A Marginal-cost Pricing Model for Transportation Networks with Multiple-class Users – An Application to the Toll Ceiling Problem Shou-Ren HU a , Hui Pei HUANG b a Department of Transportation and Communication Management Science, National Cheng Kung University, Tainan City, 70101, Taiwan a E-mail: [email protected]b THI Consultants lnc., Taichung City, 40466, Taiwan b E-mail: [email protected]Abstract: This study proposes a marginal-cost pricing model for the congestion pricing problem under different users’ attributes and their influence on the system’s total cost . The primary objective is to investigate the effects of the current double tolling policy during congested periods of the national freeway system in Taiwan. The proposed model solves the differential pricing problem by the marginal cost pricing theory under different user characteristics with varying external costs and values of time. The results of the numerical analysis indicate that the toll ceiling by law needs to be relaxed so that the ultimate goal of social welfare maximization can be achieved, and the marginal-cost pricing strategies have an effect on reducing traffic congestion by encouraging users with a high price elasticity to switch to alternative routes. The results found in this study should have implications for the government offices in preparing a desirable tolling policy. Keywords: Marginal-cost Pricing, Multiple-class Users, Asymmetric Link Cost, External Cost, Value of Time 1. INTRODUCTION As the increasing economic and social activities, traffic demand and usage of private vehicles have been increasing in the metropolitan areas. The growth and increased usage of private vehicles have resulted in many problems, such as traffic congestion, road crashes, and air and noise pollutions. To resolve these problems, traditional and advanced methods have been proposed, in which congestion pricing and/or highway tolling policy has been shown as one of the most effective policy tools to mitigate traffic congestion problems. In Taiwan, a distance-based Electronic Toll Collection (ETC) system was implemented on the national freeway system since 2013. Averagely four millions of toll transactions in a typical weekday. The distance-based ETC system uses a microwave system for the communication between an on-board RFID tag (called e-Tag, free-of-charge of installation) and a roadside gantry. Toll is charged in two ways. For those e-Tag equipped vehicles, they are charged by a pre-registered deposit account. While for those vehicles without an e-Tag, they will receive a bill via traditional mail services. The tolls for non e-Tag users are calculated by their travel distance captured/recorded by a license plate recognition (LPR) system plus an administrative fee. Totally 319 gantries are installed near an entrance and or an exit for each mainline segment. The average distance between two gantries is 3.13 kilometers. The ETC system does not only provide the tolling service, but also has the capability of conducting differential pricing strategies according to different vehicle types and/or user classes. Besides the tolling function, the ETC system also provides time-dependent vehicular origin-destination (O-D) data, travel Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017 388
16
Embed
A Marginal-cost Pricing Model for Transportation Networks ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
A Marginal-cost Pricing Model for Transportation Networks with
Multiple-class Users – An Application to the Toll Ceiling Problem
Shou-Ren HU a, Hui Pei HUANG b
a Department of Transportation and Communication Management Science, National
Cheng Kung University, Tainan City, 70101, Taiwan a E-mail: [email protected] b THI Consultants lnc., Taichung City, 40466, Taiwan b E-mail: [email protected]
Abstract: This study proposes a marginal-cost pricing model for the congestion pricing
problem under different users’ attributes and their influence on the system’s total cost. The
primary objective is to investigate the effects of the current double tolling policy during
congested periods of the national freeway system in Taiwan. The proposed model solves the
differential pricing problem by the marginal cost pricing theory under different user
characteristics with varying external costs and values of time. The results of the numerical
analysis indicate that the toll ceiling by law needs to be relaxed so that the ultimate goal of
social welfare maximization can be achieved, and the marginal-cost pricing strategies have an
effect on reducing traffic congestion by encouraging users with a high price elasticity to
switch to alternative routes. The results found in this study should have implications for the
government offices in preparing a desirable tolling policy.
Keywords: Marginal-cost Pricing, Multiple-class Users, Asymmetric Link Cost, External Cost,
Value of Time
1. INTRODUCTION
As the increasing economic and social activities, traffic demand and usage of private vehicles
have been increasing in the metropolitan areas. The growth and increased usage of private
vehicles have resulted in many problems, such as traffic congestion, road crashes, and air and
noise pollutions. To resolve these problems, traditional and advanced methods have been
proposed, in which congestion pricing and/or highway tolling policy has been shown as one
of the most effective policy tools to mitigate traffic congestion problems. In Taiwan, a
distance-based Electronic Toll Collection (ETC) system was implemented on the national
freeway system since 2013. Averagely four millions of toll transactions in a typical weekday.
The distance-based ETC system uses a microwave system for the communication between an
on-board RFID tag (called e-Tag, free-of-charge of installation) and a roadside gantry. Toll is
charged in two ways. For those e-Tag equipped vehicles, they are charged by a pre-registered
deposit account. While for those vehicles without an e-Tag, they will receive a bill via
traditional mail services. The tolls for non e-Tag users are calculated by their travel distance
captured/recorded by a license plate recognition (LPR) system plus an administrative fee.
Totally 319 gantries are installed near an entrance and or an exit for each mainline segment.
The average distance between two gantries is 3.13 kilometers. The ETC system does not only
provide the tolling service, but also has the capability of conducting differential pricing
strategies according to different vehicle types and/or user classes. Besides the tolling function,
the ETC system also provides time-dependent vehicular origin-destination (O-D) data, travel
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017
388
time and distance data at the individual vehicle level. The valuable traffic data can be used for
effective freeway control and management.
For the toll levels of the national freeway system in Taiwan, the standard toll structure set
by Ministry of Transportation and Communications (MOTC) is shown in Table 1. Differential
tolling policy have been separately implemented on specific periods. For instance, 150% and
50% of the standard toll are respectively imposed on the vehicles traveling on the peak and
off-peak hours of series holidays. Despite for that, a freeway corridor connecting major
metropolitan or attraction areas is frequently congested during specific time intervals and road
segments. This outcome raises a question if the upper bound of the differential toll rates that is
set two times of the standard toll levels (i.e. the ceiling by law) is not high enough to prevent
the traveling public from using the limited freeway resource. Thereby, the purpose of this
study is to investigate the toll rates for a social optimal situation by charging different user
classes with differential tolls. The ultimate goal of this study is to provide the government
agency and freeway bureau a reference in setting a desirable toll structure.
Table 1. Toll structure of the national freeway system in Taiwan
Unit: NT$/km
Source: Taiwan Area National Freeway Bureau (TANFB), MOTC, 2015.
Road pricing is an effective method to mitigate traffic congestion problem through
changing road users’ route choice behaviors. To deal with the congestion pricing problem, the
policy of road pricing has been implemented in several cities around the world (Palma and
Lindsey, 2011) and the theoretical researches of road pricing has been widely extended and
discussed under different charging schemes since Pigou (1920) first advanced the concept of
road pricing of congestion tax theory. The theory proposed by Pigou is named marginal-cost
pricing principle. The ultimate goal of the marginal-cost pricing theory is to internalize the
external cost in order to achieve the social optimum objective and remove and/or divert the
excessive traffic flow or demand to alternative routes (Pigou, 1920; 3. Knight, 1924; Walters,
1961; Vickrey, 1969). By imposing the extra cost on the specific users, the limited highway
resources are expected to be used in a more efficient manner, and the traffic congestion
problem can be mitigated.
Assumption on homogeneity of highway users, no matter on vehicle types or values of
time (VOTs), does enormously simplify the theoretical development and the empirical study,
but this assumption is unrealistic (Walters, 1961). In order to reflect the reality, it needs to
account for different travel costs and VOTs when solving the congestion pricing problem. The
different user classes (e.g., passenger car, truck, bus) have different characteristics (e.g.
income, trip purpose, vehicle type) and are associated with varying VOTs (Arnott, 1992;
Small and Yan, 2001; Yang and Huang, 2005; Holguín-Veras and Cetin, 2009), resulting in
different levels of externality (Arnott, 1992; Holguín-Veras and Cetin, 2009; Dafermos, 1972;
Smith, 1979; Chen and Bernstein, 2004). The time value will influence users’ route choice
behaviors. Different types of vehicle will induce different external costs due to vehicle size,
acceleration/dissertation characteristics and so on (Chen and Bernstein, 2004), and should be
charged by differential tolls. Although the issue of road pricing with heterogeneous users have
Travel distance (d)
Vehicle type
d km/day 20 km < d 200 km d 200 km
Passenger car 0 1.2 0.9
Heavy vehicle 0 1.5 1.12
Trailer 0 1.8 1.35
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017
389
been investigated by the past researches from the perspectives of different travel costs or
VOTs, relatively few studies examine different external costs and VOTs together to solve the
congestion pricing problem. This topic is worthy of investigation, especially from a modern
highway management’s perspective.
Solving the congestion pricing problem with heterogeneous users is complex in a
general network. Dafermos (1972) suggested that the multiple-class equilibrium model could
simplify to a single-class model with link flow interaction by copying the network with
number of vehicle types to express the interaction of different vehicle types traveling on the
same link. To solve the network equilibrium problem with asymmetric link cost function, the
diagonalization method was adopted; this method is as the common solution algorithm for the
asymmetric (cost) network equilibrium problem (Shiffi, 1985). The congestion pricing
problem with multiple vehicle types or user classes can be solved by this method (Yang and
Huang, 2005).
Accordingly, the primary objective of this study are to evaluate congestion tolls of
different vehicle types by accounting for the different external costs and VOTs associated with
different user classes based on a system optimum criterion. We construct a marginal-cost
congestion pricing model based on the national freeway network in Taiwan to explore a
reasonable ceiling for the current tolling structure implemented in Taiwan.
2. METHODOLGY
The road pricing model is formulated by marginal-cost pricing theory and the solution
algorithm is developed to solve the marginal-cost pricing problem on a general transportation
network with multiple-class users.
2.1 Model Assumptions
Before constructing the congestion pricing models, assumptions are made in the following.
First, the corresponding link travel time cost function is a strictly increasing, convex and
continuously differentiable function with the increasing flow on link, and the corresponding
O-D demand function is nonnegative and strictly decreasing with respect to the generalized
cost for each class m∈M between an O-D pair. Second, the route choice behaviors of road
users follow the user equilibrium principle, which means that no user can improve his/her
travel time by unilaterally changing, route. Under this principle, each road user is perfectly
rational to choose cost-minimizing paths between any O-D pair. Third, road users are
heterogeneous and classified by two kinds of vehicle types, which are cars and heavy vehicles.
The road users of class m have their own cost function and value of time. Forth, the potential
demand of an O-D pair r-s will not alter in a short time period. Fifth, congestion tolls can be
charged on all links. Sixth, the externalities other than traffic congestion are not considered.
Seventh, the roadway capacity is fixed.
2.2 Pricing Model
This study formulates the model of a system optimum condition to discuss the congestion toll
with heterogeneous users whose cost functions are asymmetric and value of time are varied
on a general transportation network. The cost function is formulated as a travel time function
which is not only affected by the flows associated with its own vehicle type (called the main
effect) but also influenced by those of other vehicles (called cross effect).This study uses
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017
390
Bureau of Public Roads (BPR) function as the cost function (Yang and Huang, 2005; Chen and
Bernstein, 2004). For each link , there is a travel time cost function of
for each vehicle type (passenger car and heavy vehicle):
(1a)
(1b)
where is traffic flow of class on link ; is travel time cost function of
; and are the parameter of the BPR function of class ; is capacity of
class on link ; E is the passenger car equivalence of heavy vehicle.
Generally, each vehicle type can drive on every lane and the capacities of different
vehicle types are the same. However, under certain situations (e.g. HOV lane), the capacities
of different vehicle types become varying. Therefore, this study uses different notations to
present the capacities of passenger car and heavy vehicle. In addition, letting the unit of
capacity to be equal to the vehicle type is favorable to the derivative of the requirements of
the optimal solution of the proposed marginal-cost pricing model. Therefore, Eq. (1a) and Eq.
(1b) can be transferred as below (Yang and Huang, 2005):
(1c)
(1d)
where , and . Eq. (1c) is the cost function of passenger
car, which is not only affected by its own vehicle type, but also heavy vehicle flow where the
effect of a heavy vehicle is equal to times of the effect of a passenger car. Eq. (1d) is the
cost function of heavy vehicle, which is not only affected by its own vehicle type, but also
passenger car where the effect of a passenger car is equal to times of the effect of a heavy
vehicle. The degrees of the cross effect depend on the parameters of and
The demand function, which is nonnegative and strictly decreasing with the increasing
generalized cost, is assumed linear and is shown below:
(2a)
where is demand function of the m-th class between an O-D pair r-s; is class m’s
demand for travel between an O-D pair r-s; is class m’s potential demand for travel
between an O-D pair r-s; is the price elastic parameter of class ; is
generalized travel cost of class between an O-D pair r-s at traffic equilibrium. The
inverse demand function is presented as:
(2b)
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017
391
where is inverse of the demand function of the m-th class between an O-D
pair r-s. The inverse demand function can be regard as the willingness-to-pay
of user for taking his/her trip.
In an elastic demand case with asymmetric link cost function, the marginal-cost pricing
problem can be formulated as a Variational Inequality (VI) formulation as below:
(3a)
s.t. (3b)
(3c)
(3d)
(3e)
Eq. (3b) is the O-D demand conservation constraint for all classes of road users between all
O-D pairs. Eq. (3c) is the definitional equation, which represents link flows in terms of path
flows. Eq. (3d) and Eq. (3e) are the non-negativity constraints of the path flow and O-D
demand.
Based on the VI formulation, the marginal-cost pricing problem can be transformed to
mathematical programming problem as below:
(4)
subject to Eqs. (3b)-(3e) where is the link flows which include all vehicle types instead of
vehicle type m, on link a. Eq. (4) is the objective function to maximize the social benefit (SB).
The first term of the right-hand side is the total user benefit (UB) which is assumed dependent
on . The second term is the total travel cost (TC) which is assumed to be dependent on
, the flows on link a. The travel time costs of different users are the respective link
travel time multiplied by their own value of time, to become a monetary unit.
2.3 Uniqueness of Solution
In order to obtain the unique solution of link flows and tolls by using the current algorithm,
the objective function and the feasible region of solution must be convex. The concavity of
the feasible region is ensured by the linear equality constraints (Eq. (3b), Eq. (3d) and Eq. (3e))
(Shiffi, 1985). Different from the proof of the model developed by Yang and Huang (2005),
besides the asymmetric link cost function; this study also considers the heterogeneous value
of time. To guarantee the uniqueness of the solution, it needs to proof the concavity of the
objective function.
This study assumes that the demand function for each class of O-D pair, is
nonnegative and strictly decreasing with the increasing generalized cost for each class of users.
The inverse of the demand function, should be also a strictly decreasing function. The
integral of a decreasing function is strictly concave and sum of the strictly concave functions
is strictly concave. The negative of a strictly concave function will be convex. Sum of two
convex functions is convex. Therefore, we can focus on examining the convexity of the TC
with the asymmetric link cost function, which introduced in Eq. (1c) and Eq. (1d) to guarantee
the concavity of the objective function by insuring the positive definiteness of the Hessian
matrix. Specifically, this study sets:
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017
392
(5)
and
( )( )
(6)
According to Eq. (6), three conditions guarantee the positive definite of the Hessian matrix or
the convexity of the TC (Yang and Huang, 2005):
(i)
(ii)
(iii)
should be small.
The first and second requirements are the same as those discussed in (Yang and Huang,
2005).The third requirement would also be the same as Yang and Huang (2005) if this study
sets the VOT are equal to one. The third requirement means that the difference of asymmetric
effect in monetary unit between passenger car and heavy vehicle should be small to ensure the
negative value of second term of Eq. (6) to be smaller than the sum of other positive term.
The first and second requirements can be guaranteed by setting proper parameters. However,
the third requirement is hard to be guaranteed. We cannot set in advance or know the value of
because traffic flows are variable.
It causes the uncertainty of concavity of the objective function. Therefore, it may incur the
multiple solutions problem.
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017
393
2.4 Optimal Condition
The optimal conditions for the system optimum multiple vehicle types with elastic demand
can be derived by the first-order condition. Let link flow as a function of path flow vector f
defined by Eq. (3c) and the Lagrangian is presented as follows:
(7a)
s.t.
(7b)
(7c)
where are the dual variables (Lagrange multiplier) corresponding to Eq. (3b). The
first-order conditions can be written as:
(8a)
(8b)
(8c)
(8d)
(8e)
(8f)
(8g)
Here, this study calculates the partial derivatives of with respect to the flow
variable (which means flow of class users on path between the O-D pair n-o) and
the demand variable (which means class p’s demand for travel between the O-D pair
n-o). According to the derivation of Eq. (8a), we can know the congestion toll as follows:
(9)
where is the generalized cost (travel time cost and toll) of class on path
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017
394
between an O-D pair n-o. Here, is the external cost that a new
entry of vehicle type p imposes on other vehicle on link b. This term is the congestion toll
( ) that should be charged based on the marginal-cost pricing theory. Therefore, it can
present that:
(10)
The derivations of first-order conditions for this model are summarized as follows:
(11a)
(11b)
(11c)
(11d)
(11e)
(11f)
(11g)
Eqs. (11a) and (11b) are derived from Eqs. (8a) and (8b) and mean that if the cost of path k, is
larger than the minimum-path travel cost for the O-D pair r-s, no user will use this path and
the flow is zero. If the cost of path k is equal to the minimum-path travel cost, the flow on
path k can be zero or positive. Eqs. (11c) and (11d) are derived from Eqs. (8c) and (8d) and
mean that if the inverse of the demand function is smaller than the
minimum-path travel cost for the O-D pair r-s, the demand of O-D pair r-s is zero. If the
demand function is equal to the minimum-path travel cost, the demand of O-D
pair r-s can be zero or positive.
2.5 Frank-Wolfe Algorithm with Diagonalization Method
The diagonalization method is the most common algorithm to address the network assignment
problem with asymmetric link cost functions. This problem is similar as the network
assignment problem with single type of user, which can be solved by the Frank-Wolfe
algorithm. The difference is that the travel time on each link is updated based on the entire
flow pattern under different vehicle types, since rather than .To
deal with this problem, the link cost function Jacobian still needs to be positive definite to
assure a unique solution in the solving process. To satisfy the positive definite of link cost
function, this method only focuses on dealing with the main effect of a vehicle type on each
link (i.e. is a function of only). The cross effect of other vehicle type(s) on the same
link are fixed on every iterative.
This study combines the Frank-Wolfe algorithm and diagonalization method to solve the
problem. The problem will be mainly solved by the structure of the Frank-Wolfe algorithm.
The concept of diagonalization method is used to find the optimal step size to guarantee the
positive definite of link cost function.
In order to solve the problem by these algorithms, the formulations of this problem are
modified as follows:
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017
395
(12a)
s.t.
(12b)
(12c)
(12d)
For computational reasons, is added as an upper bound constraint to the O-D
demand, . In general, is set to be the amount of potential demand. Due to:
(13a)
(13b)
The model can be further rewritten as:
(14a)
s.t.
(14b)
(14c)
In order to minimize the objective function by using the convex combinations algorithm, a
solution of the linear program is required at every iteration:
(15a)
s.t.
(15b)
(15c)
where is the auxiliary flow variable of class users on path
between an O-D pair r-s. A partial derivation of is presented as:
(16)
The program (15) can be expressed as:
(17a)
s.t.
(17b)
(17c)
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017
396
where,
(17d)
(17e)
is the auxiliary flow variable of class on link . is the auxiliary O-D
flow variable of class between an O-D pair r-s. The program (17) can be solved
simply by inspection. To minimize , the demand for travel between an O-D pair r-s,
should be all assigned on the path, , for which and
is negative. If is positive, the
demand for the travel between an O-D pair r-s, should not be assigned on any path. According
to the above description, the minimization rule is presented as follows:
If (18a)
If (18b)
The SO formulation is generally not an equilibrium flow pattern, because the users can be
better off by changing their paths. Therefore, beside the travel cost that travelers perceived,
the external cost, , that travelers impose on the total system should be
charged by a toll. By considering the toll under a network equilibrium status, the model can
be expressed as follows:
(19a)
s.t.
(19b)
(19c)
To minimize the objective function along the descent direction(s), the moving size in the
Frank-Wolfe convex combinations method is determined by finding . The optimal step
size can be found by solving the following program:
(20a)
s.t.
(20b)
In order to ensure the link cost function Jacobian is positive definite, this study used the
diagonalization method. Program (20) only discusses the main effect by fixing the cross effect.
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017
397
When the optimal moving size, , has been determined, and can be
updated by setting:
(21a)
(21b)
Convergence can be tested by the objective function values or the change in link flow and
demand between two successive iterations. Here, this study uses:
(22)
is the predetermined value that is based on the desired degree of accuracy.
According to the above description, the algorithmic steps of the solution algorithm can
be presented as follows:
Step 0: Initialization. Find an initial feasible flow pattern { }, { }.This study
assumes that there are no users in the network as initial feasible flow pattern and set
Step 1: Cost and benefit update. Set by using Eq. (3.1c) and
Eq. (3.1d); compute by using Eq. (3.2).
Step 2: Direction finding. Compute the shortest path, p, between each O-D pair r-s based on
{ }; execute the all-or-nothing assignment procedure according to Eq. (18) and then yield
an auxiliary flow pattern , by Eqs. (13).
Step 3: Moving-size determination. Find by solving program (20). This study uses the
golden section search to solve this program.
Step 4: Updating. Find and by solving Eq. (21).
Step 5: Convergence check. If inequality (22) is met, then terminate. Otherwise, set
and go to step 1.
3. CASE STUDY
In the case study, we aim to investigate the congestion pricing problem based on the National
Freeway No. 5 system in Taiwan. On the National Freeway No. 5 system, Hsuehshan Tunnel
which is one of the longest tunnels in the world (12.9 kilometers in length) and connects
Taipei metropolitan and Yilan county, which is one of the congested roadway segment in
Taiwan because the capacity of tunnel is merely 1,000 vehicles/phpl (Lin and Su, 2009).
Traffic congestion is more severe during weekends because many recreational trips are made
from Taipei to Yilan where many attractive spots are located.
We focus on the traffic congestion problem during peak periods on Hsuehshan Tunnel
where only passenger car and bus are allowed; truck and trailer are prohibited. There are two
lanes on each direction of the whole freeway No. 5 system. Although the parallel Taipei-Yilan
Provincial Highway No. 9 does not charge the toll, it is still included in the network because it
is the only alternative route. The simplified network is presented in Figure 1. Node 1, 2, 3 and
5 are Nangang, Shiding, Pinglin and Toucheng interchange. Node 4 is a dummy node to
distinguish the link of freeway and highway between Pinglin and Toucheng.
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017
398
Figure 1. The simplified national freeway No. 5 network.
The empirical study data were collected based on 2014, which had implemented
differential pricing strategies. The field data of traffic flow collected from TANFB and
Directorate General of Highways is used to estimate demand function. Because the
differentiated tolls are not implemented widely and frequently in the freeway system, the
elastic parameters (and demand functions) are hard to be estimated by the property of vehicle
types, travel distance and so on. On the other hand, the travel demand of bus only accounts for
5.3% of total vehicles. Therefore, this study assumes the elastic parameters of bus are the
same as passenger car.
The related parameters of cost function include free flow travel time, capacity, value of
time and so on. The parameters of a BPR function ( and ) and capacity ( ) are
borrowed from the report of Institute of Transportation, MOTC (Lin et al., 2007). The
capacity unit of a bus is transformed from the passenger car unit (PCU) to the bus unit by
using passenger car equivalence (PCE), which is equal to two (Lin et al., 2007). The real
value of free flow travel time ( ) was obtained from TANFB. In addition, VOTs of
passenger car and bus are respectively set to be NT$ 8 and NT$ 20 per minute according to
the data of TANFB and Department of Statistics of MOTC.
3.1 Experimental Design and Results
In the case study, we design two scenarios to investigate the congestion tolls and compare the
resulted link/path flows with the real traffic flow distribution under a distance-based toll
scheme implemented in the field. The scenarios include vehicles with general lane and
vehicles with exclusive lane. The assumption of exclusive lane is because the bus exclusive
lane is implemented on a part of freeway No. 5 system (e.g., Yilan to Toucheng and a part of
Shiding to Pinglin). A possible policy will be implemented comprehensively in the future.
3.1.1 Vehicles without lane restriction (Scenario 1)
In this scenario, this study assumes that passenger car and bus can drive on each lane and they
can change lanes as needed. The parameters of the cross effects are set: and
(Chen and Bernstein, 2004).
According to the result, we found that the solution of the network equilibrium condition
is local optimum due to the non-concavity of the objective function. The reason is that, in this
scenario, the third requirement of the cross effect is too large that makes the objective
function be nonconvex.
Freeway system No. 5 Taipei-Yilan Provincial
Highway No. 9
Hsuehshan Tunnel
1 2 3 5
4
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017
399
In this local optimal solution, according to the results shown in Figure 2 and Figure 3, we
can know that the tolls of passenger car and bus on the freeway segment are below the
standard toll, except for the freeway segment between node 3 (Pinglin) and node 5
(Toucheng). Both freeway and highway segments which connect node 3 (Pinglin) and node 5
(Toucheng) need to charge a huge amount of congestion toll which is above NT$ 100. The toll
of passenger car and bus on this freeway segment are thirteen and ten times higher than the
standard toll.
Figure 2. Ratio of passenger car toll between congestion toll and standard toll for scenario 1
Figure 3. Ratio of bus toll between congestion toll and standard toll for scenario 1
3.1.2 Vehicles with exclusive lane (Scenario 2)
In this scenario, this study assumes that the outside lane is a bus exclusive lane and passenger
car can only drive on the inside lane of the freeway segment. Since vehicles of different types
are traveling on specific lanes of the freeway No. 5 system, the cross effect on the freeway
segment is zero. On the highway segment, passenger car and bus are driving on the same
lanes. Therefore, the cross effect is assumed to be: γ=1 and δ=0.25.
Different from Scenario 1, the solution in this scenario is under a global optimum
condition. Every generalized path cost and O-D demand match the optimal conditions that
this study proved. As the demonstration in Figure 4, except for the link of node 2 (Shiding) to
node 3 (Pinglin), the toll levels of the other freeway segments should be increased by at least
twice of the standard toll. The toll of passenger car on this freeway segment of node 3
(Pinglin) and node 5 (Toucheng) are eighteen times higher than the standard toll. For the bus
(Figure 5), link tolls on the freeway segment are all below the standard toll levels due to the
bus exclusive lane policy, which is in favor of the transit mode when traveling on a congested
One to twice times
Below standard toll Twice to five times
Above five times
Cannot compare
1 2 3 5
4
One to twice times
Below standard toll
Twice to five times
Above five times
Cannot compare
1 2 3 5
4
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017
400
tunnel section.
The tolls of passenger car and bus are different in this scenario due to the exclusive lane
control policy. The corresponding toll levels of bus are all equal to or less than that of
passenger car. The reason is that the exclusive lane policy is implemented on the freeway
system in this scenario. It reduces the available road capacity for passenger cars. In addition,
since the cross effect of bus is not existed, the toll of passenger car is increased and the toll of
bus is decreased.
Figure 4. Ratio of passenger car toll between congestion toll and standard toll for scenario 2
Figure 5. Ratio of bus toll between congestion toll and standard toll for scenario 2
3.2 Comparison of the Two Scenarios
In this section, this study compares the link flows of the two scenarios with real traffic flow
data. The data of real link flow and O-D demand is obtained from TANFB. It is the average
peak-hour flow of weekends between August 2014 and September 2014.
According to Figure 6 and Figure 7, the traffic flow of the Freeway No. 5 system
between Pinglin and Toucheng, which is the most congested roadway segment, reduced
significantly. In scenario 1, the traffic flows on the southward and northward directions reduce
about 40% and 30%, respectively. In scenario 2, the traffic flows on the southward and
northward directions reduce respectively about 60% and 55%. A part of the traffic flow
transfers to the alternative road (Taipei-Yilan Provincial Highway No. 9). In addition, a part of
the traffic demand is canceled due to the high congestion toll. This result demonstrates that
increasing congestion toll is an effective means to mitigate the congestion problem of the
Hsuehshan Tunnel freeway segment. Further, the traffic flows between Nangang and
Toucheng on the freeway segment are also smaller than that of the collected field data. The
traffic flows of field data between Nangang and Pinglin on the highway segment all transfers
to the corresponding freeway segment. Therefore, the flows of this highway segment become
zero in scenario 1 and scenario 2. In addition, charging congestion toll with exclusive lane can
One to twice times
Below standard toll Twice to five times
Above five times
Cannot compare
1 2 3 5
4
One to twice times
Below standard toll
Twice to five times
Above five times
Cannot compare
1 2 3 5
4
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017
401
reduce more traffic flow than that without lane restriction.
Figure 7. Link flows of the real data and those of two scenarios of the northern bound
4. CONCLUSIONS AND RECOMMENDATIONS
This study investigates the differential congestion toll problem by the marginal cost pricing
theory under different vehicle types’ consideration. Different from the past research, this study
assumes that different vehicle types have distinct impacts on the others and values of time.
According to the outcome of the research, we can know that, for the freeway segment
between Pinglin and Toucheng, the tolls of both passenger car and bus need to be increased
more than ten times of the standard tolls on every scenario no matter on the southward or the
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017
402
northward direction, except for the toll of bus in scenario 2. Second, the traffic flow between
Pinglin and Toucheng, which is most congested road segment, are reduced significantly by
charging the congestion toll. Therefore, it can conclude that if the traffic congestion problem
on the freeway No. 5 system, the ceiling by law (twice of the standard toll) needs to be
relaxed when the marginal-cost pricing principle is applied in order to achieve system
optimum.
Besides the above research findings and conclusions, there are some future study
directions that are worthy of further investigation. First, this study uses the diagonalization
algorithm and Frank-Wolfe algorithm to solve this problem. However, the optimal solution
cannot always be obtained due to the uncertainty of concavity of the objective function. The
other solution algorithms need to be pursued to overcome this research limitation. In addition,
this study only considers the external cost of time loss. Incorporating other external costs into
a pricing model, such as air and/or noise pollution, pavement damage and traffic accident is
another important issue to make the road pricing policy more complete.
5. REFERENCES
Arnott, R., de Palma, A., and Lindsey, R. (1992). Route choice with heterogeneous
drivers and group-specific congestion costs. Regional Science and Urban Economics,
22, 71–102.
Chen, M., and Bernstein, D.H. (2004). Solving the toll design problem with multiple
user groups. Transportation Research Part B, 38, 61–79.
Dafermos, S.C. (1972). The traffic assignment problem for multiclass-user transportation
networks, Transportation Science, 6, 73–78.
Holguín-Veras, J., and Cetin, M. (2009). Optimal tolls for multi-class traffic: Analytical
formulations and policy implications. Transportation Research Part A, 43, 445-467.
Knight, F. (1924). Some fallacies in the interpretation of social cost. The Quarterly
Journal of Economics, 38(4), 582–606.
Lin, F.B. and Su, C.W. (2009). Traffic flow characteristics in and near the Shea-San
tunnel on national highway 5.Transportation Planning Journal, 38, 85–120. Lin, K.S. et al. (2007). National Sustainable Transportation Development: The Demand Model of
the Intercity Transportation System. Research report, MOTC-IOT-95-PDB006, Institute of
Transportation, MOTC, Taiwan. de Palma, A., and Lindsey, R. (2011). Traffic congestion pricing methodologies and
technologies. Transportation Research Part C, 19, 1377–1399.
Pigou, A. C. (1920). The economics of welfare. London, UK: Macmillan.
Shiffi, Y. (1985). Urban transportation networks: equilibrium analysis with mathematical
programming methods. New Jersey, N.J.: Prentice-Hall, Inc.
Small, K.A., and Yan, J. (2001). The value of ‘‘value pricing’’ of roads: second-best
pricing and product differentiation. Journal of Urban Economics, 49, 310–336.
Smith, M.J. (1979). The marginal cost taxation of a transportation network.
Transportation Research Part B, 13, 237–242.
Vickrey, W. S. (1969). Congestion theory and transport investment. The American
Economic Review, 59(2), 251–260.
Yang, H., and Huang, H.J. (2005). Mathematical and economic theory of road pricing.
Philadelphia, PHL: Elsevier Science
Walters, A.A. (1961). The theory and measurement of private and social cost of highway
congestion. Econometrica, 29, 676–699.
Journal of the Eastern Asia Society for Transportation Studies, Vol.12, 2017