathematical function nor is the utility function it
employs. The difficulty in developing a logit model is encountered in estimating the
considerable number of parameters for variables in the utility function. The estimation is
accomplished using one or another multivariate statistical analysis program to optimize the
accuracy of estimates of coefficients of several independent variables.
Trip Assignment
The following excerpt was taken from the Transportation Planning Handbook published in
1992 by the Institute of Transportation Engineers (pp. 115-117).
The traffic assignment process is somewhat different from the mathematical models used for
trip distribution and mode choice. Traffic is assigned to available transit or roadway routes
using a mathematical algorithm which determines the amount of traffic to allocate to each
route. The traffic allocation is usually based on the relative time to travel along each available
path, computed from the transit and roadway networks.
All or nothing
Historically all trips between two zones were assigned to the route having the minimum travel
time, regardless of the available capacity; this is termed an "all-or-nothing" assignment. Such
an approach is still used for identifying travel desire corridors as an initial step in locating
new and improved transportation facilities. For most transit assignments the all-or-nothing
approach is still used since there are rarely closely competing transit routes in an efficiently
designed transit system. Similarly the all-or-nothing approach is used for assigning high
occupancy vehicle trip assignments.
Capacity Restraint
More common today for roadway assignments is the "capacity-restrained" assignment, a
strategy which assigns traffic in steps. One option in this approach is "proportional"
assignment, which allocates a portion of the trips between every origin-destination zone pair
to the network at each step. An alternative is the "incremental" assignment, which allocates all
of the trips between a subset of zone pairs at each step. In either case the travel times between
all zone pairs are recalculated after each assignment step, considering the traffic already
assigned, to adjust the speeds on all network links. The revised speeds on all links are
determined by a speed-volume function that indicates the maximum speed likely for a
particular volume/capacity ratio. . . . Another assignment step is then computed considering
the revised travel times, after which the link speeds are again adjusted as previously. This
process is iterated until all trips have been assigned. Additional fully iterated assignments may
be necessary to reach an equilibrium in which there is little change in speeds throughout the
network at each assignment step.
that seeks the parameter values which have the greatest probability or maximum likelihood of
being accurate within acceptable tolerance of error.
Such an effort is commonly accomplished with specialized statistical computer programs
designed for just such purposes. . . . Model calibration can also be accomplished by using
values of constants and parameters from models estimated for another location that is similar
to the area being studied; this strategy is referred to as "importing" model parameters and
should be employed only by experienced practitioners.
Once satisfactory estimates of the parameters for all models have been obtained, the models
must be checked to assure that they adequately perform the functions for which they are
intended, that is, to accurately estimate traffic volumes on transit and roadways. Verifying a
calibrated model in this manner is commonly called "validation." The validation process
establishes the credibility of the model by demonstrating its ability to replicate actual traffic
patterns.
Validating the models requires comparing traffic estimated by the model to observed traffic
on the roadway and transit systems. Initial comparisons are for trip interchanges between
quadrants, sectors, or other large areas of interest. . . . The next step is to compare traffic
estimated by the models to traffic counts, including transit ridership, crossing contrived
barriers in the study area. These are commonly called screenlines, cutlines, and cordon lines
and may be imaginary or actual physical barriers. Cordon lines surround particular areas such
as the central business district or other major activity centers. . . . Transit ridership estimates
are commonly validated by comparing them to actual patronage crossing cordon lines around
the central business district. . . .
The importance of traffic and transit counts for model validation underscores the need for
careful planning, thoroughness and accuracy of a traffic and transit data collection program
that has this purpose. As with the travel surveys, the resulting models and forecasts will be no
better than the data used for model estimation and validation.
Example problems
It doesn't seem to matter how many times we read about a concept, most of us won't
remember it or fully understand it until we have worked with it. To encourage this extra level
of comprehension, we have provided an example problem for each of the applicable concepts.
The more concerned you are about your understanding of a topic, the more seriously you will
want to approach the example problem for that topic.
Cross-Classification
Gravity Model
Logit Model
Traffic Assignment
Cross Classification
The following cross-classification data have been developed for Beaver Dam Transportation
Study Area.
($000)
Income
HH (%)
Autos/HH (%)
High
Med
Low
10
30
70
48
48
20
50
50
72
24
30
10
70
20
53
40
40
20
75
32
52
15
50
50
50
19
56
25
60
70
30
10
60
30
($000)
Trip Rate/Auto
Trips (%)
Income
10
0
2.0
1
6.0
2
11.5
3+
17.0
HBW
38
HBO
34
NHB.
28
20
2.5
7.5
12.5
17.5
38
34
28
30
4.0
9.0
14.0
19.0
35
34
31
40
5.5
10.5
15.5
20.5
27
35
38
50
7.5
12.0
17.0
22.0
20
37
43
60
8.0
13.0
18.0
23.0
16
40
44
Develop the family of cross classification curves and determine the number of trips produced
(by purpose) for a traffic zone containing 500 houses with an average household income of
$35,000. (Use high = 55,000; medium = 25,000; low = 15,000)
[Solution Shown Below]
Solution
The solution to this type of problem is best described through the use of graphs and tables.
The graphs and tables used for this problem are shown below.
Income
Households (%)
HH/Zone
Total HH
13
500
65
Medium $20,000-45,000
72
500
360
High $45,000-$60,000
15
500
75
100
500
Income
Low
Medium
High
26
60
63
15
14
32
58
3+
27
100
100
100
Income
Low
Medium
High
13
12
13
18
3+
17
18
23
Income
Low
Medium
High
17
11
39
227
11
115
44
3+
20
65
360
75
Income
Low
Medium
High
34
32
273
1814
146
109
1498
783
3+
130
466
416
3474
1395
5285
Medium
High
HBW
38
37
18
HBO
34
34
38
NHB
28
29
44
100
100
100
Medium
High
HBW
158
1285
251
1695
HBO
141
1181
530
1853
NHB
116
1007
614
1738
Gravity Model
A study area consists of three zones. The data have been determined as shown in the
following tables. Assume a Kij =1.
Zone Productions and Attractions
Zone
Total
Trip Productions
140
330
280
750
Trip Attractions
300
270
180
750
82
52
50
41
39
26
20
12
Determine the number of trips between each zone using the gravity model formula and the
data given above. Note that while the Friction Factors are given in this problem, they will
normally need to be derived by the calibration process described in the Theory and Concepts
section.
[Solution Shown Below]
Solution
First, determine the friction factor for each origin-destination pair by using the travel times
and friction factors given in the problem statement.
Fij as Determined from Travel Time
Zone
39
52
50
52
26
26
50
26
39
Once you have the friction factors for each potential trip, you can begin solving the gravity
model equation as shown below. Solving for the A*F*K term in a tabular form makes this
process easier. Study the equation below and the following table.
Where:
Tij = number of trips that are produced in zone I and attracted to zone j
Pi = total number of trips produced in zone I
Aj = number of trips attracted to zone j
Fij = a value which is an inverse function of travel time
Kij = socio economic adjustment factor for interchange ij
1
11700
14040
9000
34740
15600
7020
4680
27300
15000
7020
7020
29040
AjFijKij
sum
Once the A*F*K terms for each origin-destination are tabulated, you can insert these values
into the gravity model equation and determine the number of trips for each origin-destination.
The following table illustrates this.
Zone to Zone First Iteration:
zone
47
57
36
140
189
85
57
330
145
68
68
280
380
209
161
750
given A
300
270
180
750
Since the total trip attractions for each zone dont match the attractions that were given in the
problem statement, we need to adjust the attraction factors. Calculate the adjusted attraction
factors according to the following formula:
Ajk =
Where:
Ajk = adjusted attraction factor for attraction zone (column) j iteration k.
Ajk = Aj when k=1
Cjk = actual attraction (column) total for zone j, iteration k
Aj= desired attraction total to attraction zone (column) j
j= attraction zone number
n= number of zones
k = iteration number
To produce a mathematically correct result, repeat the trip distribution computation using the
modified attraction values.
For example, for zone 1:
Zone
Aj1
380
209
161
Given A
300
270
180
Aj2
237
349
201
AjFijKij
1
sum
9237
18138
10062
37437
12316
9069
5232
26617
11842
9069
7848
28759
35
68
38
140
153
112
65
330
115
88
76
280
303
269
179
750
given A
300
270
180
750
Upon finishing the second iteration, the calculated attractions are within 5% of the given
attractions. This is an acceptable result and the final summary of the trip distribution is shown
below.
The resulting trip table is:
zone
35
68
38
153
112
65
115
88
76
Logit Model
Given the utility expression:
UK= AK - 0.05 Ta - 0.04Tw - 0.02 Tr - 0.01 C
Where:
Ta is the access time
TW is the waiting time
Tr is the riding time
C is the out of pocket cost
a) Apply the logit model to calculate the division of usage between the automobile mode (AK
= -0.005) and a mass transit mode (AK = -0.05). Use the data given in the table below for your
analysis.
Mode
Ta
TW
Tr
Auto
30
100
Transit
10
10
45
50
b) Estimate the patronage shift that would result from doubling the bus out-of-pocket cost.
[Solution Shown Below]
Solution
Part A is solved by substituting the given values into the utility function and solving the
logit model equation. The calculations and results for part A are shown in the table below.
Part B is essentially identical to part A except for the change in the out-of-pocket cost for
bus travel. The preliminary calculations for part B are shown in the table below as well,
while the final calculations are located below the table.
Part A
Mode Ta
Auto
Transit
TW
Tr
Ak
Uk
e^Uk
30
100
10
10
45
50
Part B
Mode Ta
Auto
TW
5
Tr
0
C
30
Ak
100
Uk
e^Uk
0.0050
Transit
10
10
45
100
Traffic Assignment
Assign the vehicle trips shown in the following O-D trip table to the network, using the all-ornothing assignment technique. To summarize your results, list all of the links in the network
and their corresponding traffic volume after loading.
Origin-Destination Trip Table:
Trips between Zones
From/to
1
1
-
2
100
3
100
4
200
5
150
400
200
100
500
200
100
100
150
250
150
300
400
200
100
50
350
Highway Network:
The all-or-nothing technique simply assumes that all of the traffic between a particular origin
and destination will take the shortest path (with respect to time). For example, all of the 200
vehicles that travel between nodes 1 and 4 will travel via nodes 1-5-4. The tables shown
below indicate the routes that were selected for loading as well as the total traffic volume for
each link in the system after all of the links were loaded.
Nodes
Link
Travel
Path
Time
Volume
From
To
1-2
100
1-2,2-3
11
100
1-5,5-4
11
200
1-5
150
2-1
400
2-3
200
2-4
100
2-4,4-5
11
500
3-2,2-1
11
200
3-2
100
3-4
100
3-4,4-5
13
150
4-5,5-1
11
250
4-2
150
4-3
300
4-5
400
5-1
200
5-4,4-2
11
100
5-4,4-3
13
50
5-4
350
Link
Volume
1-2
200
2-1
600
1-5
350
5-1
450
2-5
5-2
2-3
300
3-2
300
2-4
600
4-2
250
3-4
250
4-3
350
4-5
1300
5-4
700
Related links
There are many sources of information about Travel Demand Forecasting on the Internet. You
might want to further your education by taking advantage of these resources.
Oregon Department of Transportation: Transportation Planning Site
http://www.odot.state.or.us/other.asp
Travel Model Improvement Program (TMIP)
http://www.bts.gov/tmip/
National Transportation Library: Planning and Policy
http://www.bts.gov/ntl/subjects/planning.html