Anti-Ship Missile Defense For A Naval Task Group

Anti-Ship Missile Defense for a Naval Task Group
Orhan Karasakal,
1
Nur Evin zdemirel,
1
Levent Kandiller
2
1
Industrial Engineering Department, Middle East Technical University, 06531 Ankara, Turkey
2
Industrial Engineering Department, ankaya University, Balgat 06530 Ankara, Turkey
Received 24 June 2009; revised 3 February 2011; accepted 7 February 2011
DOI 10.1002/nav.20457
Published online 16 March 2011 in Wiley Online Library (wileyonlinelibrary.com).
Abstract: In this study, we present a new formulation for the air defense problem of warships in a naval task group and propose a
solution method. We dene the missile allocation problem(MAP) as the optimal allocation of a set of surface-to-air missiles (SAMs)
of a naval task group to a set of attacking air targets. MAP is a new treatment of an emerging problem fostered by the rapid increase
in the capabilities of anti-ship missiles (ASMs), the different levels of air defense capabilities of the warships against the ASM
threat, and new technology that enables a fully coordinated and collective defense. In addition to allocating SAMs to ASMs, MAP
also schedules launching of SAM rounds according to shoot-look-shoot engagement policy or its variations, considering multiple
SAM systems and ASM types. MAP can be used for air defense planning under a given scenario. As thorough scenario analysis
would require repetitive use of MAP, we propose efcient heuristic procedures for solving the problem. 2011 Wiley Periodicals,
Inc. Naval Research Logistics 58: 305322, 2011
Keywords: air defense; naval task group; missile allocation problem; weapon-target allocation
1. INTRODUCTION
Air defense has been an important area for nations and
their armed forces. Substantial resources have been devoted
to develop both defensive and offensive weapon systems. The
effective use of and defense against these weapon systems is
of utmost importance. The proliferation of anti-ship missiles
(ASMs) and the increasing frequency of littoral operations
have increased the threat posed by the ASMs to the navies.
The competing technologies of ASMs and air defense sys-
tems force the navies to update the systems and to develop
new tactics continuously. All modern navies devote consid-
erable resources to ASM defense systems [9]. Planning for
effective use of those systems in operations has to be studied
carefully in order to be able to get the highest level of ben-
et from technological developments. One particular aspect
of planning is coordinated allocation of air defense systems
within a group of ships to attacking missiles. In naval ter-
minology, a collection of naval combatants and auxiliaries
that are grouped together for the accomplishment of one or
more missions is called a task group (TG). Modern area air
defense missile systems can provide support to the collocated
ships in a TG, and new technologies such as improved tacti-
cal data links and cooperative engagement capability enable a
Correspondence to: O. Karasakal (okarasa@yahoo.com)
fully coordinated air defense within a TG. For many navies,
equipping all the platforms with adequate air defense sys-
tems is clearly not the best and most cost-effective solution.
A number of navies acquire area air defense ships that can
provide air defense support to the other ships that have lim-
ited or no effective air defense capability. Allocation of the
area defense ship capabilities to other units in the TG is an
important problem to be solved for effective use of these
platforms.
For air defense engagements, there are several prescriptive
assignment models such as sector assignment, closest point
of approach assignment, maximum number of shots assign-
ment, least engagedassignment, defense indepthassignment,
and deliberate assignment of multiple overlapping systems
(see [28]). However, they do not make use of the addi-
tional capability provided by the coordination of defensive
resources and cohesion.
The problem we consider in this study is a specic type
of weapon-target allocation (WTA). A most generic form
of the WTA problem is the following [30]. Given an exist-
ing weapon force and a set of targets, what is the optimal
allocation of weapons to targets? The WTA problem can be
viewed both from an attackers and a defenders perspec-
tive. We restrict ourselves to the defense of the friendly naval
forces with surface-to-air missiles (SAMs) and focus on the
allocation of SAMs to incoming ASMs.
2011 Wiley Periodicals, Inc.
306 Naval Research Logistics, Vol. 58 (2011)
Matlin [29] and Eckler and Burr [11] review the literature
on the missile allocation problem. A number of articles have
been published in open literature after those reviews. Burr et
al. [8] develop the optimal integer solutions for Prim-Read
defense that uses the allowable damage per attacker instead
of the number of attacking missiles and builds the model
without explicitly knowing the attack size. Soland [36] con-
siders the defense of a single target against a simultaneous
attack. He assumes that the defense has interceptor missiles
with a xed number of engagements and uses shoot-look-
shoot engagement policy. Solands [36] solution procedure
is intractable as the size of the problem gets larger. Bert-
sekas et al. [6] propose a solution method for large problems
with known attack size using neuro-dynamic programming.
Krokhmal et al. [23, 24] consider uncertainty in WTA prob-
lem using conditional value-at-risk (CVaR) risk measure. In
[23], the authors develop models for one-stage and two-stage
stochastic WTA problems with CVaR constraints. In [24],
they assume known number of targets in the rst stage and
unknown number of targets with known probability distribu-
tion of the number of second stage targets within a two stage
recourse problem setting. They take into account the proba-
bility of running out of munitions on the second stage of the
engagement using the CVaR risk measure. Al-Mutairi and
Soland [3] analyze the effectiveness of partially coordinated
area defense systems against a simultaneous attack. Due to
the computational complexity of the missile allocation prob-
lem, a number of heuristic approaches have been suggested.
Wacholder [37] proposes a solution for a one sided many-
on-many missile allocation problem using articial neural
networks combined with the Lagrange differential multipli-
ers method. Jaiswal [17] investigates a similar problem using
simulated annealing, genetic algorithms and articial neural
networks in a layered defense context.
Nguyen et al. [33] introduce the idea of using generating
functions as a simple, consistent, and easily applied tool for
evaluating the effectiveness of an air defense system. This
approach does not provide any interceptor allocation plan.
The scenario considered is similar to that of Soland [36].
The model described in [33] is based on four parameters:
the total number of available interceptors, the total number
of attackers, the maximum number of engagement oppor-
tunities against each threat, and a constant probability of
successful interception. Al-Mutairi et al. [2] propose for-
mulations for analyzing a layered defense. Simulation is
one of the tools frequently used to evaluate the effective-
ness of the air defense systems. Hoyt [16] reports a simple
Monte-Carlo simulation model of ballistic missile defense
system. SEAROAD [7, 28], JASMINE [35] and SADM [10]
are examples of naval air defense simulations. Beare [5] and
Grifths et al. [15] describe the use of analytical models to
reduce the number of scenarios that would be investigated in
detail by simulation models.
Karasakal [19] andNguyen[32] investigate the cohesionof
air defense capability within a TG. In his work, Nguyen [32]
quanties the benet from resource allocation for a naval
TG having perfect coordination among its assets. The inter-
ceptors are assumed to cover all the other ships of the TG
and are capable of defending the ships within range. Other
geometric and defense systemlimitations are not considered.
Karasakal [19] uses the geometric information such as rela-
tive bearings and distances between ships and attacking mis-
siles, and capabilities of different types of SAM systems and
ASMs such as effective ranges and speeds. Proposed models
enforce an approximate shoot-look-shoot engagement policy.
The author claims that the models can be used to investi-
gate the composition of a naval TG and the effectiveness of
different SAM systems to protect the TG.
Almeida et al. [1] present the impact of information on
the effectiveness of air defense in a time-constrained context.
Sherali et al. [34] develop algorithms to schedule a set of
illumination radars to engage incoming targets using surface-
to-air missiles in a naval TG. Armstrong [4] analyze the effect
of lethality (i.e., the relative balance between the offensive
and defensive power of rival naval forces) to the outcome of
the naval combat represented by an aggregate level missile
salvo model. Lucas and McGunnigle [26] investigate the util-
ity of simple models to provide insights for military decision
problems by comparing a simple salvo model with a complex
simulation of naval combat.
In this study, we address the issue of allocating air defense
missiles to incoming air targets in a coordinated way within
a naval TG such that the available defense capability is used
in the most effective manner. We call our version of the mis-
sile allocation problem MAP. Our aim is to develop a MAP
model for TG air defense that captures the reality of ASM
defense, generates an efcient allocation plan for SAM sys-
tems, and measures the effectiveness of air defense under a
given scenario. Ascenario is composed of the information on
the attacking ASMs and the defensive SAM systems as well
as the relative positions of the ships in TG, which is called
the formation of the TG.
MAP is a new treatment of an emerging problem fostered
by the rapid increase in the capabilities of ASMs, and the
different levels of air defense capabilities of warships against
the ASM threat. In addition to allocating SAMs to ASMs,
MAP also schedules launching of SAM rounds according to
shoot-look-shoot engagement policy or its variations, con-
sidering multiple SAM and ASM types. Our objective is to
maximize the probability that none of the ASMs can reach
their target. MAP can produce the best course of action for
defending the TG against an immediate and simultaneous
ASMthreat. It can also be used for analysis of the air defense
effectiveness of warships under different positioning at sea.
For a given attack scenario and a positioning at sea (i.e., an
escort ship and a defended ship positioned at a specied range
Naval Research Logistics DOI 10.1002/nav
Karasakal, zdemirel, and Kandiller: Anti-Ship Missile Defense 307
and bearing from the escort ship), the air defense coverage
provided by the escort ship can be calculated using MAP.
The aggregated coverage values for a set of representative
attack scenarios and positions can be used to decide on the
disposition of the TG. Interested readers are referred to [20]
and [21] for a detailed explanation of the integrated solu-
tion approach to form a screen disposition of a TG using
MAP.
Although we provide a mathematical programming model
for MAP, we do not explicitly use the model to solve MAP.
Instead, we propose efcient and fast heuristic algorithms to
solve the problem. As for manyother combinatorial optimiza-
tion problems, rst we try to construct a good initial feasible
solution and then improve this solution further. We make use
of some conventional ideas such as 2-opt exchange, which is
used for typical ordering-based problems, e.g., the traveling
salesman and scheduling problems.
After introducing our original MAP formulation and solu-
tion procedure, we discuss extensions for alternative objec-
tives and for relaxing some of the assumptions including the
shoot-look-shoot engagement policy.
The literature reviewon the MAP, according to our knowl-
edge, shows that there is no model that can be used to solve the
problemdened in this article. The existing analysis methods
mainly consist of computer models that simulate the defense
against ASM attack. The Naval Studies Boards [30] panel
on Modeling and Simulation also identied the requirement
for analytical models in anti-air warfare area.
The rest of the article is organized as follows. In the next
section, we describe MAP and present a non-linear 01 inte-
ger programming model for the problem. In Section 3, we
propose efcient heuristic solution algorithms for MAP. In
Section 4, we present and discuss the results of the compu-
tational study undertaken to evaluate the performance of the
proposed heuristics. We discuss some extensions of MAP in
Section 5, followed by concluding remarks.
2. PROBLEM DESCRIPTION AND MODEL
Consider a naval TG, composed of several ships with vari-
able air defense capabilities, defending itself against an air
attack. These ships may either be equipped with one or more
surface-to-air missile (SAM) systems or none at all. Their
air defense capability may be limited to self-defense or may
extend to area defense, i.e., a ship may defend the other ships
within its effective weapon range. Ships in a TGare typically
arrayed into a formation, called a screen, in which the most
valuable and important units (termed high value unit or HVU)
are surrounded and protected by the escorting vessels. Within
the screen, the escort ships are stationed in sectors away from
the HVU. Figure 1 depicts a generic naval TG composition
and an air attack scenario. In this scenario, a TG composed
of four ships in formation, one HVUand three escort ships, is
attacked by four ASMs. Ship 1 (HVU with no SAM system
onboard) is targeted by ASM2 and ASM3, Ship 2 is targeted
by ASM1, and Ship 4 is targeted by ASM4. There is no ASM
threat toShip3. Ships 2, 3, and4have short-range self defense
SAM systems (such as the NATO Sea Sparrow SAM), and
the effective ranges are depicted by the circular areas around
each ship. These self defense SAM systems can only engage
the ASMs targeting their own ships. Ship 2 also has a long-
range area defense system SAM2 (such as the SM-2 SAM),
and part of its effective range is depicted by the dotted area
and the arc drawn in a dashed-line. Because SAM2 is an area
defense system, it can engage ASMs targeting the other ships
as well as Ship 2. ASM1 can be engaged by both SAM1 and
SAM2. ASM2 and ASM3 can be engaged by only SAM2.
Note that SAM4 cannot engage ASM3, even if some part of
the ASM3s ight path falls into the effective range of SAM4,
since SAM4 is a self-defense systemand can only engage the
ASMs that are a direct threat to it. ASM4 can be engaged by
both SAM2 and SAM4.
The TG air defense commander will maintain the air pic-
ture and coordinate the response until the time when the ships
are forced to defend themselves using the weapons for last
line of defense, such as close-in-weapon systems and soft
kill systems. The air defense command and control ship will,
in most cases, have to coordinate the TG response to an air
threat to ensure maximum efciency and probability of suc-
cess. In this role, a set of command decision tools is required
to plan the air defense of the TG, and to schedule the force
defense as an attack develops.
Maximizing the probability of shooting down all the
incomingASMs is animportant objective for a TGair defense
commander at sea. (We consider some alternative objectives
in Section 5.) However, saving the maximum number of
SAMs (for possible future attacks) from the limited num-
ber available onboard the ships and the high price tag of each
missile have to be considered as well. The objective might
be to maximize the probability of neutralizing the incom-
ing ASMs with minimal SAM expenditure. Several missile
engagement tactics have been developed to achieve a bal-
ance between these conicting objectives. Although we do
not explicitly minimize SAM expenditures, we enforce the
shoot-look-shoot (SLS) tactic to take this secondary objective
into account. The SLS tactic requires shooting at the target
rst, then looking to see if it is killed, and shooting again only
if it is necessary to achieve the kill. In this research, we ini-
tially consider the case when the TG employs the SLS tactic.
Variations of SLS are also discussed in Section 5.
The engagement process of a SAM system to an ASM
can be divided into four phases. These are the tracking of
the target illumination radar, the solution of the re control
problem, the launch delay, and the ight time to the engage-
ment. Each engagement of SAM systems takes a constant
Figure 1. Composition of a naval TG and an air attack scenario. [Color gure can be viewed in the online issue, which is available at
wileyonlinelibrary.com.]
setup time for the rst three phases and a variable time for
the last phase, which is the ight time to the engagement. We
assume that the time required for checking whether or not the
ASM is hit at the end of one engagement overlaps with the
setup time of the next engagement, hence there is no need to
consider this time separately. Alternatively, the look time
can be added to the setup time. Each engagement takes less
time compared with the one before as the attacking ASM
approaches its target.
The maximumdistance at whichanASMintercept cantake
place is determined by the smallest of the following three: the
maximum effective range of the SAM system, the horizon of
the re control radar against the incoming ASM, and the rst
detection range of the ASM.
When the SLS ring policy is used, there are few engage-
ment opportunities (mostly less than 10) against each ASM.
In reality, each engagement does not take the same setup time,
since the target illumination radar may already be on track,
or the re control problem may have already been solved.
However, we use a conservative approach and consider that
each engagement takes a constant setup time.
In summary, MAP is concerned with the allocation of
different types of SAMs available for attacking ASMs and
schedulingthe SAMlaunches under SLSringpolicy, soas to
maximize the TGs air defense capability, which is measured
by the probability of no leaker.
The basic assumptions that are needed to develop the
missile allocation model are stated below.
1. The TG sees all of the air threats to intercept simul-
taneously. Thus, we investigate the case where the
attack size is known.
2. The ships in the TG are capable of coordinating the
allocation of the air defense, i.e., C
2
(command and
control) capability is assumed.
3. The TG has both point (or self) and area air defense
missile systems.
4. Both attacking ASMs and SAM systems onboard
ships may be of different types.
5. Different SAM systems may have different effective
ranges, i.e., layered defense is assumed. Layers may
overlap.
6. Defense systems can predict the eventual target of the
attacking ASMs, i.e., impact point prediction capa-
bility is assumed. The track of each ASMis a straight
line.
7. Defense systems can distinguish the ASMs from
each other, i.e., a SAMintended for a particular ASM
actually locks on that ASM.
8. Missile engagement policy is SLS.
9. The incoming ASMs are assumed to be classied in
terms of their speed(e.g., supersonic or subsonic) and
attack prole (e.g., sea-skimmer, high diver). Thus,
the single shot kill probability of each SAM against
each ASM is known.
10. The relative positions of the ships within the TG do
not change as the air raid continues. The ships are
thought to be stationary. This is a reasonable assump-
tion since the speed of the ships is very lowcompared
with the speed of the ASMs.
11. There are no limitations on the number of SAMs in
ight that are launched from the same SAM system
to engage different ASMs.
We discuss howto relax assumptions 6, 8, and 11 in Section 5.
Suppose that there are nincomingASMs, indexedi N =
{1, . . . , n} and there are mSAMsystems on board of warships
composing the naval TG, indexed j M = {1, . . . , m}. Let
t
i
be the time taken by ASM i to reach its known target.
Then H = max
iN
{t
i
} is the problem horizon given by the
highest time-on-target. The interval [0, H] may be divided
into t non-overlapping slots, each of unit duration , indexed
k K = {1, . . . , t }, and
k
denotes the beginning time of
slot k, k K. Let V denote the valid combinations of ASM
and SAM systems, i.e., (i, j) V if SAM system j can
engage ASM i. Each ASM i has a specied engageability
interval [q
ij
, r
ij
], which depends on the location and capabil-
ity of the SAM system j, and a successful engagement can
be achieved only during this interval. Here, q
ij
refers to the
earliest beginning time of the rst engagement, and r
ij
refers
to the latest ending time of the last engagement, i.e., the lat-
est time the last round of SAM j should intercept ASM i.
We assume that the problem data related with time have been
perturbed in such a way that each value is an integer multiple
of the unit time . Time taken by each feasible engagement is
determined as the sumof a constant setup time and a variable
ight time to the engagement. Thus, each engagement takes
a specied time according to the ASM and SAM combina-
tion (i, j) V and the starting time of the engagement. This
engagement duration is denoted by
ijk
.
The above parameters are calculated based on the scenario
specications. t
i
is found by dividing the distance between
ASM i and the ship it targets by the velocity of the ASM. q
ij
and r
ij
are calculated similarly using the minimum and max-
imum engagement ranges of SAMs and the initial detection
distances of ASMs. The constant setup time included in
ijk
is assumed to be given. The variable ight time to engage-
ment is found based on the distance between ASM i and the
target ship that has SAM j at the beginning of time slot k,
as well as the velocities of the ASM and SAM pair and the
SAM y out trajectory.
To formulate the problem, recalling that
k
denotes the
beginning time of slot k K, let us dene for each valid
combination of ASM i N and SAM j M, a set
S
ij
= {k K : (i, j) V and [
k
,
k
+
ijk
] [q
ij
, r
ij
]}.
Note that S
ij
denotes the slots for which SAM j can be
scheduled to engage ASM i. For example, suppose
ijk
has
the respective values 5,4,3 for k = 1, 2, 3. (The engagement
duration decreases as the ASM approaches its target.) Then
S
ij
= {1, 2, 3} means that [1,5], [2,5], and [3,5] are possible
engagement intervals.
Accordingly we dene the binary decision variable x
ijk
=
1 if SAM j is scheduled to start the engagement process
against ASM i at the beginning of Slot k, and x
ijk
= 0 oth-
erwise. Furthermore, to ensure that schedules of the SAMs
against an ASM do not overlap in accordance with the SLS
tactic, let us dene for each slot k K and for each ASM
i N, the set
J
ik
= {(j, ) : (i, j) V, S
ij
, and [
k
,
k
+ ]
[
+
ij
]}.
Note that for each i N and k K, J
ik
is the set of combi-
nations (j, ) such that slot k for ASM i will be blocked (to
disallow other SAMs being red during slot k) if x
ij
= 1.
For example, J
i3
= {(j, 1), (j, 2), (j, 3)} indicates that Slot
3 will be blocked if SAM j starts engaging ASM i in time
Slots 1,2,3.
We need the following additional notation and variables to
formulate the TG air defense problem.
p
ijk
: the single shot kill probability of SAMj against ASM
i when the engagement begins at the beginning of slot
k, (i, j) V and k S
ij
.
d
j
: the number of available rounds on SAM system j.
u
ij
: the upper bound on the number of engagements of
SAM system j against ASM i, (i, j) V within the
engageability interval. u
ij
is calculated by dividing
the engageability interval by the minimum duration
of a single engagement according to the SLS tactic.
Then, the TG air defense problem MAP can be formulated
as the following nonlinear integer programming model.
Max
iN
_
_
_1
kK
jM|(i,j)V
(1 p
ijk
)
x
ijk
_
_
_ (1)
subject to
kK
iN|(i,j)V
x
ijk
d
j
for all j M (2)
(j,)J
ik
x
ij
1 for all i N and k K (3)
kS
ij
x
ijk
u
ij
for all (i, j) V (4)
x
ijk
{0, 1} for all (i, j) V and k S
ij
(5)
The objective function (1) maximizes the probability of
no-leaker for the whole TG. Constraint set (2) reects the
restriction on the number of rounds available for each SAM
system. Constraints of type (3) ensure that there is no over-
lap of the engagements against each ASM. Constraints of
type (4) limit the total number of rounds that can be red for
each valid ASM and SAM combination. This constraint set
tightens the feasible space of the problem. Constraint set (5)
imposes binary restriction on the decision variables.
3. SOLUTION PROCEDURE
The non-linearity in the model can be removed by using
logarithms and exploiting the relation between the objective
function and the right-hand side of a newly added constraint
set coming from the objective function [18]. However, even
a small problem can become quite large in terms of the num-
ber of binary decision variables as the unit duration, , gets
smaller.
The WTA problem is hard to solve in terms of the com-
putational complexity. Indeed, Lloyd and Witsenhausen [25]
prove that the weapon allocation problem is NP-complete
even in its simplest form. WTA models in literature use
simplifying assumptions to reduce the problem to a level
of suitable mathematical tractability. Our MAP formulation
uses parameters and assumptions required by the specic
air defense scenario we consider. Thus, further simplifying
assumptions cannot be used in our model without sacricing
the representation of the real situation.
MAP can be used to analyze TG air defense effectiveness
in different scenarios and settings. Solving MAP for a large
number of representative cases is a prerequisite for a thor-
ough scenario analysis. Since this process requires running
MAP many times, we need fast solution procedures that yield
high-quality solutions.
The mathematical programming model presented in the
preceding section does not meet the solution time require-
ment due to its computational complexity. Thus, we focus
on heuristic solution procedures for MAP. We present here
two greedy construction and two improvement algorithms for
MAP.
The rst of those construction algorithms, the best engage-
ment construction heuristic, allocates SAM systems to
incoming ASMs according to a measure called the engage-
ment potential. Its aim is to nd the best engagement pairs in
terms of the objective. Inthe quasi-uniformconstructionalgo-
rithm, we try to engage each ASM at least once to reduce the
risk of a zero objective function value. Thus, we give prece-
dence tothe ASMthat has the lowest number of SAMsystems
that can engage it. After nding a starting solution by using
these construction algorithms, we try to improve the solution
by using improvement algorithms. One of the improvement
algorithms, opt-change (OC) algorithm, improves the initial
feasible engagement schedule by changing the target ASM
or SAM system of an engagement in the engagement list. In
the 2-opt exchange (2OX) algorithm, we aimto exchange the
target ASMs of two engagements to improve the solution.
We choose the best move in each iteration of improvement
algorithms.
By dividing the planning horizon into small time slots
and allowing engagements to start at the beginning of these
time slots, our formulation approximates the continuous time
problem. Naturally, solution of this discrete time formula-
tion may not be optimal for the continuous time problem.
However, we do not solve the formulation explicitly but nd
the optimum with enumeration. Both the enumeration proce-
dure and the heuristic algorithms come up with a sequence of
SAM engagements against incoming ASMs observing the
SLS tactic. In both methods, an engagement can start as
soon as the previous one ends according to the generated
engagement sequence. Therefore, they are not affected by
the discretization, and their results are directly comparable.
3.1. Best Engagement Construction (BEC) Algorithm
First, we present the additional notation and the variables
for the construction algorithm below.
va
i
: speed of ASM i
vs
j
: speed of SAM j
r
j
: maximum effective range of SAM j
r
j
: minimum effective range of SAM j
c
: constant setup time for an engagement
f
i
: initial detection distance of ASM i from its
known target
pf
i
: present distance of ASM i from its known
target
p
ij
: single shot kill probability of SAMj against
ASM i, (i, j) V
ep
ij
: engagement potential of SAM j against
ASM i
w
1
, w
2
, w
3
, w
4
: weights of the components of the engage-
ment potential
G
i
: set of engagement potentials of the SAMs
that can be used against ASM i
t
i
: time for ASM i to reach its target, i.e.,
time-on-target (TOT), t
i
= f
i
/va
i
T : set of TOTs of ASMs
In this algorithm, we allocate SAM rounds to ASMs
according to the engagement potential, which is a measure
of the defensive capability of a SAM system against a given
ASM. We compare each SAM system with a hypothetical
SAM, which has the best features such as the largest single
shot kill probability, highest speed, longest effective range,
and shortest effective range against a given ASM. We assign
the SAMwith the highest engagement potential to the closest
ASM in terms of TOT at each step of the algorithm.
STEP 0: Determine the ideal SAM for each ASM.
Find the best features, vs
i
, r
i
, r
i
, p
i
for ASM i using all
SAM systems that can be used against that ASM. Dene a
new SAM called ideal SAM with these best features. (Maxi-
mum range may be limited to the initial detection distance of
ASM if the detection distance is smaller than the maximum
effective range of SAM.)
vs
i
= max
j
{vs
j
: (i, j) V},
r
i
= min
_
f
i
, max
j
{ r
j
: (i, j) V}
_
,
r
i
= min
j
{r
j
: (i, j) V},
p
i
= max
j
{p
ij
: (i, j) V}.
Initialize present ASM distances to initial detection dis-
tances, pf
i
= f
i
i N.
STEP 1: Determine the engagement potential of each
SAMsystemagainst each ASMif the engage-
ment is feasible. Find the set of engagement
potentials of the SAMs that canbe usedagainst
ASM i.
ep
ij
= w
1
vs
j
vs
i
+ w
2
min
_
r
j
r
i
, 1
_
+ w
3
r
i
r
j
+ w
4
p
ij
p
i
for (i, j) V,
G
i
= {ep
ij
: (i, j) V}.
STEP 2: Determine the TOT for each ASM, and nd
the set of TOTs of ASMs, T .
STEP 3: If all ASMs have been engaged, then start
a new engagement wave. If T = { }, then
re-populate T = {t
i
: i N}.
This step ensures that the nal engagement
schedule is as uniform as possible and that the
ASMs are engaged with a more or less equal
number of SAMs.
STEP 4: Find the ASM with minimum TOT, and
remove its TOT from the engagement list, T .
a = arg min
i
T , T = T \{t
a
}.
STEP 5: If there are no SAM missiles left that can be
used against any of the ASMs, stop. If G
i
= { }
i N, then STOP.
STEP 6: If there is no SAM system that can be used
against ASM a, then return to Step 3; other-
wise, nd the SAMsystemwith the maximum
engagement potential against the ASM in the
engagement order.
If G
a
= { } then go to Step 3; otherwise,
b = arg max
j
G
a
.
STEP 7: If there is at least one SAMround of type b and
the intercept distance is larger than the min-
imum engagement range of SAM b, allocate
SAMb to ASMa. Reduce the number of avail-
able rounds of SAM b by one and go to Step
3. Otherwise, update the set of engagement
potentials and go to Step 5.
If d
b
1 and
_
pf
a
va
a
_
pf
a
va
a
c
va
a
+ vs
b
_
va
a
_
r
b
then,
If the intercept distance is larger than
the maximum engagement range of
SAM b, then reduce the intercept dis-
tance to the maximum engagement
range of SAM b, i.e.,
If
_
pf
a
va
a
_
pf
a
va
a
c
va
a
+ vs
b
_
va
a
_
r
b
, then pf
a
= r
b
,
else pf
a
= pf
a
va
a
_
pf
a
va
a
c
va
a
+ vs
b
_
va
a
.
Allocate SAM b to ASM a and set
d
b
= d
b
1. Go to Step 3.
Otherwise, G
a
= G
a
\{ep
ab
} and go to
Step 5.
In Step 7 of the algorithm, the net closing velocity of a
SAMand ASMpair is calculated as if they have a 180 degree
angle of deection. These computations are infact muchmore
complicated, involving not only the deection angles but also
various sources of error due to external factors such as wind
and temperature, and internal factors such as device and mea-
surement errors. These computations are beyond the scope of
our study, and we assume head-on intercepts for all engage-
ments for simplicity. However, longer elapsed time is the
only implication when the deection angle is ignored, and
this does not affect our methodology and workings of the
algorithms.
l[(n 1)m + (m 1)] different cases are considered
for change and enhancement in each iteration of the algo-
rithm, where l is the total number of engagements. The
computational complexity for OC algorithm is O(lmn) per
iteration.
3.2. Quasi-Uniform Construction (QUC) Algorithm
The BEC algorithm assigns the SAM with the highest
engagement potential to the closest ASM in terms of TOT.
However, if the number of missiles in magazine or launcher
is limited, the assignment rule may produce unsatisfactory
results. Note that the probability of no-leaker will be zero
by allocating anything less then one shot per ASM. This dis-
continuity, the jump from zero to a positive probability of
no-leaker value as the last ASM in the rst engagement wave
is shot at, causes difculties for our construction algorithm.
If there is an engagement schedule that has at least one shot
per ASM, then it is desirable to nd that one. This variation
makes sure that we nd the desirable engagement schedule
if there is one. To achieve this goal, we change Step 3 of the
previous algorithm as follows.
STEP 3: If T = { } and there exists at least one ASM
with no interceptor assigned, then disregard all
assignments made so far and let TOTs be the
cardinality of the corresponding set of engage-
ment potentials, i.e., T = {t
i
= |G
i
| : i N}.
Else if T
= { }, then re-populate T = {t
i
: i N}.
Here, t
i
= |G
i
| gives the number of SAM systems that
can engage ASM i. In Step 4, by choosing the ASM having
minimum t
i
value, the algorithm tries to allocate at least one
SAM shot to each ASM.
3.3. Opt-Change (OC) Algorithm
Our purpose in this algorithm is to nd the engagements
that would increase the objective function value by (1) chang-
ing the target ASM of an engagement under consideration
and (2) simultaneously considering the enhancement of the
defense effectiveness by increasing the total number of SAM
missiles launched against target ASMs. Changing the tar-
get ASM means that while one ASM will get one less shot,
another ASM will get one more shot. The ASM that gets one
less shot after the change is considered for an additional shot
observing the SLS tactic. We summarize the OC algorithm
below and give its full description in Appendix A.
STEP 0: Select an initial feasible engagement list.
STEP 1: For each engagement in the list, check the pos-
sibility of the change of target ASM. Achange
of target ASM will degrade defense against
the target ASM before the change, and will
enhance the defense against the new target
ASM. To improve the overall defense capa-
bility, we simultaneously consider enhancing
the defense against the former ASM by using
the remaining SAM rounds, if any.
STEP 2: Consider changing the defending SAM for
each engagement in the list.
STEP 3: Find the best change in Steps 1 and 2. If there
is an improvement, update the engagement list
and go back to Step 1. Otherwise, stop.
l[(n 1)m + (m 1)] different cases are considered
for change and enhancement in each iteration of the algo-
rithm, where l is the total number of engagements. The
computational complexity for OC algorithm is O(lmn) per
iteration.
3.4. 2-Opt-Exchange (2OX) Algorithm
Our purpose in this algorithm is to nd the engagement
pairs that would increase the objective function value by
exchanging the target ASMs in the pair. We also try to
increase simultaneously the number of engagements against
the ASMs under consideration with each exchange. We sum-
marize the 2OXalgorithmbelow. The full description is given
in Appendix B.
STEP 0: Select an initial feasible engagement list.
STEP 1: For each pair of engagements in the list,
check the possibility of the exchange of tar-
get ASM. Simultaneously consider enhancing
the defense against both target ASMs using the
remaining SAM rounds.
STEP 2: Consider exchanging all the scheduled
engagements of the two ASMs.
STEP 3: Find the best exchange in Steps 1 and 2.
Update the engagement list, if necessary. If
there is an improvement, go back to Step 1.
Otherwise, stop.
l(l1)
2
(2m+ 1) different neighboring engagement lists are
checked for exchange for each iteration of the algorithm. If an
exchange is made, then the algorithm starts over again. The
algorithmstops when there is no exchange possible. Note that
anundesirable exchange maybecome desirable after a change
in the engagement list. Thus, we continue until no desirable
exchange is left for the engagement list. The computational
complexity for 2OX algorithm is O(l
2
m) per iteration.
4. COMPUTATIONAL RESULTS
In the preceding section, we developed heuristic algo-
rithms for MAP. Since heuristic algorithms do not guarantee
an optimal solution, it is of interest to study the worst-case
performance of the algorithms. Submodularity concept is
usedtoprove the worst-case performance of greedytype algo-
rithms (see [22]). Submodularity states that adding an ele-
ment to a smaller set helps more than adding it to a larger set.
Formally, F(A{j})F(A) F(A{j}{k})F(A{j})
must hold for all A I and j I\(A {k}) where F is a
set function dened over a nite set I [31]. We can show that
our objective is not submodular using a simple counter exam-
ple. Let us assume a scenario with three attacking ASMs,
i N = {1, 2, 3}, one SAM system, j M = {1} having
one available missile, d
1
= 1, and p
i1
0 and u
i1
1 for all
i. In this case, our objective function value is 0. Adding one
more available missile does not change the objective function
value. However, adding the third missile suddenly produces
a positive objective function value, which contradicts with
submodularity.
Below, we empirically show the performance of our algo-
rithms by comparing their results with the optimal solutions
calculated using implicit enumeration and with the upper
bounds calculated when the optimal solution is not tractable
using implicit enumeration.
We randomly generated test problems using the informa-
tion on real weapon systems in open literature. Interested
readers are referred to [12, 13, 27] for more information
on ASM and SAM systems. We dened seven different
SAM systems, including four self-defense and three area air
defense SAM systems. Self defense SAM systems are Sea
Sparrow, Evolved Sea Sparrow(ESSM), Aster-15 and Barak.
Area defense SAM systems are SM-1, SM-2, and Aster-30.
We also assumed seven ASM systems including Harpoon,
Exocet, Polyphem, Gabriel, Penguin, SS-N-26, and Maver-
ick. We createda sample single shot kill probabilitymatrixfor
SAMand ASMsystems. We choose the single shot kill prob-
abilities by considering the speed of both SAM and ASM,
ASMs ight prole (i.e., sea-skimming or high diving) and
the technological age of the weapons. We assume single shot
kill probabilities between 0.15 and 0.9. For a given number
of ASM and SAM systems, we randomly generate the type
Table 1. Minimum, average and maximum elapsed time for
enumeration.
Number of
Number of SAM Systems
ASMs 1 2 3 4 5
1 Min
a
0.0
d
0.0 0.0 0.0 0.0
Ave
b
0.0 0.0 0.0 0.0 0.0
Max
c
0.0 0.0 0.0 0.0 0.0
2 Min 0.0 0.0 0.0 0.0 0.0
Ave 0.0 0.1 1.2 0.2 0.5
Max 0.0 0.6 3.5 0.6 1.8
3 Min 0.0 0.0 0.2 0.0 0.0
Ave 0.2 0.8 9.9 1.6 2.9
Max 1.2 2.8 40.7 4.5 9.8
4 Min 0.0 0.4 0.0 1.3 0.3
Ave 94.5 11.0 62.4 53.5 264.5
Max 414.0 41.9 269.4 216.3 1294.6
5 Min 0.2 0.1 38.6 5.0 0.1
Ave 891.8 56.0 1736.5 268.7 1731.7
Max 2191.0 228.8 7008.6 1127.9 8503.3
a
Minimum.
b
Average.
c
Maximum.
d
CPU time (sec) on a personal computer with AMD Athlon 2000+
CPU and 256 MB of RAM.
of ASMs, initial detection range of ASMs (between 5000 m
and 40,000 m or maximum range of the specic ASM if its
range is less than 40,000 m), type of ships (a ship has only
SD capability or both SD and AAD capabilities with equal
probability), and target ships of ASMs. We decide on the
number of missiles on each SAM system randomly between
1 and 10. We set the weights used for the components of
the engagement potential to one in our experiments, as rel-
ative values of all four components are between zero and
one.
We developed an enumeration algorithm to nd the opti-
mal solution for MAP. Since complete enumeration is still
very time consuming, we restrict the problem size to a maxi-
mum of ve SAM systems with a total of nine missiles in the
launchers and ve ASMs. There are 25 combinations for the
number of SAM systems and ASMs. For each combination
we generated ve problems. Problem parameters including
the types of SAM systems and ASMs, the initial number of
missile of each SAM system, the target of the ASMs and the
initial detection ranges of ASMs were generated using dif-
ferent random number streams. Table 1 shows the summary
of the computational time for the enumeration algorithm for
all 125 sample MAPs.
Table 2 presents the summary of sample MAPs in terms of
the minimum, average, and maximum percent gaps between
the optimal solution and the best of the two construc-
tion heuristic solutions. We calculate the gap as 100
(optimal-heuristic)/optimal objective function value. The
Table 2. Minimum, average and maximum % gap between the
best construction heuristic and optimal solutions.
Number of
Number of SAM Systems
ASMs 1 2 3 4 5
1 Min
a
0.0 0.0 0.0 0.0 0.0
Ave
b
0.0 0.0 0.0 0.0 0.0
Max
c
0.0 0.0 0.0 0.0 0.0
2 Min 0.0 0.0 0.0 0.0 0.0
Ave 0.0 4.0 4.8 2.1 1.5
Max 0.0 20.0 23.8 10.7 7.3
3 Min 0.0 0.0 0.0 0.0 0.0
Ave 2.4 2.0 3.0 0.4 1.2
Max 5.9 5.7 14.8 2.0 6.0
4 Min 0.0 0.0 0.0 0.0 0.0
Ave 0.0 11.0 7.6 19.2 20.0
Max 0.0 17.9 33.3 44.9 100.0
5 Min 0.0 0.0 0.0 0.0 0.0
Ave 5.1 20.8 16.3 14.0 10.3
Max 11.4 38.6 38.6 41.9 23.1
a
Minimum % gap.
b
Average % gap.
c
Maximum % gap.
QUC algorithm produced better engagement schedules for
17 out of 125 cases compared to BEC algorithm. Those
cases where the QUC algorithm produced better results do
not represent any identiable pattern. Construction heuris-
tics failed to produce the optimal solution in 38 out of 125
cases. Although the construction heuristics attained the opti-
mal solution in 70% of the test cases, we conclude that they
may frequently produce unsatisfactory results.
We ran our improvement algorithms for those 38 cases
where the construction algorithms failed to produce the opti-
mal solution. Two different combinations of the improvement
algorithms were also investigated. One of those combinations
(OC + 2OX) is running OC rst and then 2OX. The other
(2OX + OC) is running 2OX rst and OC next. The sum-
mary results of improvement heuristics are given in Table 3.
The last column of Table 3 depicts the best results of the
improvement heuristics. The best result may be viewed as
another heuristic that runs OC + 2OX and 2OX + OC, and
takes the best of the two solutions.
We provide some measures of solution quality for heuris-
tics OC+2OX, 2OX+OC, and Best in Table 4. OC+2OX
dominates 2OX + OC with respect to ve measures given
in Table 4. Best provides a slight improvement on the
OC + 2OX results. OC + 2OX attains the optimal solution
in 33 out of 38 problems. In one out of the remaining ve
cases where OC+2OX failed to achieve the optimal results,
Best was found by 2OX + OC. We statistically compared
Best and OC + 2OX against 2OX + OC heuristic using
Wilcoxon signed rank test as described in Golden and Stew-
art [14]. Wilcoxon tests showed that Best and OC + 2OX
heuristics are statistically better than 2OX+OC heurisitic at
= 0.05 signicance level.
For all small test problems presented above, our MAP
solutionprocedure producedhigh-qualitysolutions while sat-
isfying the run time requirement for MAP. We used small
test problems to be able to compare the heuristic results with
the optimal results. We restricted the number of total SAMs
to nine. Thus, the average number of missiles available on
Table 3. % Gap between optimal solution and the improvement
heuristics for the problems where constructions heuristics failed to
nd the optimal solution.
Best of Improvement algorithms
Problem construction
Number
a
heuristics OC 2OX OC+2OX2OX+OC Best
I.4.2 16.0 15.4 13.8 0.0 0.0 0.0
I.4.4 44.9 1.2 44.9 0.0 1.2 0.0
I.5.2 35.7 15.4 0.0 0.0 0.0 0.0
I.5.3 9.7 9.7 0.0 0.0 0.0 0.0
I.5.4 41.9 1.2 24.7 0.0 24.7 0.0
I.5.5 22.9 4.8 0.0 4.8 0.0 0.0
II.3.1 5.9 0.0 5.9 0.0 0.0 0.0
II.3.5 6.0 6.0 6.0 6.0 6.0 6.0
II.4.2 10.1 0.0 10.1 0.0 0.0 0.0
II.4.3 4.8 0.0 4.8 0.0 0.0 0.0
II.4.4 7.2 7.2 7.2 7.2 7.2 7.2
II.5.1 11.4 0.0 11.4 0.0 0.0 0.0
II.5.2 38.6 0.0 9.4 0.0 9.4 0.0
II.5.3 38.6 0.0 9.4 0.0 9.4 0.0
III.2.2 20.0 20.0 0.0 0.0 0.0 0.0
III.2.3 23.8 23.8 0.0 0.0 0.0 0.0
III.3.2 4.1 4.1 4.1 4.1 4.1 4.1
III.3.3 14.8 14.8 0.0 0.0 0.0 0.0
III.3.4 2.0 0.0 2.0 0.0 0.0 0.0
III.4.2 11.1 11.1 0.0 0.0 0.0 0.0
III.4.3 33.3 0.0 5.6 0.0 5.6 0.0
III.4.4 12.5 0.0 12.5 0.0 0.0 0.0
III.5.1 8.3 0.0 8.3 0.0 0.0 0.0
III.5.2 22.2 22.2 0.0 0.0 0.0 0.0
III.5.3 33.3 0.0 5.6 0.0 5.6 0.0
III.5.4 22.2 0.0 22.2 0.0 0.0 0.0
IV.4.4 31.3 0.0 0.0 0.0 0.0 0.0
IV.4.5 100.0 20.0100.0 0.0 20.0 0.0
IV.5.1 5.9 0.0 5.9 0.0 0.0 0.0
IV.5.4 5.9 0.0 5.9 0.0 0.0 0.0
IV.5.5 23.1 23.1 0.0 0.0 0.0 0.0
V.2.4 10.7 0.0 10.7 0.0 0.0 0.0
V.2.5 7.3 7.3 7.3 7.3 7.3 7.3
V.3.1 5.9 0.0 5.9 0.0 0.0 0.0
V.3.2 5.7 0.0 5.7 0.0 0.0 0.0
V.4.2 17.9 17.9 0.0 0.0 0.0 0.0
V.5.2 7.7 7.7 0.0 0.0 0.0 0.0
V.5.5 5.6 0.0 5.6 0.0 0.0 0.0
No. of optimal
found 19 12 33 27 34
a
ProblemNumber: the Roman numeral shows the problemset num-
ber, the 2
nd
and 3
rd
numerals show the number of ASMs and SAM
systems, respectively.
Table 4. Comparison of OC + 2OX and 2OX + OC heuristics with best results.
OC + 2OX 2OX + OC Best
Number of times heuristic is best or tied for best 35 31 38
Average percentage below optimal value 0.77 2.64 0.65
Average rank among three results 1.05 1.37 1.00
Worst ratio of solution to optimal value 0.93 0.75 0.93
Number of times heuristic found the optimal solution 33 27 34
the magazines for the problems with ve SAM systems falls
below two missiles per system. Since the average number of
available missiles for each system is low, using this valuable
asset against one ASMmay prevent the heuristics fromusing
it against another one more effectively in a later engagement.
This argument is generally valid for construction algorithms.
We expect that if we had a larger number of missiles per
SAM system, construction algorithms could produce better
results. This is because, as the number of SAMs carried gets
too large, solution of the problem becomes trivial; one sim-
ply allocates as many of the most effective SAMs as possible
to all ASMs. The real challenge is solving the problem with
limited resources, and that is why we create test cases with a
small number of missiles per SAM system.
Improvement heuristics increased the solution quality sig-
nicantly. However, up to this point, we investigated small
test problems to be able to compare the results of heuristics
with the optimal solution. Since we are unable to generate
optimal results for problems larger than ve SAM systems
against ve attacking ASMs due to computational burden
of implicit enumeration, we develop an upper bounding
scheme for large problems to show the quality of the heuris-
tic solutions. Large test problems also enable us to test the
performance of heuristics in terms of elapsed time.
We develop the upper bound (UB) by relaxing constraint
set (2), which reects the restriction on the number of rounds
available for each SAM system. Thus, we assume unlimited
number of rounds available for each SAM system. Objec-
tive function of the relaxed problem is an upper bound for
the original problem. By relaxing constraint set (2), the prob-
lembecomes separable on each ASMi. Sub-problems can be
solved separately and then the upper bound for the objective
function can be calculated using the probability of shooting
down each ASM i. Each sub-problem can be solved using
implicit enumeration.
The upper bounding scheme will produce optimal results
as long as we have sufcient rounds on SAM systems to
schedule the best engagement against each ASM. If there is
shortage on missile availability to schedule the best engage-
ment for each ASM, then the gap between the objective
function value and the upper bound would be large. For exam-
ple, assume that we have a scenario with one SAM system
against two ASMs. Let, d
1
= 2, u
11
= 1, u
21
= 2, p
11
= 0.4,
and p
21
= 0.6. The optimal engagement for this scenario is
one engagement against ASM1 and one engagement against
ASM2 with an optimal objective function value of 0.24. Our
upper bounding scheme produces one engagement against
ASM1 and two engagements against ASM2 with an objec-
tive function value of 0.336. If we assume d
1
= 3 rounds
available on SAM system, then optimal objective function
value and upper bound are both 0.336. This simple example
shows how the optimal objective function value gets closer
to the upper bound as we increase the available rounds on
SAM systems.
For large problems, we again use the best solution obtained
from the two construction algorithms as the starting point of
improvement heuristics. We assume that 10 rounds are avail-
able for each SAM system. Table 5 depicts the results for
those large test problems in terms of computation time. We
use ve sample problems for each ASM and SAM system
combination. Each row in Table 5 shows the average results
of those ve problems. The largest average run time recorded
for heuristics is 1.01 s for the problem with 25 ASMs and
20 SAM systems using OC + 2OX algorithm. Computation
times of the heuristics are less than half a second in 434
runs out of 450 (two construction heuristics, two individual
improvement heuristics and two permutations of improve-
ment heuristics tried for each of the 15 scenarios for ve
different problems). Elapsed times are larger than 1 second
and less than 1.6 s for only four out of 450 runs. The last col-
umn in Table 5 shows the average elapsed time for upper
bound computation. The largest average run time for the
upper bound is 5222.69 s.
Table 6 shows the minimum, average and maximum gap
between the best heuristic solution and the upper bound.
We declare optimality for 22 out of 75 problems as the gap
between the heuristic solution and the upper bound is zero.
The largest average gap is 0.048, while the largest maximum
gap is 0.164. We can say that our heuristics work well for
sample problems with 5 and 10 ASMs. We need further inves-
tigation to determine the source of larger deviations with 15,
20, and 25 ASMs. The source of deviation may be a large
optimality gap for the upper bound or a large optimality gap
for the heuristic solution. We theoretically know that, if we
have unlimited rounds available on SAM systems, the upper
bounding scheme produces optimal results. Hence, we expect
Table 5. Performance of heuristics for large problems in terms of elapsed time.
Number Number of
Elapsed time
a
(sec)
of ASMs SAM systems BEC QUC OC 2OX OC + 2OX 2OX + OC UB
5 10 0.00 0.00 0.00 0.00 0.00 0.00 19.94
15 0.00 0.00 0.00 0.00 0.00 0.01 209.27
20 0.00 0.00 0.00 0.00 0.00 0.01 1005.94
10 10 0.00 0.00 0.00 0.00 0.01 0.01 23.14
15 0.00 0.00 0.01 0.02 0.02 0.03 109.67
20 0.00 0.00 0.01 0.02 0.04 0.03 1070.19
15 10 0.00 0.00 0.02 0.03 0.05 0.05 65.13
15 0.00 0.00 0.05 0.08 0.10 0.11 274.67
20 0.00 0.00 0.11 0.12 0.17 0.16 2290.29
20 10 0.00 0.00 0.08 0.09 0.13 0.15 112.39
15 0.00 0.00 0.08 0.24 0.20 0.25 672.71
20 0.00 0.00 0.24 0.29 0.49 0.50 4352.20
25 10 0.00 0.00 0.14 0.15 0.15 0.18 80.45
15 0.00 0.00 0.13 0.37 0.36 0.44 850.49
20 0.00 0.00 0.44 0.61 1.01 0.97 5222.69
a
Average CPU time for the algorithms on a personal computer with AMD Athlon 2000 + CPU and 256 MB of RAM.
that the optimality gap for the upper bound will get smaller as
we increase the number of rounds available on SAMsystems.
If we are able to reduce the gap between the heuristic solu-
tion and the upper bound by increasing the number of rounds
available on SAM systems, larger gaps in Table 6 can be
attributed to the large optimality gap for the upper bounding
scheme. Therefore, we increase the number of rounds avail-
able on SAM systems to 20 and 30 for sample problems with
15, 20, and 25 ASMs. The results are depicted in Table 7. The
largest average gap is reduced to 0.014 for 20 rounds, and to
0.007 for 30 rounds. Although we cannot claim optimality
Table 6. Average gap between the best heuristic solution and the
upper bound.
Number Number of Minimum Average Maximum
of ASMs SAM systems gap
a
gap
a
gap
a
5 10 0.000 0.000 0.000
15 0.000 0.007 0.027
20 0.000 0.000 0.000
10 10 0.000 0.002 0.012
15 0.000 0.011 0.034
20 0.000 0.004 0.011
15 10 0.000 0.044 0.164
15 0.001 0.042 0.092
20 0.001 0.030 0.052
20 10 0.002 0.029 0.112
15 0.004 0.038 0.074
20 0.001 0.029 0.070
25 10 0.000 0.048 0.150
15 0.000 0.027 0.057
20 0.000 0.021 0.045
a
Gap = UB-best heuristic solution.
for those problems, results warrant that the heuristics work
well for all problem sizes in general.
We focused on the quality of the solutions up to here. Sensi-
tivity of the solutions to the input parameters is another issue
for better understanding of the model and quantication of
the relationships between input parameters and output. We
assumed static single shot kill probabilities for the randomly
generated SAM systems in the preceding computations. We
checked the sensitivity of the solutions to randomly gener-
ated single shot kill probabilities. We used problemI.5.5 from
Table 3 with ve ASMs and ve SAM systems. We solved
the problem 20 times, each time using a randomly generated
single shot kill probability matrix. All heuristic solutions pro-
duced optimal results with different objective function values
and engagement schedules as expected.
Table 7. Average gap for 20 and 30 rounds available on SAM
systems.
Number of Number of
Average gap
a
ASMs SAM systems 20 rounds 30 rounds
15 10 0.000 0.000
15 0.014 0.007
20 0.001 0.000
20 10 0.002 0.001
15 0.013 0.007
20 0.012 0.007
25 10 0.009 0.001
15 0.011 0.007
20 0.006 0.003
a
Gap = UB-best heuristic solution.
Table 8. Solutions of the sample case problem and the revised problems.
Obj. func.
Engagements
Change
a
value ASM1 ASM2 ASM3 ASM4 ASM5
None 0.59 SAM2 SAM2 SAM2(3)
b
SAM2(2) SAM2
SAM5 SAM5
d
2
= 6 0.56 SAM2 SAM2 SAM1 SAM2(2) SAM2
SAM2 SAM5 SAM5
SAM4
d
2
SAM2 SAM5
SAM4
d
2
SAM2 SAM5
SAM4
d
2
SAM4 SAM5
p
12
= 0.75 0.49 SAM2 SAM2 SAM2(3) SAM2(2) SAM2
SAM5 SAM5
p
12
= 0.35 0.28 SAM1 SAM2 SAM2(3) SAM2(2) SAM2(2)
SAM5
p
42
= 0.60 0.56 SAM2 SAM2 SAM2(3) SAM2 SAM2(2)
SAM5(2)
p
42
= 0.30 0.53 SAM2 SAM2 SAM2(3) SAM5(3) SAM2(2)
a
Shows the difference from the original parameters of the sample case.
b
Figure within parenthesis shows the number of engagements if it is larger than one.
We present a sample case to show the effect of input para-
meters. We assume a TG with three ships. Ship1 has self and
area defense SAM systems SAM1 and SAM2. Ship2 simi-
larly has self and area defense systems SAM3 and SAM4.
Ship 3 has only a self defense system SAM5. ASM1 and
ASM3 target Ship1, and ASM2, ASM4, and ASM5 target
Ship3. Note that Ship2 is not a target of any ASM. Thus,
SAM3 cannot engage any of the ASMs. Let the number of
available rounds on SAM systems be (d
j
) = (8, 8, 8, 16, 8).
The upper bounds on the number of engagements of SAM
systems against ASMs within the engageability interval and
the single shot kill probabilities are:
(u
ij
) =
_
_
_
_
_
_
1 2 0 0 0
0 1 0 0 1
2 3 0 1 0
0 2 0 1 3
0 2 0 1 2
_
_
_
_
_
_
(p
ij
) =
_
_
_
_
_
_
0.40 0.90 0.00 0.80 0.00
0.00 0.75 0.00 0.70 0.65
0.30 0.65 0.00 0.60 0.00
0.00 0.75 0.00 0.70 0.45
0.00 0.90 0.00 0.80 0.50
_
_
_
_
_
_
.
We solved this sample problem and additional eight prob-
lems generated by changing the available SAM rounds and
the single shot kill probabilities one at a time. We report the
revised input parameter values, the objective function values
and the engagements produced in Table 8.
The objective function value of the original sample prob-
lem is 0.59 and the resulting engagements are depicted in the
rst row of Table 8. All available rounds of SAM2, which is
the most effective area air defense system, are used. When
we reduce the available rounds of SAM2 to d
2
= 6, SAM1
and SAM4 replace SAM2 against ASM3. Note that ASM3
attacks Ship1 that is equipped with SAM2, and the other area
air defense system of the TG, which is SAM4 on board of
Ship2, replaces SAM2 for engaging ASM3. When we further
reduce the number of available rounds of SAM2 to 4, 2, and 0,
self-defense SAM systems SAM1 and SAM5 replace SAM2
at the cost of reduced probability of no leaker. When we
reduce p
12
to 0.75, we get the same engagement plan as in the
original sample case. Although p
14
= 0.80 is larger thanp
12
,
the model cannot use it because of the engageability inter-
val restrictions (i.e., u
14
= 0). When we reduce p
12
to 0.35,
SAM1 replaces SAM2. When we reduce p
42
to 0.60, SAM5
replaces one of the SAM2 engagements. All of the engage-
ments of SAM2 are replaced by SAM5 when p
42
= 0.30.
Although p
44
is larger than p
42
, the upper bound on the num-
ber of engagements of SAM4 against ASM4 prevents SAM4
fromreplacing SAM2. We conclude that changes in the input
parameters produce reasonable results. Experiments carried
out on the sample case show the importance of cohesion
within a TG and convey the additional capability provided
by a fully coordinated air defense of a TG.
5. EXTENSIONS
5.1. Alternative Objective Functions
Our formulation maximizes probability of no leaker for
all the ships in a TG, treating them as equally important. If
the defense commander places higher priority on defense of
the HVUs than of the escort ships, he may prefer his area
defense systems to be used for protecting the HVUs rather
than the escorts. An extreme case may be to maximize the
probability of shooting down only those ASMs targeting the
HVUs, ignoring the rest. If one does not want area defense
SAM system j to engage ASM i targeting an escort ship,
this can easily be implemented in our formulation by exclud-
ing respective (i, j) pairs from the valid engagements set V,
thereby setting the respective x
ijk
variables to zero. Hence
this is a special case of our formulation.
In the above treatment, only the HVUs are protected and
the escort ships are completely ignored, which means that
they can only rely on their own self-defense capabilities. If
one still wants to provide some area defense for the escorts,
but with lower priority compared to the HVUs, this can be
achieved by the following alternative objective function.
sT G
v
s
iN
s
_
_
_1
kK
jM|(i,j)V
(1 p
ijk
)
x
ijk
_
_
_ (1
)
Here, TG is the set of ships in the TG, v
s
is the value or
importance weight assigned to ship s, and N
s
is the set of
ASMs targeting ship s such that
sT G
N
s
= N. We expect
the v
s
parameter to take a higher value for a HVU than it
does for an escort ship. This objective can be interpreted as
maximizing the total expected value of TG.
The probability of no leaker may be a very small gure,
when there is a large number of attacking ASMs. In this case,
we may consider minimizing the expected number of leak-
ers or maximizing the expected number of ASMs shot down.
This leads to the simple linear objective function below.
Max
iN
kK
jM|(i,j)V
p
ijk
x
ijk
(1
)
However, we lose control of which ASM to engage with
this objective. Some of the attacking ASMs may reach their
targets with no scheduled SAM engagement. This may make
the probability of no leaker zero and HVU(s) may be the tar-
get of those leaking ASMs. If defending against all the ASMs
is desirable, we can maximize the minimum probability of
shooting down the attacking ASMs to have an engagement
plan that leaves none of the ships in TG undefended against
any ASM if possible.
Other objectives could be maximizing the survival prob-
ability by taking into account the single shot kill probabil-
ities of the ASMs against the ships in TG, maximizing the
total expected value of surviving ships, and minimizing the
damage sustained, which would require damage assessment.
5.2. Impact Point Predictability
Assumption 6 states that the target of each ASMcan be pre-
dicted. If a radar can detect an ASM it can also determine its
speed and direction. Hence, target prediction is possible at the
time of detection as long as the ASMs ight path is a straight
line and the ships are not too close to each other. However,
when an ASM follows pre-programmed way points, there is
no way of predicting its eventual target before it is in the nal
leg of the pre-programmed ight path. In this case, there is
nothing to do or no problem to solve at the time of initial
detection.
If impact point prediction is impossible because of the long
engagement ranges, our approach can still be used. When the
engagement range is very long, we can reasonably assume
that the distances between ships are much smaller (they are
close together) compared with the distances of the ASMs
from the TG as a whole. Then, all distances (and respective
times) can be computed fromthe ASMto the center of the TG.
In this case, one can consider using only the long range area
defense systems, i.e., the valid engagements set V includes
only those (i, j) pairs such that ASM i is within the range
of SAM j. Our formulation and heuristic algorithms remain
unchanged. Naturally, this may result in some unnecessary
missile allocations and waste of defense resources, but this
is the price of uncertainty since the exact target of an ASM
is unknown at the time of detection and allocation.
5.3. SLS Engagement Policy
As an alternative to the SLS policy (assumption 8) one
might want to use shoot-shoot-look (SSL), shoot-shoot-
shoot-look or other variations. Here, we describe how our
approach can be adapted for the SSL policy without loss of
generality. If the defense commander wants to re two rounds
of SAM j at ASM i before he stops to look, the engage-
ment duration
ijk
should include two constant setup times
and one variable ight time to interception. The overall miss
probability of this double engagement is (1 p
ijk
)
2
. In this
case, our binary decision variable x
ijk
indicates whether or
not a double engagement starts in time slot k. The objective
function and constraint set (2) need to be changed as follows.
Max
iN
_
_
_1
kK
jM|(i,j)V
(1 p
ijk
)
2x
ijk
_
_
_ (1
)
subject to
kK
iN|(i,j)V
x
ijk
d
j
/2 for all j M (2
)
When two rounds are to be red at the same ASM from
different SAM systems, determination of the engagement
duration before looking is not as straightforward as above.
Instead, one can simply duplicate the incoming ASMs to
use our formulation and heuristic algorithms as they are.
Assuming there are two ASMs instead of one, our approach
can allocate two rounds of different SAMs to these ASMs,
which in fact are the same ASM. This treatment also allows
using different engagement policies against different ASMs.
The commander may use the SLS tactic against some of the
ASMs and SSL against others.
5.4. Multiple SAMs in Flight
Assumption 11 states that there can be multiple SAMs in
ight launched from the same SAM system to engage dif-
ferent ASMs. Although some modern systems (e.g., Aster
missile system where each missile has its own radar and
Evolved Sea Sparrow Missile systemESSM) are capable of
illuminating multiple ASMs simultaneously, this assumption
may be restrictive for systems that must illuminate the target
ASM by a traditional single-beam radar. We can relax this
assumption by dening a new set L
jk
symmetrical to set J
ik
.
L
jk
= {(i, ) : (i, j) V, S
ij
, and [
k
,
k
+ ]
[
+
ij
]}
L
jk
is the set of combinations (i, ) such that time slot k for
SAM system j is blocked (to prevent the SAM system from
having multiple missiles in ight during slot k). To impose
the restriction, we need the new constraint set (6).
(i,)L
jk
x
ij
1 for all j M and k K (6)
6. CONCLUSION
In this study, we introduced a new formulation for the air
defense problem and developed an efcient solution method
for it. The mathematical programming model that was devel-
oped was not explicitly used to solve MAP. Although mathe-
matical programming models do guarantee an optimal solu-
tion (without loss of generality), they usually take much more
than a few seconds in which we have to solve MAP for its
repetitive application in scenario analysis.
Our solution approach for MAP uses construction and
improvement heuristics. We have developed two greedy con-
struction algorithms for MAP. The rst of those algorithms,
BEC heuristic, allocates SAM systems to incoming ASMs
according to a measure called the engagement potential. In
the QUC algorithm, we aim to engage each threat ASM at
least once. Thus, we give precedence to the ASM that has the
lowest number of SAMs allocated.
We developed two improvement heuristics, OC and 2OX.
Our purpose in the OC algorithm is to nd the engagements
that would increase the objective function value by changing
the target ASM of an engagement and simultaneously con-
sidering the enhancement of defense by increasing the total
number of SAM missiles launched against target ASMs. Our
purpose in the 2OX algorithm is to nd the engagement pairs
that would increase the objective function value by exchang-
ing the target ASMs of the SAMs. With each exchange, we
again try to increase the number of engagements against the
ASMs under consideration.
We tested our solution approach against the optimumusing
125 small sample problems. The procedure yielded highly
successful results. We attained 121 optimal solutions out of
the 125 test problems. We generated 12 large test problems
to be able to test the performance of our heuristics in terms
of computation time. The largest run time recorded was 1.17
seconds. Computation times of the improvement heuristics
were less than half a second in 44 out of 48 runs made for 12
large problems.
As the computation times of our construction and improve-
ment algorithms are short, we suggest running both construc-
tion algorithms since they do not dominate each other. Then,
one can try to improve the better of the two construction
solutions by running different combinations of improvement
algorithms. If one combination is to be selected, we suggest
using OC+2OXas it gives better results according to Table 4
and the Wilcoxon test.
MAP can be used in the decision-making process for the
procurement of new air defense ships, and in evaluating the
capabilities of ships in inventory and the effectiveness of
present tactics. The coverage a ship provides for an HVU
or other ships can also be estimated by using MAP through
repetitive scenario analysis. When there is no or limited infor-
mation on the attack direction, various representative sce-
narios can be developed and solved using MAP. The results
can be aggregated to estimate the coverage provided from
one sector to another on the sea surface. The ships can then
be located in these sectors so as to maximize the cover-
age provided for them [20]. According to the computational
studies carried out, MAP also has the potential for on-line
and real-time use in realistic tactical situations as part of the
threat evaluation and weapon allocation module of a TG air
defense system.
The proposed solution approach can be enhanced along
several directions. A comparison of the proposed method-
ology with the existing air defense policies and procedures
may reveal more insight into the utility of the approach.
In addition to computational time requirement, the solution
quality of MAP for large problems needs to be investigated
provided that an exact solution procedure for large prob-
lems is developed. Solution procedures may be developed
to include other engagement policies. We solve MAP in
a static environment, assuming simultaneous attack. How-
ever, both simultaneous and sequential attack waves may
occur in the dynamic environment of a real combat situa-
tion. High resolution simulation models can be developed to
investigate the best use of MAP solutions in such a dynamic
environment.
APPENDIX A: OC ALGORITHM
STEP 0: Select an initial feasible engagement list, E.
E = {(i
1
, j
1
), (i
2
, j
2
), . . . , (i
l
, j
l
)|k L = {1, . . . , l}, (i
k
, j
k
) V,
t
(i
k
,j
k
)
< t
(i
k
,j
k
)
k L and k
L i
k
= i
k
and |k| <|k
|}
where i N, j M, t
(i
k
,j
k
)
is the time of the engagement (i
k
, j
k
)
and V is the set of valid combinations of ASM and SAM sys-
tems, i.e., (i, j) V if SAM system j can engage ASM i. Let
the corresponding objective function value be Z(E).
STEP 1: Set k = 0, E
= E, where E
is the best engagement schedule

that has been found so far. Set the logical variables add1 and
add2 to false.
STEP 2: Set k = k +1. i.e., take the next engagement in the engagement
list of E.
STEP 3: Check the possibility of the change of target ASM for the
engagement (i
k
, j
k
) in the engagement list for all possible tar-
gets except i
k
, i.e., set F = N\{i
k
} = {f
1
, f
2
, . . . , f
n1
}. Let
h H = {1, 2, . . . , n 1} and set h = 1.
STEP 4: If h > 1, then set h = h+1. If (f
h
, j
k
) V, then go to Step 5 to
nd a SAM missile for ASM i
k
to enhance the defense against
it, otherwise go to Step 11.
STEP 5: If there is at least one SAM system that can engage ASM i
k
and
has missiles left, then check the possibility of enhancement using
all possible SAMs, i.e., set G = M\{j
k
} = {g
1
, g
2
, . . . , g
m1
}.
Let t T = {1, 2, . . . , m 1} and set t = 1. Set the logical
variable change to false.
STEP 6: If t > 1, then set t = t + 1.
STEP 7: If d
gt
> 0 and (i
k
, g
t
) V, then go to Step 8; otherwise, go to
Step 9. Note that d
gt
is the number of available rounds on SAM
system g
t
.
STEP 8: Dene a new engagement list,

E = {E\{(i
k
, j
k
)}}
{(f
h
, j
k
), (i
k
, g
t
)}. Note that, (i
k
, g
t
) will be the last engagement
of the engagement list. Check the feasibility of new engagement
list

E, and calculate the objective function value, Z(

E). If

E is
feasible and Z(

E) > Z(E
), then change the engagement list,

E
=

E, update the objective function value Z(E
) = Z(

E).
Set g
= g
t
, the variables add1 and change to true, and
add2 to false.
STEP 9: If t = m 1, then go to Step 10. Else, go to Step 6.
STEP 10: If the variable change has value false, then dene a new
engagement list,

E = {E\{(i
k
, j
k
)}} {(f
h
, j
k
)}. Check the fea-
sibility of new engagement list

E, and calculate the objective
function value, Z(

E). If

E is feasible and Z(

E) > Z(E
), then
change the engagement list E
=

E, update the objective func-
tion value Z(E
) = Z(

E). Set the variable add1 and add2
to false.
STEP 11: If h = n 1, then go to Step 12. Else, go to Step 4.
STEP 12: Consider changing the defending SAM for the engagement
(i
k
, j
k
). Set t = 1.
STEP 14: If d
gt
> 0 and (i
k
, g
t
Step 16.

E = {E\{(i
k
, j
k
)}} {(i
k
, g
t
)}.
Note that we change (i
k
, j
k
) to (i
k
, g
t
) in the engagement list
E. Check the feasibility of new engagement list

E, and cal-
culate the objective function value, Z(

E). If

E is feasible and
Z(

E) > Z(E
), then change the engagement list, E
=

E,
update the objective functionvalue Z(E
) = Z(

E). Set g
= g
t
,
g
+
= j
k
, the variables add1 to false and add2 to true.
STEP 16: If t = m 1, then go to Step 17. Else, go to Step 13.
STEP 17: If k = l, then go to Step 18. Else, go to Step 2.
STEP 18: If Z(E) = Z(E
), then stop.
Otherwise, set E = E
, Z(E) = Z(E
), if variable add1
has value true, then set l = l + 1, d
g
= d
g
1, if variable
add2 has value true, then set d
g
= d
g
1, d
g
+ = d
g
+ +1
and go to Step 1.
APPENDIX B: 2OX ALGORITHM
STEP 0: Select an initial feasible engagement list, E.
E = {(i
1
, j
1
), (i
2
, j
2
), . . . , (i
l
, j
l
)|k L = {1, . . . , l}, (i
k
, j
k
) V,
t
(i
k
,j
k
)
< t
(i
k
,j
k
)
k L and k
L i
k
= i
k
and |k| < |k
|}
where i N, j M, t
(i
k
,j
k
)
is the time of the engagement (i
k
, j
k
)
and V is the set of valid combinations of ASM and SAM sys-
tems, i.e., (i, j) V if SAM system j can engage ASM i. Let
the corresponding objective function value of the engagement
list E be Z(E). Set E
= E, where E
is the best engagement

schedule that has been found so far.
STEP 1: Set k = 1 and h = 1. Set the logical variables add1 and add2
to false. Those logical variables are used to control whether
the best engagement schedule that may be found has additional
launhes against ASMs exchanged or not.
STEP 2: Check the possibility of exchange of SAM allocation of the
engagements k and k +h in the engagement list. If {(i
k
, j
k+h
) /
V or (i
k+h
, j
k
) / V} go to Step 18.

E = {. . . , (i
k
, j
k+h
), . . . ,
(i
k+h
, j
k
), . . .}. Check the feasibility of new engagement list
E, and calculate the objective function value, Z(

E). If

E is
infeasible, then go to Step 18.
STEP 4: If Z(

E) > Z(E
), then reset the best engagement list, E
=

E,
update the objective function value, and set variables add1 and
add2 to false.
STEP 5: Check for additional assignment against ASM i
k
, i.e., set G =
M = {g
1
, g
2
, . . . , g
m
}. Let t T = {1, 2, . . . , m} and set
t = 1. Note that we do not exclude SAM j
k
from consider-
ation, since change in a previous engagement may enable us
to launch the same engagement (i
k
, j
k
) as the last engagement
against ASM i
k
. Set the logical variable change to false.
The variable change is used to control whether ASM i
k
has
additional launches against itself.
STEP 7: If d
gt
> 0 and (i
k
, g
t
Step 10. Note that d
gt
is the number of available rounds on SAM
system g
t
.
STEP 8: Dene a new engagement list, E = E {(i
k
, g
t
)}. Note that,
(i
k
, g
t
) will be the last engagement of the engagement list. Check
the feasibility of newengagement list E, and calculate the objec-
tive function value, Z(E). If E is feasible, then set the variable
change to true, g
change
= g
t
and go to Step 9; otherwise, go
to Step 10.
STEP 9: If Z(E) > Z(E
), then reset the best engagement list, E
= E,
update the objective function value Z(E
) = Z(E). Set g1
=
g
t
, the variable add1 to true and add2 to false.
STEP 10: If t = m, then go to Step 11. Else, go to Step 6.
STEP 11: Check for additional assignment against ASM i
k+h
. Set t = 1.
STEP 13: If the variable change has value true, go to Step 14;
otherwise, go to Step 16.
STEP 14: If g
t
= g
change
and d
gt
> 1, then go to Step 15, else if
g
t
= g
change
and d
gt
> 0, then go to Step 15; otherwise, go
to Step 17.
STEP 15: If (i
k+h
, g
t
) V then dene a new engagement list, E =
E {(i
k+h
, g
t
)}. Check the feasibility of new engagement list
E, and calculate the objective function value, Z(E). If E is
feasible and Z(E) > Z(E

E
= E, update the objective function value Z(E
) = Z(E),
set g2
= g
t
, the variables add2 true; otherwise, go to
Step 17.
STEP 16: If d
gt
> 0, then dene a new engagement list, E = E
{(i
k+h
, g
t
)}. Check the feasibility of new engagement list E,
and calculate the objective function value, Z(E). If E is fea-
sible and Z(E) > Z(E

E
= E, update the objective function value Z(E
) = Z(E).
Set g2
= g
t
, the variable add2 to true and add1 to
false.
STEP 17: If t = m, then go to Step 18.
Else, go to Step 12.
STEP 18: If k + 1 = l, then go to Step 19.
Else,
if k + h = l then set t = 1, k = k + 1 and go to Step 2.
if k + h < l then set h = h + 1 and go to Step 2.
STEP 19: If Z(E) = Z(E
), then go to step 20.

Otherwise, set E = E
, Z(E) = Z(E
), if variable add1
has value true, then set l = l +1, d
g1
= d
g1
1, if variable
add2 has value true, then set l = l +1, d
g2
= d
g2
1, and
go back to Step 1.
STEP 20: For each possible ASMpair, try changing all the engagements of
those ASMs. If there is an improvement, update the engagement
list E, and go back to Step1; otherwise, stop.
ACKNOWLEDGMENT
The authors thank the associate editor and the anonymous reviewers for
their insightful comments and valuable suggestions on improving this article.
REFERENCES
[1] R. Almeida, D.P. Gaver, and P.A. Jacobs, Simple prob-
ability models for assessing the value of information in
defense against missile attack, Nav Res Logist 42 (1995),
535547.
[2] D.K. Al-Mutairi, T.M. Al-Khamis, and M.S. Abdel-Hamed, A
Bayesian analysis of layered defense systems, J Appl Probab
34 (1997), 449457.
[3] D.K. Al-Mutairi and R.M. Soland, Attrition through a par-
tially coordinated area defense, Nav Res Logist 52 (2005),
7481.
[4] M.J. Armstrong, Effects of lethality on naval combat models,
Nav Res Logist 51 (2004), 2843.
[5] G.C. Beare, Linear programming in air defense modelling,
J Oper Res Soc 38 (1987), 899905.
[6] D.P. Bertsekas, M.L. Homer, D.A. Logan, S.D. Patek, and N.S.
Sandell, Missile defense and intercept allocation by neuro-
dynamic programming, IEEE Trans Syst, Man Cybernet 30
(2000), 4251.
[7] A.F. Bloemen and R.R. Witberg, Anti-air warfare research for
naval forces, Nav Forces 21 (2000), 2023.
[8] S.A. Burr, J.E. Falk, and A.F. Karr, Integer prim-read solu-
tions to a class of target defense problems, Oper Res 33 (1985),
726745.
[9] W.S. Carus, Cruise missile proliferation in the 1990s, The
Washington Papers/159, Praeger Publishers, Westport, CT,
1992, pp. 101.
[10] S.J. Chapman, Ship Air Defense Model Users Guide: Single
Ship Version, British Aerospace Australia Ltd. No: AD51-
001 (1999).
[11] A.R. Eckler and S.A. Burr, Mathematical models of target
coverage and missile allocation, Military Operations Research
Society, Alexandria, VA, 1972.
[12] FAS (Federation of American Scientist), Air Defense
Effectiveness, 2009. http://fas.org/spp/aircraft/part03.htm, last
accessed on 9 June 2009.
[13] GlobalSecurity.org, US Missiles, 2009, http://www.
globalsecurity.org/military/systems/munitions/missile.htm,
last accessed on 9 June 2009.
[14] B.L. Golden and W.R. Stewart, Empirical Analysis of Heuris-
tics, The Traveling Salesman Problem, E.L. Lawler, J.K.
Lenstra, A.H.G. Rinnooy Kan, and D.B. Shmoys (Editors),
Wiley, 1985, pp. 207249.
[15] J.D. Grifths, S.A. Marshall, T.G. Weale, and J.E. Williams, A
simple stochastic model of air-naval combat, J Oper Res Soc
42 (1991), 289300.
[16] H.C. Hoty, A simple ballistic-missile-defense model to help
decision makers, Interfaces 15 (1985), 5462.
[17] N.K. Jaiswal, Military operations research: Quantitative deci-
sion making, Kluwer Academic Publishers, Norwell, Massa-
chusetts, 1997.
[18] O. Kwon, D. Kang, K. Lee, and S. Park, Lagrangean relaxation
approach to the targeting problem, Nav Res Logist 46 (1999),
640653.
[19] O. Karasakal, Air defense missile-target allocation mod-
els for a naval task group, Comput Oper Res 35 (2008),
17591770.
[20] O. Karasakal, L. Kandiller, and N.E. zdemirel, Abranch and
bound algorithm for sector allocation problem of a naval task
group, Industrial Engineering Department Technical Report
No: 09-08, Middle East Technical University, Ankara, Turkey,
2009.
[21] O. Karasakal, Optimal air defense strategies for a naval task
group, Ph.D. Dissertation, Industrial Engineering Department,
Middle East Technical University, Ankara, Turkey, 2004.
[22] A. Krause, J. Leskovec, C. Guestrin, J. VanBriesen, and C.
Faloutsos, Efcient sensor placement optimization for secur-
ing large water distribution networks, J Water Resour Plann
Manag 134 (2008), 516526.
[23] P. Krokhmal, R. Murphey, P. Pardalos, S. Uryasev, and G.
Zrazhevsky, Robust decision making: Addressing uncertain-
ties in distributions, S. Butenko, et al. (Editors), Coopera-
tive control: Models applications and Algorithms, Kluwer
Academic Publishers, 2003, pp. 165185.
[24] P. Krokhmal, R. Murphey, P. Pardalos, and S. Uryasev, Use of
conditional value at-risk in stochastic programs with poorly
dened distributions, S. Butenko, et al. (Editors), Recent
developments in cooperative control and optimization, Kluwer
Academic Publishers, 2004, pp. 225243.
[25] S.P. Lloyd and H.S. Witsenhausen, Weapons allocation is
NP-complete, Proceedings of the 1986 Summer Computer
Simulation Conference, 1986, pp. 10541058.
[26] T.W. Lucas and J.E. McGunnigle, When is model complex-
ity too much? Illustrating the benets of simple models with
Hughes salvo equations, Nav Res Logist 50 (2003), 197217.
[27] R. Macfadzean, Surface-based air defense system analysis,
Artech House, Norwood, MA, 1992, pp. 336339.
[28] P.F.M. Martin, V. Dongen, and J. Kos, The analysis of ship air
defense: The simulation model SEAROADS, Nav Res Logist
42 (1995), 291309.
[29] S. Matlin, A review of the literature on the missile-allocation
problem, Oper Res 18 (1970), 334373.
[30] Naval Studies Board, Technology for the United States Navy
and Marine Corps, 20002035 Becoming a 21
st
Century Force
Vol.9: Modeling and Simulation, National Research Council,
Washington, D.C., 1997. <http://www.nap.edu/html/tech_21st
/msindex.htm, Last accessed on 9 June 2009>.
[31] G.L. Nemhauser, L.A. Wolsey, and M.L. Fisher, An analysis
of approximations for maximizing submodular set functions-I,
Math Programming 14 (1978), 265294.
[32] B.U. Nguyen, A model of cohesion for air defense,
DOR(CAM) Research Note RN9601, Department of National
Defense, Ottawa, Canada, 1996.
[33] B.U. Nguyen, P.A. Smith, and D. Nguyen, An engagement
model to optimize defense against a multiple attack assuming
perfect kill assessment, Nav Res Logist 44 (1997), 687697.
[34] H.D. Sherali, Y. Lee, and D.D. Boyer, Scheduling target illu-
minators in naval battle-group anti-air warfare, Nav Res Logist
42 (1995), 737755.
[35] P.A. Smith, L. The, and B.U. Nguyen, The joint anti-ship mis-
sile intercept evaluation (JASMINE) model: Model descrip-
tion, ORA Project Research Report No. 9521, Department of
National Defense, Ottawa, Canada, 1995.
[36] R.M. Soland, Optimal terminal defense tactics when several
sequential engagements are possible, Oper Res 35 (1987),
537542.
[37] E. Wacholder, Aneural network-based optimization algorithm
for the static weapon-target assignment problem, ORSA J
Comput 1 (1989), 232246.

Anti-Ship Missile Defense For A Naval Task Group

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Anti-Ship Missile Defense For A Naval Task Group

Uploaded by

Copyright:

Available Formats

Anti-Ship Missile Defense for a Naval Task Group

is the best engagement schedule

), then change the engagement list,

), then change the engagement list, E

is the best engagement

E, and calculate the objective function value, Z(

), then reset the best engagement list, E

), then reset the best engagement list, E

), then change the engagement list,

= E, update the objective function value Z(E

), then change the engagement list,

= E, update the objective function value Z(E

), then go to step 20.

You might also like