Computers and Operations Research 105 (2019) 226–236

Contents lists available at ScienceDirect

Computers and Operations Research

journal homepage: www.elsevier.com/locate/cor

Survey in Operations Research and Management Science

The Weapon-Target Assignment Problem

Alexander Kline∗, Darryl Ahner, Raymond Hill
Department of Operational Sciences, Air Force Institute of Technology, WPAFB, OH 45433, United States

a r t i c l e i n f o a b s t r a c t

Article history: Research addressing the Weapon Target Assignment (WTA) Problem, the problem of assigning weapons
Received 4 April 2018 to targets while considering their effective probability of kill, began with Manne’s seminal work in 1958.
Revised 18 September 2018
In the years following, improved modeling and solution techniques have been developed, along with im-
Accepted 22 October 2018
provements in computing power, which have enabled researchers to consider more complex variants of
Available online 30 October 2018
the problem, to include models with fewer assumptions and models in which time is a parameter. Herein,
Keywords: we review the various model formulations, exact algorithms, and heuristic algorithms for the static and
Weapon Target Assignment Problem dynamic WTA. We place the formulations into a comparable form and use this form to provide insight
Static WTA into the evolution of the defense-related WTA problem. The solution methods are comparatively analyzed
Dynamic WTA and an analysis of the influence of past work is conducted. More recent developments are introduced and
discussed.
Published by Elsevier Ltd.

1. Introduction able for a single exchange. The solution to the SWTA informs the
defense on how many of each weapon type to shoot at each tar-
Projectile weapons have been a consistent threat of hostilities get. In the SWTA, no subsequent engagements are considered since
throughout history. Military advantage has always been aided by time is not a dimension considered in the problem.
the capacity to inflict damage from a distance. In the 20th century, By contrast, the DWTA includes time as a dimension. Variants
missile technology advanced to the point that an adversary had of the DWTA include the two stage DWTA and the shoot-look-
the potential to attack a protected asset from great distances. To shoot DWTA. The two stage DWTA replicates the SWTA in its first
neutralize this stand off threat, the concept of air defense evolved. stage, but includes a second stage wherein a number of targets of
However, as the ability to reduce a missile threat increased, so too various types are known only to a probability distribution. In this
did the quantity and quality of missiles available, and research into variant, the solution to the DWTA informs the defense on how to
the effective allocation of air defense resources emerged. allocate the weapons in the first stage and how many to reserve
Originally introduced into the field of operations research by for the second stage in order to minimize the probability of de-
Manne (1958), the Weapon Target Assignment (WTA) Problem, or struction. The shoot-look-shoot variant also replicates the SWTA,
Missile Allocation Problem (MAP) as it is sometimes known, seeks however it enables the defense to observe which targets may have
to assign available interceptors to incoming missiles so as to min- survived the engagement (leakers) and allows for a subsequent en-
imize the probability of a missile destroying a protected asset. gagement opportunity. The solution to this variant similarly in-
While much of the literature on the WTA focuses on the defen- forms the defense on how to allocate the weapons and how many
sive perspective, some have considered the offensive perspective weapons to reserve to reengage any leakers.
(Sikanen, 2008), wherein the objective is to maximize the proba- The WTA has been solved to optimality with exact algorithms.
bility of destroying enemy protected assets. However, as Lloyd and Witsenhausen (1986) showed that the WTA
There are two distinct categories of the WTA: the Static WTA is NP-Complete, the majority of solution techniques seek to find
(SWTA) and the Dynamic WTA (DWTA). Originally modeled by near optimal solutions in real-time, or “fast enough to provide an
Manne (1958), the SWTA defines a scenario wherein a known engagement solution before the oncoming targets reached their
number of incoming missiles (targets) are observed and a finite goals” (Leboucher et al., 2013). These real-time solution techniques
number of interceptors (weapons), with known probabilities of are products of heuristic algorithms or are solved using exact algo-
successfully destroying the targets (probabilities of kill), are avail- rithms applied to transformations of the formulation.
The rest of this paper proceeds as follows. In Section 2, we
review the various formulations for both the SWTA and DWTA.
∗
Corresponding author. We examine the basic formulations of each and explore the trans-
E-mail address: akline@afit.edu (A. Kline).

https://doi.org/10.1016/j.cor.2018.10.015
0305-0548/Published by Elsevier Ltd.
 A. Kline, D. Ahner and R. Hill / Computers and Operations Research 105 (2019) 226–236 227

formations which have been implemented. We also review novel However, it is common to write the formulation in terms of the
formulations which have sought to model and solve the problem probability of survival qi j = 1 − pi j
in unique settings. In Section 3, we review the exact algorithms

n 
m
x
that have been used to solve the SW TA and DW TA. Some of these S1 min Vj qi ji j
algorithms provide optimal solutions to the original formulations j=1 i=1
whereas others refer to the transformed formulations identified in

n
Section 2. In Section 4, we review the heuristic and metaheuris- s.t. xi j ≤ wi , for i = 1, . . . , m
tic solution techniques for the SWTA and DWTA. In Section 5, we j=1
discuss the state of the WTA and present a metric with which we
focused this examination of the literature. xi j ∈ Z+ , for i = 1, . . . , m, j = 1, . . . , n .
The nonlinear objective function in S1 seeks those assignments
to minimize the expected value of survival. The assignments are
2. Formulations
integer and the total number of weapon i cannot exceed the num-
ber of weapons on hand, wi . This formulation is used frequently,
There have been many different formulations of the WTA.
(e.g., (Ahuja et al., 2007),(Lemus and David, 1963),(Lee et al.,
Early literature sought to transform the nonlinear formulation from
2002a),(denBroeder et al., 1959)) and is often the initial formula-
Manne (1958) due to the computational limitations with nonlinear
tion used when implementing a transformation.
programming. As computational power increased, transformations
A simpler version of S1 is given by denBroeder et al. (1959),
which were better suited to global optimization tools emerged.
who assumes that all weapons have the same probability of kill
Burr et al. (1985) introduced the DWTA which captured the value
for target j, pi j = p j ∀i = 1, . . . , m. His formulation differs from S1
of subsequent engagements. Similar to the SWTA, variations to the
in the objective, which is
original DWTA occur throughout the literature.
Herein, we examine some of the formulations for both the 
n
x
S2 min Vjq j j
SW TA and DW TA. For purposes of clarity in both formulation
j=1
and in presentation, we map the formulations presented by
their authors into the terms of the formulation developed by This formulation simplifies S1 and is easily optimized by a greedy
Manne (1958). Namely, variables that are shared between multiple assignment technique. However, its assumption of homogeneity
formulations are defined as follows: greatly reduces the applicability of the formulation.
Kwon et al. (1999) utilize a similar model to S1 but reformulate
pi j : the probability weaponi destroys target j the problem into an integer program with a linear objective func-
qi j : the probability weapon i fails to destroys target j tion and nonlinear constraints. They use a negative cost parameter,
V j : the destructive value of target j cij , for assigning weapon i to target j which they seek to minimize
as follows:
xi j : the number of weapons of type i assigned to target j 
min ci j xi j
K : the number of protected assets
(i, j )∈A
ak : the value of asset k 
s.t. xi j ≤wi for i = 1, . . . , m
n : the number of targets
{ j=1,...,n|(i, j )∈A}
m : the number of weapon types   xi j
1− 1 − pi j ≥d j for j = 1, . . . , n
wi : the number of weapons of type i
{i=1,...,m|(i, j )∈A}
ci j : a cost parameter for assigning a weapon of type i to target j
∀(i, j ) ∈ F,
xi j ≤ui j
F : the set of feasible assignments
xi j ≥0 ∀(i, j ) ∈ F
γ jk : the probability target j destroys asset k
where dj is the minimum desired probability of kill for target
s j : the maximum number of weapons that can be assigned j, uij is an upper bound on the number of weapons i that can
to target j be assigned to target j, and F is the set of all feasible assign-
t : the number of stages ments. Kwon et al. (1999) then multiply a large number θ to a
logarithmic transformation of the nonlinear constraint and round
down to the largest integer contained in order to generate the fol-
2.1. SWTA formulations lowing linear approximation, where ai j = −θ ln(1 − pi j ) > 0 and
b j = −θ ln(1 − d j ) > 0
The original formulation as defined by Manne (1958) considers 
a scenario where a defender has wi of i = 1, . . . , m weapon types S3 min ci j xi j
with which to defend against j = 1, . . . , n targets. Each weapon (i, j )∈A

type i has a probability pij of killing target j and each target j has a s.t. xi j ≤wi for i = 1, . . . , m
destructive value Vj . With decision variables xij indicating the num- { j=1,...,n|(i, j )∈A}
ber of weapons of type i to assign to target j, the SWTA is formu- 
lated: ai j xi j ≥b j for j = 1, . . . , n
{i=1,...,m|(i, j )∈A}

n m 
 xi j
min Vj 1 − pi j xi j ≤ui j ∀(i, j ) ∈ A,
j=1 i=1 xi j ≥0 ∀(i, j ) ∈ A

n
This formulation is linear and is computationally simpler than S1
s.t. xi j ≤ wi , for i = 1, . . . , m
and a solution is more easily attained. Because the formulation is
j=1
an approximation, however, its solution is not guaranteed to be op-
xi j ∈ Z+ , for i = 1, . . . , m, j = 1, . . . , n . timal for S1.
228 A. Kline, D. Ahner and R. Hill / Computers and Operations Research 105 (2019) 226–236

A different transformation to S1 is put forth by 

n

Ahuja et al. (2007) by applying a logarithmic transformation to s.t. xi j = 1 for i = 1, . . . , m

the objective. Letting di j = − ln(qi j ), their formulation becomes j=1


m

n
xi j = 1 for j = 1, . . . , n
−y j
S4 min Vj2 i=1
j=1 
1 weapon j is assigned to target i

n xi j =
s.t. xi j ≤wi for i = 1, . . . , m 0 otherwise
j=1 for i = 1, . . . , m, j = 1, . . . , n

m
This allows for solution techniques which exploit the special struc-
di j xi j = y j for j = 1, . . . , n
ture of the formulation, but is only of use under certain rigid situ-
i=1
ations.
xi j ∈Z+ for i = 1, . . . , m, j = 1, . . . , n
Two additional variants to the SWTA formulations are also in
y j ≥0 for j = 1, . . . , n the literature. One, defined by Shang et al. (2007), considers the
value, ak , of a protected asset k = 1, . . . , K and the probability with
With this transformation, Ahuja et al. (2007) have an objective
which a target j will destroy this asset γ jk . Given the probability
which is the sum of separable convex functions. They utilize this
that weapon i will destroy target j, pij , they formulate
transformation to model the SWTA as a network flow problem,
which is addressed later in Section 2.1. Further, as is shown by 
K 
nk

W
Kline et al. (2017b), utilizing S4 within a commercial global op- S8 min ak γ jk ( 1 − pi j xi j )
timization solver such as BARON is more reliable than when utiliz- k=1 j=1 i=1
ing S1, which has roughly a 21% false optimality rate. 
K
Others simplify the problem by limiting the number of weapons s.t. nk = n
of each type to wi = 1, making the problem a binary program. k=1
Li et al. (2009) propose the objective function 
n
xi j = 1, i = 1, . . . , m

n 
m
S5 min Vj ( 1 − pi j xi j ), j=1

j=1 i=1 xi j ∈ {0, 1}, i = 1, . . . , m, j = 1, . . . , n

with an added constraint which limits xij to a binary decision where each protected asset has incoming targets 1, . . . , nk and
variable. S1 can be transformed to S5 by setting the number of there are a total of n targets. This formulation considers the im-
weapon types to the total number of weapons. That is, if wi = 3 portance of different protected assets, which is relevant in a mis-
and m = 5, the problem could be transformed for S5 by setting sile defense problem, but it adds complexity to the problem.
wi = 1 and m = 15. This increases the number of decision variables Karasakal (2008) does not consider a target value but rather
of the problem, though the transformation to a binary program al- treats each target to be of identical destructive capacity. He defines
lows for more efficient solution techniques. the set of feasible solutions as F, limits the number of weapons
A more simplified formulation is put forth by that can be assigned to target j as sj , and defines a formulation
Rosenberger et al. (2005), who model the SWTA as a knap-
sack problem. They define a positive cost parameter cj , which is  
S9 max 1− ( 1 − pi j )xi j
earned when assignment j is selected. Their model assumes that
j=1,...,n {i=1,...,m|(i, j )∈F }
no two weapons can be assigned to the same target and is 
 s.t. xi j ≤ wi for i = 1, . . . , m
S6 max c jx j { j=1,...,n|(i, j )∈F }
j∈J 
 xi j ≤ s j for j = 1, . . . , n
s.t. x j ≤ 1 i = 1, . . . , m {i=1,...,m|(i, j )∈F }
j∈Si
 0 ≤ xi j ≤ ui j , ∀i, j ∈ F and xi j is integer
x j ≤ 1 j = 1, . . . , n
This formulation treats all protected assets and targets as having
j∈T j
 equal value and simply seeks to maximize the expected destruc-
1 assignment j is selected tion to incoming targets.
xj =
0 otherwise
2.2. DWTA formulations
In the first constraint, the set Si is the subset of all feasible assign-
ments of which weapon i is assigned. Similarly, the set Tj in the 2.2.1. Shoot-look-shoot
second constraint is the subset of all feasible assignments which There are two variants of the DWTA, each of which have unique
assigns a weapon to target j. While simpler than S1, this formu- formulations. The first variant is the shoot-look-shoot scenario,
lation, like S2, carries more assumptions which limit its ability to wherein weapons are assigned to targets in a first engagement and
model and solve complex missile defense problems. a subsequent engagement allows assigning remaining weapons to
Malcolm (2004) proposed a formulation with the same binary any surviving targets. This problem was discussed by Eckler and
decision variables in which the objective is similar in structure to Burr (1972), who do not define a model but define the probability
S5. He shows that, when weapon assignments are restricted to ex- that n targets are destroyed over t stages, which is equivalent to
actly one target and m = n, the objective can be written as the probability that at least n weapons do not fail over t stages, as
 

n 
m 
t
t
S7 min − Vj xi j pi j P (t ) = (1 − p)t−i pi , (1)
i
j=1 i=1 i=n
 A. Kline, D. Ahner and R. Hill / Computers and Operations Research 105 (2019) 226–236 229

for a problem wherein all probabilities of kill are the same and where EFFi/j is the earliest feasible fire time for weapon i to target
all targets are of the same value. Eckler and Burr (1972) identify j, LFFi/j is the latest feasible fire time for weapon i to target j, and
that the most desired strategy to the t stage problem under this d (Pjout , Pi0 ) is the Euclidean distance that the weapon, i0 , must fly
assumption will be equivalent to finding the number of weapons to over the protected area to intercept target j.
use in each stage which minimize the number of stages necessary Using these three parameters for each pairing,
to achieve some acceptable value of P(t). Leboucher et al. (2013) creates a cost matrix for all of the
Soland (1987) provides a model for the Eckler and possible assignments
Burr (1972) scenario with the assumption that all weapons ⎡ ⎤
E1/1 E2,1 ··· E|I |/1
have the same probabilities of kill and all targets have the same
⎢ E1/2 E2,2 ··· E|I |/2 ⎥
H=⎢ ⎥ ,
value. Given a nondecreasing function g(nq ) which defines the
⎣ .. .. .. .. ⎦
expected fraction of targets destroyed, where nq is the number of . . . .
unintercepted targets, the state of the system S(nq , d, t) defines E1/| J | E2,| J | ··· E|I |/|J |
the fraction of targets destroyed given nq targets, d weapons, and
t remaining engagements. The state space is bound by in which the cost of an assignment H(Ei/j ) is

S(0, d, t ) = 0, for d = 0, 1, . . . , D, t = 1, . . . , T , H (Ei/ j ) = α1 f1 (Ei/ j ) + α2 f2 (Ei/ j ) + α3 f3 (Ei/ j ),

S(nq , d, 0 ) = g(nq ), for nq = 0, 1, . . . , n, d = 0, 1, . . . , D. where α1 + α2 + α3 = 1 and (α 1 , α 2 , α 3 ) ∈ [0, 1]3 . He presents a
He defines the transition probability that j targets survive having formulation to determine the assignments
used i weapons as P(j|nq , i, d, t), where

m 
n

nq D3a min H (Ei/ j )xi j

P ( j|nq , i, d, t ) = qtjI (1 − qtI )nq − j . i=1 j=1
j

m 
n
I defines the spread of weapons to targets, or I = i
in the case s.t. xi j ≤ m
nq
 i=1 j=1
i i
that nq is integer. If it is not integer, then nq + nq nq − i of the
  and defines, for each possible firing sequence (FS), three parame-
i i
targets receive nq weapons and the remaining i − nq nq tar- ters. First, given the time at which target j is engaged, FTj , the pa-
  
i rameter measuring the firing time is f4 (F S ) = Tj=1 F T j . Second, a
gets receive nq weapons. His model seeks to minimize the num- 
parameter identifying any constraint violation is f5 (F S ) = m i=1 ci ,
ber of weapons required to ensure that the expected number of where ci is 1 if the assignment of weapon i violates a constraint
surviving targets is less than some “nondecreasing maximum dam- and 0 otherwise. Lastly, the parameter representing idle time of
age function f” (Soland, 1987) the system, given the time at which weapon i is fired, FTi , is
 −1
D1 min D f 6 (F S ) m
i=1 (F Ti+1 − F Ti ). Leboucher et al. (2013) present a formu-
lation whose solution gives the optimal firing sequence of the as-
s.t. S(nq , D, T )≤ f (nq ), nq = 1, . . . , n.
signment solution
Hosein and Athans (1989) provide a different model than
Soland (1987), but with the same underlying assumptions. They D3b min F (F S )

define nk (t) as the number of targets aimed at protected asset k ( f 4 (F S ) + 1 ) ∗ f 6 (F S ) f 5 (F S ) = 0
at stage t, or the number of surviving targets after t − 1 engage- s.t. F (F S ) =
∞ f 5 (F S ) = 0 .
ments. They compute the probability that the number of targets
surviving into the second stage is j(2) given the assignment in the The formulation presented by Leboucher et al. (2013) enables
 stage (2is) x(1) , for all i =∗ 0, 1, . . . , m and j = 0, 1, . . . , n(1 ) as
first (1)
observation of surviving targets following an engagement which
P n ( 2 ) = j |x . Defining Js (n(2 ), m2 ) as the optimal solution to can be reengaged in a subsequent iteration and can be used to
the second stage with m2 weapons available, they define the for- solve a shoot-look-shoot problem.
mulation
D2 max Jd = E [Js∗ (n(2 ), m2 )] 2.2.2. 2-stage
x(1) ∈ZK+ n (2 )
The second variant of the DWTA is the 2 stage, or more gen-
s.t. |x(1) | + m2 = m. erally the multi-stage, problem, which differs from the shoot-look-
shoot in that it does not allow the reacquisition of leakers. That
The objective is the expected value of the optimal solution in the
is, in the shoot-look-shoot problem, a given number of targets are
second stage, thus the optimal solution to the problem is to find
repeatedly engaged until all have been destroyed or a limit to the
the number of weapons to use in the first stage, m1 , and assign
number of iterations is met. In the 2 stage problem the given num-
them to the appropriate targets, x(1) , in such a way that the second
ber of targets is only engaged once before a subsequent stage oc-
stage can be solved to optimality given the number of surviving
curs. In the second stage, the number and type of incoming targets
targets and the number of unused weapons, m2 .
is known only to a probability distribution.
A different approach, which considers the available win-
Chang et al. (1987) model the T stage WTA by considering the
dows in which targets can be engaged, is proposed by
value of each stage as defined by the formulation S1 and taking
Leboucher et al. (2013). He computes random paths of randomly
the sum over the T stages.
located targets using Bézier curves, which allow for the calcula-
tion of the time to impact for each target and the earliest point at 
T  m 
 xi j (t )
which each weapon can engage the target. For each weapon-target D4 min E V j (t ) 1 − pi j (t )
xi j (t )
pairing, he computes: t=1 j∈At i=1

f1 (Ei/ j ) = EF Fi/ j , (i ∈ I ), ( j ∈ J ) s.t. At+1 = (At ∪ Lt ) ∩ Kt

f2 (Ei/ j ) = LF Fi/ j − EF Fi/ j , (i ∈ I ), ( j ∈ J ) 
n
Mi (t ) = Mi (t − 1 ) − xi j (t − 1 ) i = 1, . . . , m
f3 (Ei/ j ) = d (Pjout , Pi0 )
j=1
230 A. Kline, D. Ahner and R. Hill / Computers and Operations Research 105 (2019) 226–236


n(t )
xi j (t ) ≤ Mi i = 1, . . . , m 
T 

j∈At xi j (t ) ≤ wi for i = 1, . . . , m
t=1 j=1
xi j (t ) ∈ Z+ i = 1, . . . , m, j = 1, . . . , n
xi j (t ) ≤ fi j (t ) for i = 1, . . . , m, j = 1, . . . , n,
where At is the set of targets in stage t, Mi (t) is the number of in-
t = 1, . . . , T
terceptors of type i at stage t, Lt is the set of new targets observed
in stage t and Kt is the set of targets killed in stage t. where
Burr et al. (1985) present a formulation for a multi-stage prob-
lem in which all weapons have the same probability of kill for all K (t ) : number of existing assets at time t
equally valued targets. That is, a known number of targets arrives n(t ) : number of existing targets at time t
in the first stage and a number arrive in each of the number of m(t ) : number of available weapons at time t
subsequent stages, both of which are known only to a probabil-
ak : the value of asset k
ity distribution. Given the attack strategy a(k), which identifies the
number of targets aimed at asset k, and a defense strategy d(k, j), γ jk : the probability target j destroys asset k
which defines how many weapons to shoot at the target j threat- pi j (t ) : the probability weapon i destroys target j at time t
ening asset k, Burr et al. (1985) formulate the problem
ni : maximum number of targets weapon i can shoot at
a (k )

K  each stage
D5 min d (k, j )
d (k, j ) s j : maximum number of weapons that can be assigned
k=1 j=1
to target j at each stage
V 
K
s.t. V (d, a ) ≤ a(k ), for all a wi : total number of weapons of type i
n
k=1 fi j (t ) : 1 if weapon i can be assigned to target j, 0 otherwise.
where V is the sum of all protected asset values, n is the total num-
This is one of the more complex and realistic models that can be
ber of targets, and V(d, a) is the expected damage to the protected
found within the literature. It allows for expansion into a shoot-
asset given a targets and deployment strategy d
look-shoot problem and considers many parameters which are rel-
  evant to modeling missile defense. However, as the complexity is
a (k )

K  
d (k, j )
V (d, a ) = ak 1 − 1−q . higher than other formulations, finding solutions is more compu-
k=1 j=1 tationally expensive than for simpler formulations.
A model proposed by Khosla (2001) considers the required time
This formulation seeks to ensure that the expected damage to the for weapon system control and defines the following terms:
protected assets is less than the total value of all protected assets.
Murphey (20 0 0) formulates the multi-stage problem by defin- n : Number of current threats
ing n(t) as the number of targets which arrive at time t = 1, . . . , T m : Number of current weapon systems
and c(t) as a nondecreasing function which represents a cost of
T : Total number of time points in time interval
waiting. With the assumption that all weapons have the same
probability of kill for target j, his formulation is T V ( j ) : Threat value of threat j

n(t )
OW (i, j ) : Option weight of weapon system i for threat j

T  x (t )
D6 min c (t ) Vjq j j LB(i, j ) : Begin launch time for weapon system i for threat j
t=1 j=1 LE (i, j ) : End launch time for weapon system i for threat j
n(t )

T  GT (i, j ) : Guidance time interval for interceptor for weapon
s.t. x j (t ) = m,
system i to engage threat j
t=1 j=1
IR(i ) : Inventory resource of weapon system i (number of
V j ∈ V ∈ Rn+ j = 1, . . . , n ( T )
interceptors)
x(t ) ∈ Zn+(t ) t = 1, . . . , T
GR(i ) : Guidance resource capacity of weapon i
where m is the total number of weapons and V is the set of all tar-
where GT(i, j) defines the amount of time the guidance system
get values. This formulation is seeking to minimize the value of the
must be allocated to weapon i in targeting target j and GR(i) is the
assignments and the inclusion of a non-decreasing cost of waiting
number of guidance systems available. The option weight OW(i, j)
function will bias the solution to make earlier assignments unless
serves to add a benefit to preferred pairings; a bias for weapon i
these assignments are to targets of a very small value relative to
to be assigned to target j.
those of later assignments.
Khosla (2001) defines a mixed integer program with only a few
Xin et al. (2011) allow for different probabilities of kill for each
of the considerations discussed thus far, proposing that expanded
weapon to each target and further allow for different probabilities
models including additional time constraints such as reload time.
of kill between stages. Their formulation for stages t = 1, . . . , T is
  Using a decision variable L(i, j, t ) = 1 if t denotes the launch time
K (t ) n(t ) (t ) 
   
T m xi j (h) for an interceptor from weapon system i to engage threat j and 0
D7 min Vk 1 − γ jk 1 − pi j ( h ) otherwise, he models this problem, with a weight factor α ∈ [0, 1],
k=1 j=1 h=t i=1 as
n(t )
 
m 
n 
T
s.t. xi j (t ) ≤ ni for i = 1, . . . , m, t = 1, . . . , T D8 max [α T V ( j ) + (1 − α )OW (i, j )]L(i, j, t )
j=1 i=1 j=1 t=1
m(t )
 
m 
T
xi j (t ) ≤ s j for j = 1, . . . , n, t = 1, . . . , T s.t. L(i, j, t )≤ 1, for j = 1, . . . , n
i=1 i=1 t=1
 A. Kline, D. Ahner and R. Hill / Computers and Operations Research 105 (2019) 226–236 231


n 
T x jt ∈ {0, 1} for j = 1, . . . , n(t ), t = 1, . . . , T
L(i, j, t )≤ IR(i ), for i = 1, . . . , m
j=1 t=1
where there are T stages, n(t) targets per stage, and m available
weapons. Further, the value of target j is Vj and its time discount
I f L(i, j, t ) = 1, R(i, j, t ) = 1, ∀ t ≤t ≤ t + GT (i, j ) factor at stage t is λtj . Given that xjt is a binary decision variable,

n 
T
the expression is the sum over all stages of the sum over all as-
R(i, j, t )≤ GR(i ), for i = 1, . . . , m signments. The time discounting factor imposes a bias on earlier
j=1 t=1
assignments. The product of the probability, target value, time dis-
In this model, Khosla (2001) ensures that only one weapon system count factor, and decision variable result in either a value of 0 or
is assigned to each target, the total number of interceptors does the time discounted expected value of the assignment. Though a
not exceed the inventory, and the guidance time required for the superior assignment may occur as the target is closer to the pro-
assignments does not exceed the number of guidance systems. He tected asset (pj (t) is a function of time), the value of the target de-
nests this model into a framework which updates the number of creases since a miss will reduce the ability to reengage.
targets after each completed iteration and, as such, can be used
in a 2 stage scenario where the number of targets in subsequent 3. Exact algorithms
stages is stochastically determined.
This model simplifies some of the parameters and considera- There are few cases in the literature of exact algorithmic so-
tions posed in D7, but includes a discretized time step which pro- lutions to the WTA. The problem suffers due to its complexity
vides the firing sequence inherent to, but previously not consid- as an NP-Complete problem (Lloyd and Witsenhausen, 1986) and,
ered, missile defense. However, due to the exponential growth of like routing problems, are simply hard to solve. For the SWTA, the
the number of decision variables with the increase in resolution, number of possible permutations of assigning m weapons to n tar-
Khosla (2001) identifies that even modest sized problems are very gets is nm , assuming that all weapons must be assigned, which, as
computationally expensive. the number of weapons and targets increases, grows exponentially
Ahner and Parson (2015) address the SWTA formulation, orig- and searching all possible solutions quickly becomes computation-
inally proposed by Murphey (20 0 0), to compute the value of the ally intractable. Because the DWTA includes either a shoot-look-
first stage and includes in the objective the expected value of the shoot or multiple stage (or both) framework, it further increases
second stage, with the maximum number of weapons, b < M, that the number of permutations. The literature implementing exact so-
can be used in any stage, as follows lution techniques generally fall into one of two categories: small
 n1   
 (1 )  x(1)  problems and problems wherein assumptions reduce the complex-
D9 min V j 1 − p(j1) j + E Z2 x ( 2 ) , ω j ity.
x ω∈
j=1

s.t. x (1 ) ≤ b 3.1. SWTA

x j ∈ Z+ , for j = 1, . . . , n
denBroeder et al. (1959) showed the first optimal solution tech-
(1 ) nique in their Maximum Marginal Return (MMR) algorithm. As-
where x j defines the number of weapons fired at target j in the
suming that the probability of kill for any weapon to target j is
first stage and the total number of weapons fired in the first stage
n1 (1 ) the same, they showed that an optimal solution can be found by
is x(1 ) = j=1 x j . Further, the second stage, Z2 , is a function of
assigning xi j = 1 where {i, j} ∈ arg max(V j pi j ) and then updating
the remaining weapons, x(2 ) = M − x(1 ) , and a random occurrence, V j = V j (1 − pi j ) and p(i, · ) = p(·, j ) = 0 and repeating the process
ω ∈  of the number of targets and the type of each target. until all weapons have been assigned. Further, when the proba-
n 2 ( ω )  bilities of kill are the same for all weapons to all targets, pi1 j1 =
   (2 )  x(2)
Z2 x (2 )
,ω j
=min V j (ω ) 1 − p(j2) j pi2 j2 , ∀i1 , i2 ∈ I, j1 , j2 ∈ J, the optimal solution is found by divid-
x (2 )
j=1 ing the weapons evenly across all targets (Hosein et al., 1988).
2(
Malcolm (2004) developed and solved S7, where he defined his
 ) (2 )
n ωj

n1
constraint coefficient matrix A as totally unimodular. This ensures
s.t. x(j1) + xj = M
that every vertex of the convex polytope that defines the feasible
j=1 j=1
solution space is an integer solution. As such, he uses the Simplex
(2 )
x ≤b Method to quickly find the optimal solution.
(2 ) Smaller problems were solved through an exhaustive search al-
xj ∈ Z+ , for j = 1, . . . , n
gorithm for S1 by Johansson and Falkman (2009). In comparing the
Unlike other formulations, this formulation accounts for the uncer- objective function value of every feasible solution, they show that
tainty of a second stage which must be considered when allocating a problem with 9 weapons and 8 targets took 13 min to run to
available weapons to the first stage. However, it is a homogeneous completion and that adding one additional target took 43.7 min to
model which assumes that all weapons have the same probability run to completion, which they present to illustrate the combinato-
of kill for a target j. rial explosion in run time as a function of problem size.
Sikanen (2008) models the DWTA wherein all weapons have Several cases of using a branch and bound algorithm are found
the same probability of kill for all targets in the literature. Rosenberger et al. (2005) solved S6 for up to
n(t ) 8 weapons and 4 targets. Ahuja et al. (2007) implemented three

T 
D10 max p j (t )λtjV j x jt lower bounding strategies to increase the efficiency of fathoming
xi1 ,...,xiT
t=1 j=1 nodes: a generalized network flow solution, an MMR solution, and
a minimum cost flow solution. Kline (2017) developed a branch

T
s.t. x jt ≤ 1 for j = 1, . . . , n(t ) and bound algorithm to solve S1 and was able to find optimal solu-
t=1
tions for up to 10 weapons and 10 targets. Beyond this, the size of
n(t )
the problem precluded convergence within 7 days of computation.

T 
Karasakal (2008) utilizes linear integer programming techniques
x jt ≤ m
to find optimal solutions to two linear transformations of S9.
t=1 j=1
232 A. Kline, D. Ahner and R. Hill / Computers and Operations Research 105 (2019) 226–236

Bogdanowicz (2012) develops and utilizes an algorithm by are of well known heuristic algorithms, such as the very large scale
which he searches through known effective weapon-target pair- neighborhood (VLSN) search or the Genetic Algorithm (GA), but
ings to find an optimal solution. Utilizing the Joint Munition Ef- others are of new design, seeking to exploit the special structure
fectiveness Manual (JMEM), he defines the desired minimal effect of the WTA.
of any one pairing to reduce the number of sets through which he
searches for an optimal set of pairings, given the number and type 4.1. SWTA
of weapons and targets.
The heuristic algorithms applied to the SWTA often fall into one
3.2. DWTA of several groups. Herein, we will explore some of the varying ap-
proaches within these groups.
Burr et al. (1985) puts forth an optimal algorithm for D4 given
a scenario wherein one target per stage is observed and a defender 4.1.1. MMR
will assign weapons for up to k − 1 stages, after which he will sur- Kolitz (1988) implemented the MMR algorithm and, unlike
render the protected asset. Given a maximum total expected dam- denBroeder et al. (1959), did not assume that all weapons had
age of 1, the defender must limit the expected damage per stage the same probability of kill for any target j, but rather that each
to no greater than r = 1k , where each weapon has a probability of weapon’s probability of kill for any target j was independent.
kill of p = 1 − q. His algorithm is to set the minimum number of Julstrom (2009), Madni and Andrecut (2009), and
weapons to ensure damage does not exceed r for the first stage Gelenbe et al. (2010) implement the MMR algorithm as a
as comparative benchmark in testing their heuristic approaches.
 
ln(r ) Ahuja et al. (2007) utilize the MMR algorithm as one of three
d (1 ) = lower bounding schemes for their branch and bound algorithm.
ln(q )
and all subsequent stages as 4.1.2. Genetic algorithms
⎡  ⎤ There have been several implementations of the GA in the
1−rk
ln 1 − k SWTA, each with a minor adjustment yet the same in structure
⎢ −1
(
1−qd (i ) ) ⎥
d (k ) = ⎢ ⎥,
i=1
and execution. Metler et al. (1990) was the first to implement
⎢ ln(q ) ⎥
⎢ ⎥ the GA for the SWTA. Lee et al. (2002b), Zhihua et al. (2009),
Lee and Lee (2005), Bogdanowicz et al. (2013), Li et al. (2009),
where 1 < k ≤ k − 1. Fu et al. (2006), Lee et al. (2003), Lu et al. (2006), and
Soland (1987) gives an optimal solution to the shoot-look-shoot Wu et al. (2008) are among the many subsequent researchers that
model D1 in which the number of weapons i to assign! to the total utilized the GA for the SWTA.
number of targets a in the first stage is simply ai and to pre-
!
serve the remaining i − ai weapons for the surviving targets. If 4.1.3. VLSN
the problem allows for more than two stages, he iteratively per- The very large scale neighborhood (VLSN) search metaheuristic
forms this allocation, utilizing the largest integer contained in the is used by Ahuja et al. (2007) and Lee (2010) to improve upon in-
fraction of available weapons to surviving targets in the immediate formed feasible solutions. Their VLSN algorithms execute a heuris-
stage and preserving the remaining weapons for the subsequent tic search to efficiently find a quality solution and then they define
stage. local search neighborhoods within which to search for superior so-
Hosein (1989) proves that the optimal solution to D2 is to lutions.
spread the number of weapons used for each stage t, mt , as evenly
as possible, which is similar to Soland (1987). He therefore seeks 4.1.4. Ant Colony Optimization
to optimize over the decision variables mt the minimum value of The Ant Colony Optimization (ACO) is another heuristic that is
the final stage. frequently implemented. It was first used by Lee et al. (2002a),
Ahner and Parson (2015) generate an optimal strategy for D9 and Yanxia et al. (2008), Lee and Lee (2003), Shang (2003),
through the implementation of the Concave Adaptive Value Esti- Shang et al. (20 07), Shang (20 08), Huang and LI (2005), and
mation (CAVE) algorithm with a modified MMR algorithm which Su et al. (2008) among others have used the ACO to solve the
they call the MMR Plus Algorithm. The CAVE algorithm estimates SWTA.
the value of second stage assignments by utilizing random realiza-
tions of the number of targets in the second stage and iteratively 4.1.5. Other heuristic algorithms
updating the subgradient of a concave value estimation, the CAVE Other techniques that do not fall into more generalized group-
function. Their MMR Plus algorithm assigns weapons to known tar- ings have been demonstrated to efficiently find quality solutions
gets in the first stage and, by comparing marginal returns of as- to the SWTA. Day (1966) solves an integer relaxed NLP and
signments to the CAVE function, indicates how many weapons to utilizes rounding schemes. Wacholder (1989) implemented neu-
preserve for the second stage. Though the CAVE Algorithm is an ral networks to find robust solutions. Ahuja et al. (2007) used
approximation technique, Ahner and Parson (2015) prove the con- a network flow based construction heuristic to find near opti-
vergence to the optimal solution in the DWTA wherein all weapons mal solutions to some of the larger problems in the literature.
have the same probabilities of kill to target j. Tokgöz and Bulkan (2013) compared the results of GA, Simu-
Sikanen (2008) uses dynamic programming to solve D10. He lated Annealing (SA), Variable Neighborhood Search (VNS), and
uses a backwards induction process to recursively define the policy Tabu Search algorithms. Johansson and Falkman (2010) use Par-
which will optimize the problem. ticle Swarm Optimization (PSO) and compare computational re-
sults to the GA, MMR, and exhaustive search algorithms. Similarly,
4. Heuristic algorithms Zeng et al. (2006) compares PSO with GA and a GA improved by
greedy eugenics. Kwon et al. (1999) solves S3 using a Lagrangian
Due to the computational complexity of the WTA, much of the relaxation Branch and Bound Algorithm. Kline (2017) implemented
literature focuses on heuristic algorithms which provide real time the filtered beam search heuristic on S1, developed a heuristic
solutions rather than guaranteed optimal solutions. Many of these based upon the optimal solution to the quiz problem and improved
 A. Kline, D. Ahner and R. Hill / Computers and Operations Research 105 (2019) 226–236 233

on these initial solutions using a metaheuristic which iteratively problem and the solution techniques implemented. In the earliest
blocked assignment pairings which may have prevented superior works, reference to the limited capacity to solve large nonlinear
solutions (Kline et al., 2017a) and also developed a heuristic with problems (Day, 1966) resulted in attention on simplified formu-
similarities to the Hungarian Algorithm (Kline et al., 2017b). See lations of the SWTA (denBroeder et al., 1959) and solution tech-
Hill and Pohl (2010) for a description of GA, SA, ACO, Tabu Search, niques capable given the computational capacity of the day (i.e.,
and PSO. (Lemus and David, 1963), (Day, 1966)). Eckler and Burr (1972) pro-
posed and discussed the possibility of solving dynamic variants of
4.2. DWTA the SWTA but were unable to generate algorithms to solve such
problems.
Less attention has been given to the DWTA as compared to the As computational power increased, so too did the ability to
SWTA. Thus, there are fewer heuristic algorithms shared among solve problems of increased complexity. Burr et al. (1985) mod-
researchers. Often, hybrid heuristic algorithms are used to inform eled and solved one of the earliest DWTA problems, as did
one another in execution. Chang et al. (1987), Soland (1987), and Hosein et al. (1988). Mean-
Metler et al. (1990) propose three greedy heuristics, the first of while, models of the SWTA with fewer assumptions were solved
which is simply the MMR algorithm. In the second heuristic, the with novel approaches (i.e., Kwon et al., 1999; Metler et al., 1990;
expected value of each pairing is computed and the selection of a Wacholder, 1989).
random number determines the assignment based upon a proba- This pattern continued into the 20 0 0s, with model develop-
bility mass function for which assignments with higher expected ments either capturing additional parameters which more closely
values have higher probabilities. The third heuristic proposed by resemble reality (i.e., Shang et al., 2007 and Karasakal, 2008)
Metler et al. (1990) is called the ALIAS Algorithm. This algorithm or models which enabled faster optimal or near optimal solu-
first updates the value of a target in stage t by dividing the value tions (i.e., Malcolm, 2004, Ahuja et al., 2007, and Ahner and Par-
of the group by the total number of targets of type j in stage t. son, 2015). Once developed, these models were solved using newer
It then assigns weapons to targets based upon an MMR procedure, approaches (i.e., Bertsekas et al., 20 0 0; Kline et al., 2017a; Wu
updating the probabilities of kill and repeating until an assignment et al., 2008) or combinations of existing approaches which could
violates one of the constraints or the maximum number of itera- be implemented efficiently (i.e., Ahuja et al., 2007; Lee et al.,
tions has occurred. 2002a; Su et al., 2008, and Xin et al., 2010).
Chang et al. (1987) developed a heuristic algorithm for D4 As computational power continues to grow, the WTA will likely
which utilizes a heuristic subroutine to solve the first stage. An continue to be the subject of research which improves upon ex-
iterative process then decrements the number of weapons to use isting solution techniques. Dynamic models which consider the
in the first stage based upon its marginal contribution until the time dependence of weapon utilization and target flight paths
contribution is greater than some  , at which point the number of have been proposed (Khosla, 2001; Leboucher et al., 2013) but
weapons for the first stage is fixed and the second stage is consid- have received less attention than existing models. Improvements
ered. This process iterates until either all weapons have been as- to the solution techniques in these models are yet to emerge,
signed or all stages have been considered. As a subroutine to solve and as remarked by Khosla (2001), “in spite of the two-step ap-
the first stage, Chang et al. (1987) use three different heuristics: proach [outlined in Khosla (2001)], each of the optimization prob-
the MMR, an iterative linear network programming algorithm, and lems still have a huge search space even for a modest number
a nonlinear network flow algorithm. of threats, weapon systems, and time points.” Methods of im-
Murphey (20 0 0) develops a decomposition algorithm to solve proving on the two-step approach are yet to emerge. Similarly,
D9. In this heuristic, he solves the first stage by some heuristic al- Leboucher et al. (2013) remarks on the exponential growth of the
gorithm, saving the first stage solution and expected second stage problem and proposes a two-step solution technique, adding that
solution. After this he solves the second stage primal and dual for- an additional problem is “to be able to quantify the quality of one
mulation across all possible second stage target outcomes. He uses proposed solution.”
these solutions to define the expected objective function value of The future of the WTA will need to address the aforementioned
the second stage. If this value exceeds the expected second stage difficulties of the scheduling-focused DWTA with techniques capa-
objective function value previously determined, he adds a cut to ble of exploiting the special structure of the problem. Additionally,
the problem and repeats the process. there exist many parameters of the problem which are removed
Xin et al. (2010) solve D7 using Virtual Permutation (VP), TS, due to the increased computational complexity they would bring
GA, and ACO. In a subsequent work, they developed a rule-based that could be introduced using novel modeling techniques.
heuristic to solve D7 in which they consider the saturation of
the constraints in order to inform the greedy selection process by
5.2. Recent developments
which they assign weapons to targets in a stage t (Xin et al., 2011).
Leboucher et al. (2013) use a Hungarian Algorithm to solve
While the focus of the research discussed heretofore focuses
the assignment pairings for D3a and uses a GA-PSO hybrid al-
primarily on the static and dynamic allocation of interceptors to
gorithm to solve D3b in order to determine the firing order of
offensive missiles, recent research has provided different frame-
the assignments. Khosla (2001) uses a GA-SA hybrid algorithm
works through which this problem is addressed. We briefly discuss
to solve D8. Chen et al. (2009) implements a GA to solve D7.
these recent developments here.
Bertsekas et al. (20 0 0) uses Neuro-dynamic programming to ob-
tain near optimal policies which he compares to optimal policies
obtained through dynamic programming. 5.2.1. Sensor Weapon Target Assignment Problem
Missile defense depends on the accuracy and reliability of sen-
5. Discussion sors to identify the type and position of each incoming missile so
as to appropriately defend a protected asset. Much of the literature
5.1. Evolution of WTA disregards the allocation of sensors and assumes the defender’s
omniscience. However, different approaches concerning the consid-
Research on the WTA has evolved since the work of eration of a finite number of sensors are found within the litera-
Manne (1958) with developments in both the formulation of the ture.
234 A. Kline, D. Ahner and R. Hill / Computers and Operations Research 105 (2019) 226–236

Bogdanowicz et al. (2007) develop a model that seeks to max- Gülpınar et al. (2018) framed their model and solution technique
imize the sum of the benefits of assigning each sensor to each for a dynamic resource allocation problem on much of the same
target and each weapon to each target. Zi-fen et al. (2011) com- literature that is outlined in Sections 2–4 of this survey. Çetin and
bine the auction algorithm based technique developed by Esen (2006) model and solve a media allocation problem with an
Bogdanowicz et al. (2007) to reduce the limitations that an im- objective function which, if Vj is the audience type value, pij is
perfect network topology would introduce. the probability that audience j views advertisement i, and deci-
Others have considered the effect that sensors have on sion variable xij is the number of advertisements of type i to as-
the probability of detecting incoming missiles. Jian and sign to audience j, is the formulation S1. Onay et al. (2016) model
Chen (2015) models the damage probability of an interceptor neuromarketing with S1 as an objective function where Vj is the
as the probability that a sensor will identify the missile and the value of the brain stimulus, pij is the probability that stimu-
destructive capacity of the weapon with which the sensor is lus i affects the brain region j, and decision variable xij is the
paired. Xin et al. (2018) extends this by modeling the probability number of stimulants of type i to assign to brain region j. An-
of successful engagement as the product of the interceptor’s other application using objective functions similar to S1 is can-
probability of kill and the sensor’s probability of detection. cer treatment. The targeting of cancer cells with medication is
modeled and solved by Çetin (2007) and Esen et al. (2008) us-
5.2.2. Multi-objective programs ing WTA research. Both Alighanbari (2004) and Bertuccelli and
Each of the formulations presented in Section 2 seeks to maxi- How (2011) model and solve unmanned aerial vehicle (UAV) as-
mize the probability of destruction of all of the incoming missiles signment planning problems with static and dynamic WTA models.
in some capacity. While the DWTA formulations include parame- Lastly, Gelenbe et al. (2010) use WTA research to model and solve
ters and constraints that promote the preservation of some of the a problem of dispatching ambulances to emergencies with an ob-
interceptors for subsequent salvos or subsequent shots to a leaker, jective function that is similar to formulations S1 and S3.
solving each of these formulations results in the consumption of all
available resources. Though it is important to defend a protected 5.4. Analysis of literature influence
asset, there may be situations in which it is beneficial to conserve
interceptors. As such, there has been research into the simultane- Matlin (1970) put forth the first survey of the WTA literature,
ous maximization of damage and minimization of shots. characterizing the problem with five components and defining el-
Li et al. (2015) model the DWTA with an objective that si- ements of the problem which structured its subsequent research.
multaneously maximizes the expected damage and minimizes Due to the high volume of literature at present, a strategy to focus
the ammunition consumption. They compare the performance the considered literature for this survey was necessary, else an ex-
of two solution techniques for this bi-objective program and haustive list of the literature would obfuscate the state of the WTA
later develop and compare a third technique (Li et al., 2017a). and how it came to be. We considered the relevance of any work
Li et al. (2017b) solve a similar formulation with a modified ant in the literature to be a function of its usefulness to subsequent
colony algorithm. Li et al. (2018) include a third objective which research and used a rate of citation metric as a tool to limit our
seeks to maximize the value of each weapon type and use a ge- discussion heretofore.
netic algorithm to solve the multi-objective model. Table 1 shows that much of the work used in this survey with
higher citation rate, given the current year of 2018, focuses on
5.2.3. Game theory approaches heuristic solutions to the SWTA. The entries within this table rate
While all of the research discussed thus far addresses the re- approximately 2 citations per year or more and demonstrate, by
sponse to an adversary with no consideration of the adversary’s re- their consistent impact on research, the importance and substan-
action, there has been research on this game theory aspect to mis- tial contribution they have made to the WTA literature.
sile defense. In contrast to the discussion regarding the research At the same time, we find that this metric, while helpful in re-
utilizing sensors and the research of multi-objective programs, the ducing the volume of literature to consider, can lead us to consider
research of game theory approaches does not conform to similar some works as less relevant due to the lower citation rate which
models. is sensitive to original publication date. Despite this reduced rate,
Shan and Zhuang (2013) develop a model that considers the works in Table 2 have a large number of citations and are foun-
the impact of defensive resource allocation in the face of dational in much of the literature we consider highly relevant. As
strategically focused and non-strategically focused adversaries. such, we include these works.
Golany et al. (2015) develop a model that seeks to place defen-
sive resources in order to defend multiple assets and extend this 6. Conclusion
model with a superior solution technique in Golany et al. (2017).
Similarly, Boardman et al. (2017) models such a scenario and con- The WTA has a rich breadth of literature which serves to im-
siders interceptor probabilities of kill. prove upon the theory and techniques necessary to efficiently solve
Shalumov and Shima (2017) models a scenario wherein the pro- these complex problems. Early works sought to find methods to
tected assets are maneuvering aircrafts. Their model considers the transform the problem into a simpler form, assume many of the
flight paths of the aircrafts, the trajectories of the missiles, and the complexities away, or do both in order to generate a formulation
probabilities of kill of the interceptors in order to best guide the which was manageable with the computational capacity of the day.
aircrafts and their defensive actions. Within the simulation they The theories and techniques proposed by the earliest researchers,
run, Shalumov and Shima (2017) test different assignment algo- such as Manne (1958) and denBroeder et al. (1959), inform much
rithms within a small scale two agent game. of the current research and built a foundation upon which sub-
sequent researchers were able to extend the theory and solution
5.3. Alternate applications techniques of the WTA.
In this survey, we have provided nine static models and ten dy-
The WTA literature informs research beyond missile defense. namic models for the WTA which have had an impact on the liter-
Often, WTA works are cited for their modeling or solution ature and have provided insights into the problem from a model-
techniques, as they are applicable in many assignment prob- ing perspective. Additionally, we have reviewed some of the exact
lems with quantifiable rewards or costs and limited resources. algorithms, heuristic algorithms, and metaheuristic algorithms for
 A. Kline, D. Ahner and R. Hill / Computers and Operations Research 105 (2019) 226–236 235

Table 1
WTA literature by citation density .

Author Year Citations Citation rate SWTA or DWTA Exact or Heuristic

Lee, Z 2003 257 17.13 SWTA Heuristic

Ahuja 2007 187 17 SWTA Both
Lee, Z 2002 249 15.56 SWTA Heuristic
Lee, Z 2005 117 9 SWTA Heuristic
Lloyd 1986 230 7.19 SWTA Exact
Xin 2011 50 7.14 DWTA Heuristic
Xin 2010 55 6.88 DWTA Heuristic
Karasakal 2008 57 5.7 SWTA Exact
Ahner 2015 15 5 DWTA Exact
Gelenbe 2010 38 4.75 SWTA Heuristic
Chen 2009 42 4.67 DWTA Heuristic
Bertsekas 20 0 0 83 4.61 DWTA Both
Rosenberger 2005 59 4.54 SWTA Exact
Lee, M 2010 36 4.5 SWTA Heuristic
Yanxia 2008 42 4.2 SWTA Heuristic
Zeng 2006 50 4.17 SWTA Heuristic
Wacholder 1989 119 4.10 SWTA Heuristic
Lee, Z 2002 62 3.88 SWTA Heuristic
Bogdanowicz 2013 16 3.2 SWTA Heuristic
Eckler 1972 144 3.13 Both Exact
Khosla 2001 52 3.06 DWTA Heuristic
Murphey 20 0 0 54 3 DWTA Heuristic
Lee, Z 2003 43 2.87 SWTA Heuristic
Johansson 2011 18 2.57 SWTA Heuristic
Matlin 1970 115 2.40 SWTA Exact
Madni 2009 21 2.33 SWTA Heuristic
Hosein 1988 63 2.1 DWTA Exact
As of September 2018

Table 2
Included foundational WTA literature by citation count.

Author Year Citations Citation rate SWTA or DWTA Exact or Heuristic

Manne 1958 118 1.97 SWTA Exact

Hosein 1989 46 1.59 DWTA Exact
denBroeder 1959 83 1.41 SWTA Exact
Soland 1987 31 1 DWTA Exact
Day 1966 53 1.02 SWTA Heuristic
As of September 2018

the static and dynamic WTA. Some of these algorithms are widely Conference on Applied Mathematics. World Scientific and Engineering Academy
used in the literature, such as the branch and bound algorithm and Soc.(WSEAS) Stevens Point, WI, pp. 92–96.
Bogdanowicz, Z.R., 2012. Advanced input generating algorithm for effect-based
or the genetic algorithm. Others, such as the algorithm developed weapon–target pairing optimization. IEEE Trans. Syst. ManCybern. Part A 42 (1),
by Bogdanowicz (2012) or the rule based heuristic developed by 276–280.
Xin et al. (2010) were created to solve the WTA, efficiently exploit- Bogdanowicz, Z.R., Tolano, A., Patel, K., Coleman, N.P., 2013. Optimization of
weapon–target pairings based on kill probabilities. IEEE Trans. Cybern. 43 (6),
ing the special structure of the problem. 1835–1844.
The only consistent aspect of the WTA since its introduction Burr, S.A., Falk, J.E., Karr, A.F., 1985. Integer prim-read solutions to a class of target
into the field is its enduring relevance. As defensive strategies im- defense problems. Oper. Res. 33 (4), 726–745.
Çetin, E., 2007. A queuing theoretical model for anticancer tool selection.. In: Inter-
prove to enhance the capacity to mitigate the risk that ballistic
national Mathematical Forum, 2, pp. 2675–2685.
missiles present, the technology of these ballistic missiles also im- Çetin, E., Esen, S.T., 2006. A weapon–target assignment approach to media alloca-
proves. Additionally, while only minimally addressed here, many tion. Appl. Math. Comput. 175 (2), 1266–1275.
Chang, S.-c., James, R.M., Shaw, J.J., 1987. Assignment algorithm for kinetic energy
non-defensive applications will continue to benefit from the lively
weapons in boost phase defence. In: 26th IEEE Conference on Decision and Con-
research surrounding the Weapon Target Assignment Problem. trol, 1987, 26. IEEE, pp. 1678–1683.
Chen, J., Xin, B., Peng, Z., Dou, L., Zhang, J., 2009. Evolutionary decision-makings
References for the dynamic weapon-target assignment problem. Sci. China Ser. F 52 (11),
2006–2018.
Ahner, D.K., Parson, C.R., 2015. Optimal multi-stage allocation of weapons to targets Day, R.H., 1966. Allocating weapons to target complexes by means of nonlinear pro-
using adaptive dynamic programming. Optim. Lett. 9 (8), 1689–1701. gramming. Oper. Res. 14 (6), 992–1013.
Ahuja, R.K., Kumar, A., Jha, K.C., Orlin, J.B., 2007. Exact and heuristic algorithms for denBroeder, G., Ellison, R., Emerling, L., 1959. On optimum target assignments. Oper.
the weapon-target assignment problem. Oper. Res. 55 (6), 1136–1146. Res. 7 (3), 322–326.
Alighanbari, M., 2004. Task Assignment Algorithms for Teams of UAVs in Dynamic Eckler, A.R., Burr, S.A., 1972. Mathematical Models of Target Coverage and Missile
Environments. Massachusetts Institute of Technology. Allocation. Technical Report AD-A953 517. Military Operations Research Society,
Bertsekas, D.P., Homer, M.L., Logan, D.A., Patek, S.D., Sandell, N.R., 20 0 0. Missile de- Alexandria, VA.
fense and interceptor allocation by neuro-dynamic programming. IEEE Trans. Esen, Ö., Çetin, E., Esen, S.T., 2008. A mathematical immunochemoradiotherapy
Syst. ManCybern. Part A 30 (1), 42–51. model: a multiobjective approach. Nonlinear Anal. Real World Appl. 9 (2),
Bertuccelli, L.F., How, J.P., 2011. Active exploration in robust unmanned vehicle task 511–517.
assignment. J. Aerosp. Comput. Inf.Commun. 8 (8), 250–268. Fu, T.-p., Liu, Y.-s., Chen, J.-h., 2006. Improved genetic and ant colony optimiza-
Boardman, N.T., Lunday, B.J., Robbins, M.J., 2017. Heterogeneous surface-to-air mis- tion algorithm for regional air defense WTA problem. In: Innovative Computing,
sile defense battery location: a game theoretic approach. J. Heuristics 23 (6), Information and Control, 2006. ICICIC’06. First International Conference on, 1.
417–447. IEEE, pp. 226–229.
Bogdanowicz, Z., Coleman, N., et al., 2007. Sensor-target and weapon-target pairings Gelenbe, E., Timotheou, S., Nicholson, D., 2010. Fast distributed near-optimum as-
based on auction algorithm. In: Proceedings of the 11th WSEAS International signment of assets to tasks. Comput. J. 53 (9), 1360–1369.
236 A. Kline, D. Ahner and R. Hill / Computers and Operations Research 105 (2019) 226–236

Golany, B., Goldberg, N., Rothblum, U.G., 2015. Allocating multiple defensive re- Li, P., Wu, L., Lu, F., 2009. A mutation-based GA for weapon-target allocation prob-
sources in a zero-sum game setting. Ann. Oper. Res. 225 (1), 91–109. lem subject to spatial constraints. In: International Workshop on Intelligent Sys-
Golany, B., Goldberg, N., Rothblum, U.G., 2017. A two-resource allocation algorithm tems and Applications, 2009. IEEE, pp. 1–4.
with an application to large-scale zero-sum defensive games. Comput. Oper. Res. Li, Y., Kou, Y., Li, Z., 2018. An improved nondominated sorting Genetic Algorithm III
78, 218–229. Method for solving multiobjective weapon-target assignment Part I: the value
Gülpınar, N., Çanakoğlu, E., Branke, J., 2018. Heuristics for the stochastic dynamic of fighter combat. Int. J. Aerosp. Eng. 2018.
task-resource allocation problem with retry opportunities. Eur. J. Oper. Res. 266 Li, Y., Kou, Y., Li, Z., Xu, A., Chang, Y., 2017. A modified pareto ant colony optimiza-
(1), 291–303. tion approach to solve biobjective weapon-target assignment problem. Int. J.
Hill, R.R., Pohl, E.A., 2010. Heuristics and Their use in Military Modeling. Wiley En- Aerosp. Eng. 2017.
cyclopedia of Operations Research and Management Science. Lloyd, S.P., Witsenhausen, H.S., 1986. Weapons allocation is np-complete.. In: 1986
Hosein, P.A., 1989. A Class of Dynamic Nonlinear Resource Allocation Problems. Summer Computer Simulation Conference, pp. 1054–1058.
Technical Report LIDS-TH-1922. Massachusetts Inst Of Tech Cambridge Lab For Lu, H., Zhang, H., Zhang, X., Han, R., 2006. An improved genetic algorithm for tar-
Information And Decision Systems. get assignment, optimization of naval fleet air defense. In: The Sixth World
Hosein, P.A., Athans, M., 1989. The Dynamic Weapon-Target Assignment Problem. Congress on Intelligent Control and Automation, 2006, 1. IEEE, pp. 3401–3405.
Technical Report LIDS-P-1887. Laboratory for Information and Decision Systems, Madni, A.M., Andrecut, M., 2009. Efficient heuristic approaches to the weapon-tar-
Massachusetts Institute of Technology. get assignment problem. J. Aerosp. Comput. Inf.Commun. 6, 405–414.
Hosein, P.A., Walton, J.T., Athans, M., et al., 1988. Dynamic Weapon-Target Assign- Malcolm, W.P., 2004. On the Character and Complexity of Certain Defensive Re-
ment Problems with Vulnerable C2 Nodes. Technical Report LIDS-P-1786. Labo- source Allocation Problems. Technical Report.
ratory for Information and Decision Systems, Massachusetts Institute of Tech- Manne, A.S., 1958. A target-assignment problem. Oper. Res. 6 (3), 346–351.
nology. Matlin, S., 1970. A review of the literature on the missile-allocation problem. Oper.
Huang, S.-c., LI, W.-m., 2005. Research of ant colony algorithm for solving target Res. 18 (2), 334–373.
assignment problem. Syst. Eng. Electron. 1, 023. Metler, W.A., Preston, F.L., Hofmann, J., 1990. A Suite of Weapon Assignment Al-
Jian, W., Chen, C., 2015. Sensor-weapon joint management based on improved gorithms for a SDI Mid-course Battle Manager. Technical Report AD-A229 189.
genetic algorithm. In: 34th Chinese Control Conference (CCC), 2015. IEEE, Naval Research Lab Washington DC.
pp. 2738–2742. Murphey, R.A., 20 0 0. An approximate algorithm for a weapon target assignment
Johansson, F., Falkman, G., 2009. An empirical investigation of the static weapon– stochastic program. In: Approximation and Complexity in Numerical Optimiza-
target allocation problem. In: Proceedings of the 3rd Skövde Workshop on In- tion. Springer, pp. 406–421.
formation Fusion Topics, pp. 63–67. Onay, O., et al., 2016. A mathematical approach to neuromarketing: a weapontarget
Johansson, F., Falkman, G., 2010. Real-time allocation of defensive resources to rock- assignment model. Int. J. Acad. Res.Bus. Social Sci. 6 (1), 164–173.
ets, artillery, and mortars. In: 2010 13th Conference on Information Fusion. IEEE, Rosenberger, J.M., Hwang, H.S., Pallerla, R.P., Yucel, A., Wilson, R.L., Brungardt, E.G.,
pp. 1–8. 2005. The generalized weapon target assignment problem. 10th International
Julstrom, B.A., 2009. String-and permutation-coded genetic algorithms for the static Command and Control Research and Technology Symposium.
weapon-target assignment problem. In: Proceedings of the 11th Annual Con- Shalumov, V., Shima, T., 2017. Weapon–target-allocation strategies in multiagent tar-
ference Companion on Genetic and Evolutionary Computation Conference: Late get–missile–defender engagement. J. Guidance Control Dyn. 40 (10), 2452–2464.
Breaking Papers. ACM, pp. 2553–2558. Shan, X., Zhuang, J., 2013. Hybrid defensive resource allocations in the face of par-
Karasakal, O., 2008. Air defense missile-target allocation models for a naval task tially strategic attackers in a sequential defender–attacker game. Eur. J. Oper.
group. Comput. Oper. Res. 35 (6), 1759–1770. Res. 228 (1), 262–272.
Khosla, D., 2001. Hybrid genetic approach for the dynamic weapon-target allocation Shang, G., 2003. Ant colony algorithm for weapon-target assignment problem [j].
problem. In: Battlespace Digitization and Network-Centric Warfare, 4396. Inter- Comput. Eng. Appl. 3, 025.
national Society for Optics and Photonics, pp. 244–260. Shang, G., 2008. Solving weapon-target assignment problems by a new ant colony
Kline, A., 2017. Real-Time Heuristic Algorithms for the Static Weapon-Target Assign- algorithm. In: Computational Intelligence and Design, 2008. ISCID’08. Interna-
ment Problem. Air Force Institute of Technology. tional Symposium on, 1. IEEE, pp. 221–224.
Kline, A., Ahner, D., Lunday, B., 2017. A Heuristic and Metaheuristic Approach to Shang, G., Zaiyue, Z., Xiaoru, Z., Cungen, C., 2007. Immune genetic algorithm for
the Static Weapon Target Assignment Problem. Technical Report COA-01-17. Air weapon-target assignment problem. In: Intelligent Information Technology Ap-
Force Institute of Technology Center for Operational Analysis. plication, Workshop on. IEEE, pp. 145–148.
Kline, A., Ahner, D., Pachter, M., 2017. A Greedy Hungarian Algorithm for the Sikanen, T., 2008. Solving weapon target assignment problem with dynamic pro-
Weapon-Target Assignment Problem. Technical Report COA-02-17. Air Force In- gramming. Independent Res.Projects Appl.Math. 32.
stitute of Technology Center for Operational Analysis. Soland, R.M., 1987. Optimal terminal defense tactics when several sequential en-
Kolitz, S.E., 1988. Analysis of a maximum marginal return assignment algorithm. In: gagements are possible. Oper. Res. 35 (4), 537–542.
27th IEEE Conference on Decision and Control, 1988. IEEE, pp. 2431–2436. Su, M., QIAN, H., WANG, X.-f., 2008. Immune memory-based ant colony algorithm
Kwon, O., Kang, D., Lee, K., Park, S., 1999. Lagrangian relaxation approach to the for weapon-target assignment solution. Comput. Eng. 4, 078.
targeting problem. Nav. Res. Logist. 46 (6), 640–653. Tokgöz, A., Bulkan, S., 2013. Weapon target assignment with combinatorial opti-
Leboucher, C., Shin, H.-S., Siarry, P., Chelouah, R., Le Ménec, S., Tsourdos, A., 2013. A mization techniques. Int. J. Adv. Res.Artif. Intell. 2 (7), 39–50.
two-step optimisation method for dynamic weapon target assignment problem. Wacholder, E., 1989. A neural network-based optimization algorithm for the static
Recent Advances on Meta-Heuristics and Their Application to Real Scenarios. weapon-target assignment problem. ORSA J.Comput. 1 (4), 232–246.
InTech. Wu, L., Wang, H.-y., Lu, F.-x., Jia, P., 2008. An anytime algorithm based on modified
Lee, M.-Z., 2010. Constrained weapon–target assignment: enhanced very large scale ga for dynamic weapon-target allocation problem. In: IEEE Congress on Evolu-
neighborhood search algorithm. IEEE Trans. Syst. ManCybern. Part A 40 (1), tionary Computation, 2008. IEEE, pp. 2020–2025.
198–204. Xin, B., Chen, J., Peng, Z., Dou, L., Zhang, J., 2011. An efficient rule-based constructive
Lee, Z.-J., Lee, C.-Y., 2005. A hybrid search algorithm with heuristics for resource heuristic to solve dynamic weapon-target assignment problem. IEEE Trans. Syst.
allocation problem. Inf. Sci. 173 (1), 155–167. ManCybern. Part A 41 (3), 598–606.
Lee, Z.-J., Lee, C.-Y., Su, S.-F., 2002. An immunity-based ant colony optimization al- Xin, B., Chen, J., Zhang, J., Dou, L., Peng, Z., 2010. Efficient decision makings for
gorithm for solving weapon–target assignment problem. Appl. Soft Comput. 2 dynamic weapon-target assignment by virtual permutation and tabu search
(1), 39–47. heuristics. IEEE Trans. Syst. Man Cybern.Part C (Appl. Rev.) 40 (6), 649–662.
Lee, Z.-J., Lee, W.-L., 2003. A hybrid search algorithm of ant colony optimization Xin, B., Wang, Y., Chen, J., 2018. An efficient marginal-return-based constructive
and genetic algorithm applied to weapon-target assignment problems. In: In- heuristic to solve the sensor-weapon-target assignment problem. IEEE Trans.
ternational Conference on Intelligent Data Engineering and Automated Learning. Syst. Man Cybern. (99) 1–12.
Springer, pp. 278–285. Yanxia, W., Longjun, Q., Zhi, G., Lifeng, M., 2008. Weapon target assignment problem
Lee, Z.-J., Su, S.-F., Lee, C.-Y., 2002. A genetic algorithm with domain knowledge for satisfying expected damage probabilities based on ant colony algorithm. J. Syst.
weapon-target assignment problems. J. Chin. Inst. Eng. 25 (3), 287–295. Eng. Electron. 19 (5), 939–944.
Lee, Z.-J., Su, S.-F., Lee, C.-Y., 2003. Efficiently solving general weapon-target assign- Zeng, X., Zhu, Y., Nan, L., Hu, K., Niu, B., He, X., 2006. Solving weapon-target assign-
ment problem by genetic algorithms with greedy eugenics. IEEE Trans. Syst. ment problem using discrete particle swarm optimization. In: The Sixth World
Man Cybern.Part B (Cybernetics) 33 (1), 113–121. Congress on Intelligent Control and Automation, 2006, 1. IEEE, pp. 3562–3565.
Lemus, F., David, K., 1963. An optimum allocation of different weapons to a target Zhihua, S., Zhu, F., Duolin, Z., 2009. A heuristic genetic algorithm for solving con-
complex. Oper. Res. 11 (5), 787–794. strained weapon-target assignment problem. In: IEEE International Conference
Li, J., Chen, J., Xin, B., Chen, L., 2017. Efficient multi-objective evolutionary algo- on Intelligent Computing and Intelligent Systems, 2009, 1. IEEE, pp. 336–341.
rithms for solving the multi-stage weapon target assignment problem: a com- Zi-fen, L., Xiang-min, L., Jin-jin, D., Jin-zhu, C., Feng-xia, Z., 2011. Sensor-weapon-tar-
parison study. In: Evolutionary Computation (CEC), 2017 IEEE Congress on. IEEE, get assignment based on improved swt-opt algorithm. In: 2011 IEEE 2nd Inter-
pp. 435–442. national Conference on Computing, Control and Industrial Engineering (CCIE), 2.
Li, J., Chen, J., Xin, B., Dou, L., 2015. Solving multi-objective multi-stage weapon tar- IEEE, pp. 25–28.
get assignment problem via adaptive NSGA-II and adaptive MOEA/D: a com-
parison study. In: 2015 IEEE Congress on Evolutionary Computation (CEC). IEEE,
pp. 3132–3139.