Download as pdf or txt
Download as pdf or txt
You are on page 1of 24

Journal of Advanced Transportation, Vol. 43, No. 4, pp.

481-504
www.advanced-transport.com

Multi-Objective Highway Alignment Optimization


Using A Genetic Algorithm
Avijit Maji
Manoj K. Jha

The available highway alignment optimization algorithms


use the total cost as the objective function. This is a single
objective optimization process. In this process, travel-time,
vehicle operation, accident, earthwork, land acquisition, and
pavement construction costs are the basic components of the
total cost. This single objective highway alignment
optimization process has limited capability in handling the
cost components separately. Moreover, this process cannot
yield a set of alternative solutions from a single run. This
paper presents a multi-objective approach to overcome these
shortcomings.
Some of the cost components of highway alignments are
conflicting in nature. Minimizing some of them will yield a
straighter alignment; whereas, minimizing others would make
the alignment circuitous. Therefore, the goal of the multi-
objective optimization approach is to handle the trade-off
amongst the highway alignment design objectives and present
a set of near optimal solutions. The highway alignment
objectives, i.e., cost functions, are not continuous in nature.
Hence, a special genetic algorithm based multi-objective
optimization algorithm is suggested.. The proposed
methodology is demonstrated via a case study at the end.

Keywords: Multi-objective optimization, highway alignment


optimization, genetic algorithm, user cost, construction cost

Introduction

Transportation plays an important role in the socio-economic


development of a region. It consists of various modes or means of travel
from one point to another. The modes vary from simple walking to
aircraft. This classification is termed the modal classification. The basic
modal classes are comprised of highways, railways, waterways and
airways. In highways, motorized and non-motorized vehicles use the

Avijit Maji and Manoj K. Jha, Department of Civil Engineering, Morgan State
University, Baltimore, MD, USA
Received: April 2007 Accepted: October 2008
482 A. Maji and M.K. Jha

road network to travel from one point to another. The railways use the
guided path and rail tracks to travel from one point to another. This type
of transportation requires a huge initial investment. One of the oldest
modes of transport is the waterways. Many country reached culmination
of their prosperity for a good waterways transportation system. On the
other hand, an airway is the newest mode in the category. This needs a
huge initial establishment investment and the operation cost is high too.
Other than highways, all other modes operate from some fixed points.
People and freights need to be transported to those specified points to use
the service.
A highway has the capability to provide a door to door service, which
makes it unique of its kind. Moreover, the other modes need a centralized
control system for safe operation but highways do not need such a
centralized control system. This makes the highways flexible compared
to others and no doubt it can be considered as a key element of the
transportation system.
Two important aspects of highway design are the layout of the roads
or alignment and the structural design of the road or pavement design.
Both aspects have their own importance and play a major role in being a
deciding factor for worthiness of highway projects. Alignment design
comes first in the process and based on the soil characterization and
expected traffic of the chosen alignment, pavement is designed.
Optimized alignment design can save highway project cost. Highway
alignment is a combination of horizontal and vertical profile of highways.
Circular sections, tangent sections and the section joining them called
transition sections, are the basic components of the profiles. Alignment
design is generally considered as separate projections of horizontal and
vertical profiles. This design procedure is basically a 2-dimensional (2D)
approach. On the other hand, comparatively difficult 3D approach of the
design considers the horizontal and vertical alignments simultaneously.
The highway alignment design is not only about the proper design of the
geometric elements, but aspects like user and construction costs play a
key role in the alignment design process. The user and construction costs
are mainly comprised of earthwork cost, right-of-way cost, pavement
construction cost, vehicle operating cost, accident cost, travel-time cost,
etc. From financial perspective the best alignment is the one which yields
minimum cost in all respect. Apart from the costs, there are factors like
environmental impact, historical impact, induced demand impact, socio-
economic impact and political impact, which influence the alignment
design. These factors have their own unit for comparison and might not
Multi-Objective Highway… 483

be converted in monetary term with precision. This leads to a situation


where the alignment with minimum cost may not turn out to be the best
option with respect to some of the factors discussed (see, Jha 2003).
One way to solve this type of problem is by giving due weightage to
different factors and converting them in monetary terms for comparison.
This type of problem can be easily solved by a single objective
optimization process. Another way of approaching this type of problem
is by considering each factor as separate objectives in the optimization
process. This technique is commonly known as multi-objective
optimization and analyzing the factors is known as multi-objective
analysis. Unlike single objective optimization, multi-objective
optimization yields a set of solutions in the design process. Therefore,
multi-objective optimization of highway alignment will provide designer
the freedom of choosing a highway alignment alternative based on the
importance of the objectives from the set of solutions. The set of
alternatives in multi-objective optimization are generated from a single
run of the optimization process.
The proposed research aims to develop a multi-objective highway
alignment optimization methodology which can yield a set of design
alternatives from a single run of the design process. The highway
alignment optimization considers minimization of land acquisition cost,
earthwork cost, pavement construction cost, travel-time cost, vehicle
operation cost and accident cost as the objective. Minimizing travel-time
cost, vehicle operation cost and accident cost will have the tendency to
make the alignment straighter. On the other hand, minimizing land
acquisition cost, earthwork cost and pavement construction cost will
generally make the alignment circuitous. In this research the costs having
the tendency of making the alignment straight or circuitous are grouped
together to form separate objectives. We will demonstrate the process to
optimize both objectives simultaneously and find different alternative
alignments which will give the designer an opportunity to choose an
alignment from a set depending on the requirement.

Highway Alignment Optimization

Highway alignment optimization models were evolved and developed


in the last three decades. It is realized that the horizontal alignment
design process is more complex and needs substantial amount of data
compared to simple optimization of vertical alignments (OECD 1973).
484 A. Maji and M.K. Jha

Inclusion of different cost factors makes it more complicated. So far, the


highway alignment design is considered as a single-objective
optimization problem and optimized using calculus of variations,
network optimization, dynamic programming and genetic algorithm. In
all of these methodologies the highway alignment cost components or
objectives are tied together in a single objective.
Calculus of variation is based on pure mathematical modeling. In this
process, two spatial points (start and end) are connected by a curve and
integration of the cost function is minimized (Wan 1995). Therefore, the
cost function should be continuous between the two points of interest,
which is very unlikely in real-world problems. Similar concept was used
in the Optimum Curvature Principle (OCP) for horizontal highway
alignment design (Howard et al. 1968). Real world application of this
model can be found in maritime route through dynamic ice field
(Thomson and Sykes 1988) and horizontal alignment of an expressway
in flat south Florida (Shaw and Howard 1982). The discontinuous nature
of cost functions and the inability of simultaneously optimizing more
than one objective, make the process unsuitable for multi-objective
optimization process.
Network optimization method uses the concept of optimizing network
problems. In this optimization process, the search space is divided into
small cells and the highway alignment is represented as a network of
location and costs. The nodes of the cells represent the location and the
links represent the costs. Researchers used this methodology successfully
in developing horizontal alignment (Turner and Miles 1971, Turner 1978,
Athanassoulis and Calogero 1973). Later a two-stage method was
developed to incorporate the vertical alignment along with the horizontal
alignment (Parker 1977). The results obtained by this methodology
produce a piecewise linear trajectory. This can be defined as a corridor
not as a highway alignment (Jong 1998). Also, this methodology needs
to calculate the cost information for each link which is computationally
intensive in nature and needs considerable amount of storage space.
Moreover, this method cannot use the cost information of more than one
objective and therefore, it will not be the ideal candidate for solving the
multi-objective highway alignment optimization problem.
The complexity of highway alignment optimization can be better
handled by dynamic programming. In this process, the main problem is
divided into a number of sub-problems in such a way that the
contribution to the objective function value from each sub-problem is
independent and additive (Jong 1998). The search space is subdivided
Multi-Objective Highway… 485

into a number of orthogonal spaces and the objective cost function is


optimized for each orthogonal space one after another. The alignments
for the optimized cost function in each orthogonal space are joined to
form the final alignment. Researchers used this methodology to optimize
horizontal as well as 3-dimensional alignments (Trietsch 1987, Hogan
1973, Nicholson et al. 1976). The precision of this method can be
increased by increasing the number of orthogonal spaces. But this
increases the computation burden and does not yield a smooth alignment.
Therefore, this method can be used for a small search area with not too
complicated cost functions and therefore might not be efficient to solve
multi-objective highway alignment problems.
Numerical search and calculus of variation based model, developed
by Chew et al. (1989) can optimize 3-dimensional highway alignments.
This method generates a smooth alignment at the end (Jong 1998). But, it
cannot guarantee the global optimum solution (Jong 1998). Moreover,
the convergence of the model depends on the predefined initial set of
solution, which makes it a semi-automatic process. Difficulties in
including location dependent discontinuous cost information make it
cumbersome for real-world applications. Therefore, this model will not
be an ideal candidate for the multi-objective optimization of highway
alignments.
With the development of optimization algorithms, computers and
software the highway alignment optimization methodologies have also
become sophisticated. Jong (1998) used genetic algorithm (GA) to
optimize the highway alignments. In this approach the search space is
divided into orthogonal sections perpendicular to the line joining the start
and end points. The Points of Intersection (PI) generated randomly on
those orthogonal sections give the basis to form a smooth alignment
joined by tangent, transition and circular sections. Based on the
generated alignment, different cost components of the objective
functions, such as right-of-way cost, travel-time cost, earthwork cost,
structure cost, socio-economic cost, etc. are calculated. These cost
components are added to the total cost of the highway alignment. The
total cost information is used to generate a new set of PIs using genetic
algorithm (GA) operators. Therefore, this is basically a single objective
optimization approach. The process is repeated until no improvement is
noted in the minimum cost over successive number of generations. Later
incorporation of a GIS database to calculate the land acquisition cost and
earthwork cost (Jha 2000; Jha et al. 2006) made the process more
realistic. Though GAs do not guarantee the optimal solutions, extensive
486 A. Maji and M.K. Jha

research by Jong et al. (2000) and Jong and Schonfeld (2003) indicate
that a reasonably good solution can be obtained within 100 generations
of search. But, this research work was limited to optimization of the total
cost of the alignment. Therefore, simultaneous optimization of all cost
components is not guaranteed. This method yields a single optimal
alignment at the end of the optimization process and thus lacks the ability
of delivering multiple set of solutions.

Multi-Objective Optimization

The multi-objective optimization is the search and optimization of


multiple conflicting objectives. The search leads to a set of optimal
solutions known as Pareto-optima solutions (Deb et al. 2005). Each set of
solution represents a trade-off among different objectives. A solution, in
which a value of at least one objective is better than the rest of the
solutions, is known as non-dominated solution. The Pareto-optimal front
is formed by the non-dominated set of solutions. There exist many
algorithms and application case studies of multi-objective optimization
(Deb 2002), but none of them are applied to complex decision-making
problems like highway alignment optimization. Most of the available
algorithms avoid complexity of the real-world problem and transform
any given multi-objective problem to a single-objective problem by
assigning weights to the objectives. Many studies have been conducted
on the adequacy of those conversions (Deb 2002). The theories and
algorithms for classical single-objective optimization can be applied to
the transformed single-objective function, but there is a fundamental
difference between the multi-objective and transformed single-objective
optimization (Deb 2002). This gives a scope for studying and applying
multi-objective optimization on highway alignments.
Some of the highway alignment costs might be conflicting in nature;
minimizing one might increase the other. For example, an alignment
joining the start and end points in straight line may reduce the travel-time,
vehicle operation, and accident costs; but, might increase the earthwork,
pavement construction, and land acquisition costs if the alignment
traverses through undulated surface profile. Moreover, there are other
alignment sensitive issues like environmental impact, historic site, and
parkland area, which need to be simultaneously optimized along with
cost optimization (Jha and Maji 2007). In a single-objective optimization
process these factors are transformed into monetary values with higher
Multi-Objective Highway… 487

weightage to minimize their impact. However, in real world problems the


sensitive issues are often traded with the project cost to some extent. The
single-objective optimization process does not give precise control on the
trade-off and the final solution greatly depends on the assigned
weightage to different objectives. Therefore, just optimizing the total
cost does not give designer the opportunity to explore the possibility of
trading-off different impacting factors. This objective can be achieved by
multi-objective optimization of the highway alignment.

Genetic Algorithm

Multi-objective optimization problems can be optimized either by


classical optimization techniques or by evolutionary algorithms, such as
a Genetic Algorithm (GA). A preference based approach, where a
relative preference vector is used to scalarize the multi-objectives, might
be followed by the classical optimization techniques to solve the multi-
objective problems (Deb 2002). Alike any other classical optimization
technique it will be a point by point evaluation of solutions and will yield
a single solution at the end. Therefore, classical optimization techniques,
such as gradient based methods and other non-conventional methods,
such as simulated annealing are difficult to formulate for multi-objective
problems (Fonseca and Fleming 1995). On the other hand, evolutionary
algorithms, such as GAs generate a set of solutions in each iteration,
which is known as a population. If the multi-objective optimization
problem has more than one solution, then they can be easily tracked from
the set of non-dominated solutions found during the optimal search
process. Hence, a GA has the potential to deliver a set of optimal
solution in its final solution. Moreover, the ability to find more than one
optimal solution in one simulation run makes GAs unique in solving
multi-objective problems (Deb 2002).
The GA was first used by David Schaffer (1984) as a tool for solving
multi-objective problems. This was known as vector-evaluated genetic
algorithm (VEGA) and it is a modified single-objective GA. Later, more
sophisticated algorithms like non-dominated sorting GA (NSGA)
(Srinivas and Deb 1994), niched Pareto-GA (NPGA) (Horn et al. 1994),
strength Pareto evolutionary algorithm (SPEA) (Zitzler and Thiele 1999),
Pareto archived evolution strategy (PAES) (Knowles and Corne 2000)
and Pareto envelope based selection algorithm (PESA) (Corne et al. 2000)
were developed. Apart from these, there are some less popular
488 A. Maji and M.K. Jha

algorithms like distance based GA by Osyczka and Kundu (1995),


modified distance based GA (Osyczka and Kundu 1996), parameter
space investigation (PSI) (Statnikov et al 2006) and macro-evolutionary
multi-objective genetic algorithm (MMGA) (Chen et al. 2006), which
can solve multi-objective problems without generalizing the problem to a
single-objective optimization problem. Some of these methods are
computationally efficient, some have high convergence to the Pareto-
optimal front while others can generate well distributed solutions on the
Pareto-optimal front (Deb et al. 2005). The performance of the
algorithms greatly depends on the type of problem, objective function,
constraint, and number of objectives. Therefore, a GA based solution
procedure to solve multi-objective highway alignment optimization
problem is highly desirable.

Problem Formulation

Constructing a highway alignment incurs land acquisition cost,


earthwork cost, construction cost, travel-time cost, vehicle operation cost,
and accident cost. These costs depend on the alignment and the
alignment depends on the location of Points of Intersection (PIs) between
the start and end points of the alignment (see, Jha et al. 2006). Generally,
the travel-time cost, vehicle operation cost and accident cost, together
considered as user cost, reduces with the minimization of curvature in the
highway alignment. Hence, minimizing the user cost will exhibit the
tendency of making the alignment straighter. On the other hand,
minimizing land acquisition cost, earthwork cost and construction cost
will exhibit the tendency of making the alignment circuitous. Therefore,
the multi-objective cost function based on user and construction costs for
the highway alignment can be formulated as shown in Eq. (1). Eqs. (2)-
(3) formulate the user cost and construction cost information in detail.
The details of individual cost functions are discussed in Maji (2008) and
skipped here for brevity.

f (PI1, PI2,.......,PIn ) = [ fUC(PI1, PI2,.......,PIn ), fCC(PI1, PI2 ,.......,PIn ),]


T
(1)

f UC (PI 1 , PI 2 ,......., PI n ) = f TTC (PI 1 , PI 2 ,......., PI n )


+ f VOC (PI 1 , PI 2 ,......., PI n ) (2)
+ f AC (PI 1 , PI 2 ,......., PI n )
Multi-Objective Highway… 489

f CC (PI 1 , PI 2 ,......., PI n ) = f LAC (PI 1 , PI 2 ,......., PI n )


+ f EWC (PI 1 , PI 2 ,......., PI n ) (3)
+ f PCC (PI 1 , PI 2 ,......., PI n )
where,
PI i = Coordinate of the i th point of intersections
n = Number of PIs in the alignment
f UC (•) = Cost function for user cost
f CC (•) = Cost function for construction cost
f LAC (•)
= Cost function for land acquisition cost

f EWC (•)
= Cost function for earthwork cost

f PCC (•)
= Cost function for pavement construction cost

f TTC (•)
= Cost function for travel-time cost

f VOC (•)
= Cost function for vehicle operation cost

f AC (•) = Cost function for accident cost

For a 3-dimensional (3D) highway alignment, the PIs contain the x, y


and z coordinates in a 3D space. With the variation of the PI coordinate,
the alignment changes and so does the cost associated with it. In general,
the PI coordinates are searched in a pre-specified 3D search space.
Therefore, the values of x, y and z coordinates are limited to the feasible
boundary of the search space and they can be represented as follows:

PI i = ( xi , y i , z i ) (4)

xl ≤ x i ≤ x u
yl ≤ yi ≤ yu (5)
zl ≤ zi ≤ zu

where,
490 A. Maji and M.K. Jha

xl = Lower bound of x coordinate in the search space


xi = x coordinate of i th point of intersections
xu = Upper bound of x coordinate in the search space
yl = Lower bound of y coordinate in the search space
yi = y coordinate of i th point of intersections
yu Upper bound of y coordinate in the search space
=

zl = Lower bound of z coordinate in the search space


zi = z coordinate of i th point of intersections
zu = Upper bound of z coordinate in the search space

The primary goal of highway alignment optimization is to


simultaneously minimize the construction and user costs associated with
the alignment. From users perspective the newly designed alignment
should yield minimum user cost represented in the form of travel-time
cost, vehicle operation cost, and accident cost. On the other hand, the
agency or the government would like to build an alignment with
minimum right-of-way, construction (including pavement), and
earthwork costs. In most of the situations a single alignment solution
might not be able to optimize both of these objectives. At this point the
designer would like to have a set of optimized alignments which
represent different levels of trade-off between the objectives discussed.
Therefore, both objectives need to be minimized simultaneously and
mathematically it is represented as follows:

Min f (PI 1 , PI 2 ,......., PI n ) (6)

Subject to,
xl ≤ x i ≤ x u
yl ≤ yi ≤ yu
zl ≤ zi ≤ zu

The PIs in the objective function are the decision variables of the
problem. They define the highway alignment, which in turn, governs or
influences different costs associated with the alignment. Therefore, the
Multi-Objective Highway… 491

costs are indirect function of the PIs of the highway alignment. This
makes the multi-objective highway alignment optimization a unique
problem in the field of multi-objective optimization.

Solution Procedure

The multi-objective optimization problem discussed in this paper is


solved with the help of a genetic algorithm (GA). As discussed before,
the PIs are the decision variables for the objective function formulated.
The alignment is determined by the PIs, and they in turn affect the cost
functions. But, the objective function, cost function in this case, cannot
be formulated as a function of the PIs. In conventional multi-objective
optimization problems, objective functions are the function of the
decision variables. Therefore, a special multi-objective GA process is
applied to solve this particular problem. The methodology adopted for
this paper is discussed in subsequent paragraphs.
The PI coordinates are encoded in the chromosomes. Therefore, the
alleles of each chromosome are continuous real numbers defined within
the search space of the problem. The construction and user costs are
evaluated for each alignment generated by the coordinate information
stored in the allele of the chromosomes. Based on the fitness function
values, non-dominated and dominated solutions are sorted. Separate
selection pressures of the non-dominated and dominated solutions are
found based on the rank of the total cost. Next generation of
chromosome set is generated by crossover and mutation of the parent set
of solution. Parent sets for crossover are selected: one from non-
dominated solution set and the other from dominated solution set based
on the selection pressure of individual solution in the group. A set of four
varieties of crossover operators is applied in the reproduction process of
the offspring.
The four different crossover operators are simple crossover, two point
crossover, arithmetic crossover, and heuristic crossover. Simple
crossover is the most widely used crossover operator in GAs. This is
analogues to one point cross over where a point in chromosome is
selected randomly and two chromosomes, each from non-dominated and
dominated sets are crossed. This process assumes inheritance of good
genes from the parent pair to the offspring pair (Jong 1998) and helps the
solution set to converge to the Pareto-optimal front. In two point
492 A. Maji and M.K. Jha

crossover, two crossover points in a set of parent chromosome is


generated randomly and the genes in between the points are exchanged
to create a set of offspring. A combination of good information is
expected to obtain through this process (Jong 1998). The concept of
arithmetic crossover lies in the definition of convex set theory where, any
linear combination of convex set points will generate a convex set point
(Nash and Sofer, 1996). It is also known as convex crossover, linear
crossover, and intermediate crossover (Gen and Chang 1997;
Michalewicz 1996). Arithmetic crossover generates one offspring from a
pair of parent chromosomes. Heuristic crossover uses the fitness value to
determine the direction of search (Jong 1998). This operator also
produces one offspring from a set of parent chromosomes and the
resulting offspring may not always be a feasible solution (Jong 1998).
Based on the total cost information of non-dominated and dominated
set together, a parent solution is selected for the mutation. In this case
also, a set of four varieties of mutation operators are applied in the
offspring generation process. They are: the most widely used uniform
mutation, specially developed straight mutation, non-uniform mutation,
and whole non-uniform mutation. The straight mutation helps to
straighten the alignment between two randomly selected PIs (Jong 1998).
The philosophy behind this operator is that the user cost component is
lesser for most straight and direct alignments. Non-uniform mutation
(Michalewicz 1996) or dynamic mutation (Gen and Cheng 1997) is a
special type of mutation where the range of mutation decreases with the
increase in the number of generation. This operator helps to fine tune the
result. The whole non-uniform mutation, also introduced by Michalewicz
is a modified non-uniform mutation (Jong 1998). Here the concept of
non-uniform mutation is applied in a random sequence to all the genes in
a given chromosome. This operator will produce totally different
offspring from its parent. This helps the candidate alignments to jump
from present search area to another search area and maintains diversity in
the solution (Jong 1998).
The process of generating offspring and evaluating them in the light
of user and construction costs is repeated till the solution converges to
the Pareto-optimal front or no new non-dominated solution is found in
consecutive five generations. At the end, the set of Pareto-optimal front
solution is reported. The set of final solution will give the designer the
flexibility of choosing an alignment based on the trade-off information of
user and construction costs.
Multi-Objective Highway… 493

Case Study

The multi-objective optimization approach was applied to an example


with existing highway alignment information. The multi-objective
alignment optimization algorithm discussed in preceding sections was
codes in C computer language and ArcView GIS’s Avenue programming
language, similar to that performed in our previous single-objective
optimization applications (see, Jha et al. 2007; Jha and Schonfeld 2000 &
2004).
The study area for the example study is a section of Montgomery
County, Maryland near the Brookeville town. This is nearly 10 miles
south of I-70 and 3 miles north of MD 108. A previous study on this
project area was done by the Maryland State Highway (MSHA 2001) for
a bypass on existing MD 97. The study area is digitized in a Geographic
Information System (GIS) using Maryland property view 2003 GIS
database and Microstation base maps for Brookeville. The GIS database
provides a wide variation of land property value giving the opportunity
of testing the effect of land acquisition cost. The terrain height in the
study area ranges from 330 ft to 510 ft which affects the earthwork cost
in the multi-objective optimization process. The 2D coordinates of start
and end points of the proposed alignments is set to (1295645,
548735,470) and (1294512, 552574, 407) as a default on the south and
north sections of MD 97 in Brookeville, respectively. Apart from this,
other information like the width of the right-of-way, average design
speed, number of lanes, width per lane, etc. are needed in the highway
alignment optimization process. This particular study is done for a design
speed of 50 mph and 40 ft average right-of-way width. The design speed
information is used in calculating the minimum curve radius of the
alignment. On the other hand, the right-of-way width is used to calculate
the earthwork and pavement construction costs.
The newly designed highway alignment alternatives will consist of 2
lanes with a 9 ft wide shoulder on both sides. A sensitivity analysis of the
same study area shows that alignment optimized for 5 PIs or more
produce a better alignment (Kang et al. 2006), therefore the new
alignment is optimized for 7 PIs. Based on the standard empirical
vehicle operation cost model, travel-time cost model, and accident cost
model, the total user cost is calculated. The land acquisition cost is
estimated from the GIS database of land property value and the
earthwork cost is calculated based on the land-use profile and vertical
alignment profile information. The pavement construction cost depends
494 A. Maji and M.K. Jha

on the horizontal and vertical profiles of the alignment. Therefore, it is


estimated based on the highway alignment profile information. With all
these information the highway alignment is optimized using a multi-
objective GA.

Results

Single objective optimization can be considered as a special case of


multi-objective optimization where the number of objective function is
limited to one. For a better comparison of the results, the given problem
is optimized for user and construction costs separately considering the
problem as a single objective optimization problem. The best alignment
from user and construction costs perspective is shown in Figures 1 and 2,
respectively. Figure 1 confirms that the user cost will be minimized for a
straight alignment. Similarly, Figure 2 confirms that the alignment will
become circuitous with the minimization of the construction cost. The
given multi-objective problem is also solved as a single objective
optimization of the total cost, result of which is shown in Figure 3. The
optimized alignment is similar to that obtained for construction cost
minimization (Figure 2); but, it is comparatively straighter. This is
because of the inclusion of the user cost in the objective function.
The multi-objective highway alignment optimization problem is
formulated as simultaneous minimization of user and construction cost
functions represented in Eq. (6) and later solved based on the solution
procedure discussed in Solution Procedure section. The final solution
converged to a Pareto-optimal front as shown in Figure 4. For this
particular problem, 28 non-dominated solutions were obtained. A set of 5
representative non-dominated construction and user cost objectives
representing the Pareto-optimal front is shown in Table 1. The
alignments corresponding to the non-dominated solution set are shown in
Figure 5. By reviewing the different cost components it is observed that
the accident cost is zero for a straight alignment and is highest for the
alignment with minimum construction cost. On the other hand, the land
acquisition cost is lowest in the alignment with lowest construction cost
and is highest in the alignment with lowest user cost.
Multi-Objective Highway… 495

Figure 1. Best highway alignment form the user cost perspective


496 A. Maji and M.K. Jha

Figure 2. Best highway alignment from the construction cost


perspective
Multi-Objective Highway… 497

Figure 3. Best highway alignment obtained as a single objective


optimization process
498 A. Maji and M.K. Jha

61.0

60.5
User Cost (Million $)

60.0

59.5

59.0

58.5

58.0
0 100 200 300 400 500 600 700
Construction Cost (Billion $)

Figure 4. Pareto-optimal front for the multi-objective highway


alignment optimization
Note: please refer to the note on Table 1.

Table 1. Representative solution set of the Pareto-optimal front


Alter- Construction
User Cost Total Cost
native Cost Remarks
(Million $) (Billion $)
# (Billion $)
1 687.694 58.60589 687.7526 Best user cost
2 108.4089 58.69015 108.4676 2nd best user cost
3 51.38845 59.4993 51.44795 Intermediate result
2nd best
4 21.10651 60.07859 21.16659
construction cost
Best construction
5 13.20284 60.85548 13.2637
cost
Note on Table 1: The very high construction cost in this artificial case
study is the indication of the fact that we used some very high unit costs.
By reducing the order of magnitude of the unit costs one would expect to
obtain more reasonable costs. However, this will not affect the validity of
the developed methodology.
Multi-Objective Highway… 499

Figure 5. Representative highway alignments obtained by multi-


objective optimization

Discussion

The multi-objective highway alignment optimization methodology


presented in this paper provides a set of competing alternative alignments
rather than a single optimized alignment. This set of alternative solutions
500 A. Maji and M.K. Jha

represents a trade-off between construction and user costs. The Pareto-


optimal front obtained in the solution process is not a continuous front
like any other conventional Pareto-optimal front. Since the objective
functions, i.e., construction and users costs, are point-wise discrete in
nature, the Pareto-optimal front will be a set of points of non-dominated
construction and user cost values. Moreover, the construction cost is the
dominating part of the alignment cost; therefore, the overall alignment
cost is minimized with the minimization of the construction cost.
The construction cost depends on the land property value and land
parcels are not uniform in shape and size. Therefore, the intermediate
variation of construction cost in the non-dominated solutions is wide in
nature. This might explain the wide gap in between the best user cost
solution and the second best user cost solution as shown in Figure 4.
When the alignment traverse through low land cost area; the change in
land acquisition cost with alternative alignments is less. Hence,
comparatively higher number of non-dominated solutions is found close
to the best construction cost solution.
By carefully inspecting Figure 5 and the corresponding Table 1, it can
be safely concluded that the straighter alignment reduces user cost but
increases the construction cost. Moreover, the alignments greatly depend
on the land profile and the land property value. Therefore, the
alternatives obtained in the multi-objective optimization process are
study area-specific. But, in any situation the straight alignment will
generally yield the minimum user cost. The construction cost will depend
on the study area. A study area with almost level lands and minimum
land property cost along the straight line joining the start and end points
will yield a straight alignment for minimum user and construction costs.
In this situation all alternatives will converge to a single solution.
The generation of competing alternative highway alignments obtained
by multi-objective optimization approach, is an optimistic demonstration
of the methodology described in this paper. The present work is limited
to only two objectives, i.e., construction and user costs. Additional
objectives, such as impacts to sensitive environmental sites, historic sites,
and socio-economically sensitive sites can also be incorporated in the
multi-objective optimization process. This will make the optimization
process more complicated and extensive research is needed before
implementation.
Multi-Objective Highway… 501

Acknowledgements

The authors are thankful to Dr. Min-Wook Kang, research scientist at


ATRC, Inc. for his effort in preparing the GIS map for the example study
area from the town of Brookeville. The authors also acknowledge Dr.
Paul Schonfeld, Professor of Civil Engineering at the University of
Maryland, College Park for his continued collaboration in the highway
alignment optimization work with the Morgan State University. This
work was completed at the Morgan State University, Center for
Advanced Transportation and Infrastructure Engineering Research
(CATIER) (www.eng.morgan.edu/~catier). It is part of the first author’s
doctoral dissertation research.

References

Athanassoulis, G.C. and Calogero, V. (1973). Optimal Location of a


New Highway from A to B – A Computer Technique for Route
Planning, PTRC seminar proceedings on cost models and
optimization of highways (Session L9), London.
Chen, L., McPhee, J. and Yeh, W.W.-G. (2006). A Diversified
Multiobjective GA for Optimizing Reservoir Rule Curves, Advances
in Water Resources (In Press).
Chew, E.P., Goh, C.J. and Fwa, T.F. (1989). Simultaneous Optimization
for Horizontal and Vertical Alignments for Highways,
Transportation Research Part B, Vol. 23B, No. 5, pp 315-329.
Corne, D., Knowles, J. and Oates, M. (2000). The Pareto Envelop-based
Selection Algorithm for Multi-Objective Optimization, In
Proceedings of the Sixth International Conference on Parallel
Problem Solving from Nature VI (PPSN-VI), pp. 839-848.
Deb, K. (2002). Multi-Objective Optimization using Evolutionary
Algorithms, John Wiley & Sons, Ltd, Chichester, ISBN 9814-12-
685-3.
Deb, K., Mohan, M. and Mishra, S. (2005). Evaluation the ∈-
Domination Based Multi-Objective Evolutionary Algorithm for a
Quick Computation of Pareto-Optima Solutions, Evolutionary
Computation, vol. 13, no. 4, pp. 501-525.
Fonseca C.M. and Fleming P.J (1995). An Overview of Evolutionary
Algorithms in Multiobjective Optimization, Evolutionary
Computation, vol. 3, no. 1, pp. 1–16.
502 A. Maji and M.K. Jha

Gen, M. and Cheng, R. (1997). Genetic Algorithms and Engineering


Design, John Wiley & Sons Inc., New York, ISBN 0471127418.
Hogan, J.D. (1973). Experience with OPTLOC – Optimum Location of
Highways by Computer, PTRC seminar proceedings on Cost
Models and Optimization in Highways (Session L10), London.
Horn, J., Nafploitis, N. and Goldberg, D. (1994). A Niched Pareto
Genetic Algorithm for Multi-Objective Optimization, In Proceedings
of the first IEEE Conference on Evolutionary Computation,
Michalewicz, Z. edition, Piscataway, N.J., IEEE press, pp. 82-87.
Howard, B.E., Bramnick, Z. and Shaw, J.F.B. (1968). Optimum
Curvature Principle in Highway Routing, Journal of the Highway
Division, ASCE, Vol. 94., No. HW1, Proceeding paper no. 5987, pp
61-82.
Jha, M.K. (2000). A Geographic Information Systems Based-Model for
Highway Design Optimization, PhD Dissertation, Department of
Civil Engineering, University of Maryland, College Park.
Jha, M.K., P. Schonfeld, J.-C. Jong, and E. Kim (2006). Intelligent Road
Design, WIT Press, SouthHampton, UK, ISBN: 1-84564-003-9, 448
pp.
Jha, M.K. (2003). Criteria-Based Decision Support System for Selecting
Highway Alignments. Journal of Transportation Engineering. 129(1),
33-41.
Jha, M.K. and Maji, A. (2007). A Multi-Objective Genetic Algorithm for
Optimizing Highway Alignments, Proceedings of the 2007 IEEE
Symposium on Computational Intelligence in Multi-Criteria
Decision Making, Honolulu, HI.
Jha, M.K., C. Davis, and M.-W. Kang (2007). State-of-the-art Intelligent
Road Design Model with Genetic Algorithms, Geographic
Information Systems, and CADD, Advances in Transportation
Studies An International Journal, Section A&B, 13, 41-52.
Jha, M.K. and P. Schonfeld (2000). Integrating Genetic Algorithms and
GIS to Optimize Highway Alignments. Transportation Research
Record, Journal of the Transportation Research Board, 1719, 233-
240.
Jha, M.K. and P. Schonfeld (2004). A Highway Alignment Optimization
Model using Geographic Information Systems. Transportation
Research, Part A, 38(6), 455-481.
Jong, J.C. (1998). Optimizing Highway Alignment with Genetic
Algorithms, PhD Dissertation, Department of Civil Engineering,
University of Maryland, College Park.
Multi-Objective Highway… 503

Jong, J.C., Jha, M.K. and Schonfeld, P. (2000). Preliminary Highway


Design with Genetic Algorithms and Geographic Information
Systems, Computer-Aided Civil and Infrastructure Engineering, Vol.
15, No. 4, pp. 261-271.
Jong, J.C. and Schonfeld, P. (2003). An Evolutionary Model for
Simultaneously Optimizing 3-Dimensional Highway Alignments,
Transportation Research Part B, Vol. 37, No. 2, pp. 107-128.
Kang, M.-W., Jha, M.K. and Schonfeld, P. (2006). Three-Dimensional
Highway Alignment Optimization for Brookeville Bypass,
proceedings of the 85th Annual Transportation Research Board
Meeting, Washington, D.C.
Knowles, J.D. and Corne, D.W. (2000). Approximating the Non-
dominated Front using the Pareto Archived Evolution Strategy,
Evolutionary Computation Journal, vol. 8 no. 2, pp. 149-172.
Maji, A. (2008). Multi-Objective Highway Alignment Optimization,
Ph.D. Dissertation, Department of Civil Engineering, Morgan State
University, Baltimore.
Michalewicz, Z. (1996). Genetic Algorithm + Data Structures =
Evolution Programs, Third Edition, Springer-Verlag, New York.
MSHA (2001). MD 97: Brookeville Transportation Study, Maryland
State Highway Administration, Baltimore, MD.
Nash, S.G. and Sofer, A. (1996). Linear and Nonlinear Programming,
McGraw-Hill Companies Inc, New York, ISBN-007-046065-5.
Nicholson, A.J., Elms, D.G. and Williman, A. (1976). A Variational
Approach to Optimal Route Location, Highway Engineers, Vol. 23,
pp. 22-25.
OECD (1973). Optimization of Road Alignment by the Use of
Computers, Organization of Economic Co-operation and
Development.
Osyczka, A and Kundu, S. (1995). A New Method to Solve Generalized
Multicriteria Optimization Problems using the Simple Genetic
Algorithm, Structural Optimization, vol. 10, no. 2, pp. 94-99.
Osyczka, A and Kundu, S. (1996). A Modified Distance Method for
Multicriteria Optimization, Using Genetic Algorithms, Computers
and Industrial Engineering, vol. 30, no. 4, pp. 871-882.
Parker, N.A. (1977). Rural Highway Route Corridor Selection,
Transportation Planning and Technology, Vol. 3, pp. 247-256.
Schaffer, J.D. (1984). Some Experiments in Machine Learning using
Vector Evaluated Genetic Algorithms, PhD Thesis, Nashville,
Vanderbilt University, TN.
504 A. Maji and M.K. Jha

Shaw, J.F.B. and Howard B.E. (1982). Expressway Route Optimization


by OCP, Transportation Engineering Journal of ASCE, Proceedings
of the American Society of Civil Engineers, ASCE, vol. 10-8, No.
TE3, pp 227-243.
Srinivas, N and Deb, K. (1994). Multi-Objective Function Optimization
using Non-dominated Sorting Genetic Algorithms, Evolutionary
Computational Journal, vol. 2, no. 3, pp. 221-248.
Statnikov, R., Bordetsky, A., Statnikov, A. and Yanushkevich, I. (2006).
Multicriteria Analysis Tools in Real-Life Problems, vol. 52, pp. 1-32.
Thomson, N.R. and Sykes, J.F. (1988). Route Selecting through a
Dynamic Ice Field Using the Maximum Principle, Transportation
Research Part B, Vol. 22B, No. 5, pp 339-356.
Trietsch, D. (1987). A Family of Methods for Preliminary Highway
Alignment, Transportation Science, Vol. 21, No. 1, pp. 17-25.
Turner, A.K. (1978). A Decade of Experience in Computer Aided Route
Selection, Photogrammetric Engineering and Remote Sensing, Vol.
44, pp. 1561-1576.
Turner, A.K. and Miles, R. D. (1971). A Computer Assisted Method of
Regional Route Location, Highway Research Record, No. 348, pp.
1-15.
Wan, F.Y.M. (1995). Introduction to the Calculus of Variations and its
Applications, Chapman & Hall, New York, ISBN-10: 0-412-05141-
9.
Zitzler, E. and Thiele, L. (1999). Multi-Objective Evolutionary
Algorithms: A Comparative Case Study and Strength Pareto
Approach, IEEE Transactions on Evolutionary Computation, vol. 3
no. 4, pp. 257-271.

You might also like