Proceedings of the 33rd Chinese Control Conference

July 28-30, 2014, Nanjing, China

Cooperative Targets Assignment of Multi-UCAVs Considering Kill

Probability
BAI Ruiguang1, CHEN Qiushuang1, QUAN Xiongwen1, JI Xiaotao1,2
1. College of Computer and Control Engineering, Nankai University, Tianjin 300071, China
E-mail: [email protected]
E-mail: [email protected]
E-mail: [email protected]
2. Procurement of Military Supplies in Northeast, Shenyang 110026, China
E-mail: [email protected]

Abstract: Cooperative targets assignment of multiple Unmanned Combat Aerial Vehicles (UCAVs) is the basis and key issue of
multi-UCAVs cooperative combat. The kill probability is taken into consideration in cooperative targets assignment problem, for
which a multi-objective MIP model is proposed. The objectives include minimizing the number of launch points taking part in a
mission and minimizing the launch time range. An improved efficacy coefficient method is developed to transform the problem
to a single-objective MILP model. Numerical experiments show the effectiveness of the proposed approach.
Key Words: cooperative targets assignment, multi-UCAVs, kill probability, efficacy coefficient method

is also used in UCAVs-target assignment, which can reduce

1 Introduction the model dimension and improve the load balance.
In modern battles, Unmanned Combat Aerial Vehicles The existing literature mainly focuses on attack efficacy,
(UCAVs) have been widely used for combat purpose. They and obtaining solutions with maximum efficacy or
have many advantages compared to traditional piloted minimum cost, while paying less attention to the kill
aircrafts. However, the situations and missions are more and probability. Wu [4] considered kill probability in the
more complicated in battlefield, and hostile air defense objective function, but cannot guarantee the specified kill
firepower has been enhanced overtime, which bring more probability constraint. The complexity and uncertainty in
challenges for UCAV to approach and hit the targets. It is modern battlefield makes it possible that a mission is failed.
usually impossible for a single UCAV to accomplish a So it is essential that the kill probability be considered in the
complex mission and guarantee that the target is neutralized. UCAVs-target assignment.
For these reasons, multi-UCAVs cooperative combat has In this paper, a multi-objective mixed integer
become popular in modern battlefield. As the basis and key programming model is formulated for the UCAV
issue of multi-UCAVs cooperative combat, multi-UCAVs assignment, aiming at minimizing the number of launch
cooperative targets assignment refers to the problem of points and the time range of launching. The kill probability
assigning UCAVs to targets subject to various constraints, of targets is introduced as constraints. The capacity and
such as targets hitting sequence, time limit and the UCAVs launching intervals of launch points, and the time window of
range, et al, to minimize the total cost in one mission. hitting targets are also considered. An improved efficacy
Cooperative targets assignment of multi-UCAVs is an coefficient method is proposed to transform the model to a
NP-hard combinatorial optimization problem [1], some single-objective MILP model, which is much easier to solve.
effective methods have been proposed in recent literature. 2 Multi-UCAVs Cooperative Targets Assignment
Zhao [2] presented a uniform model and a unified processing
Model
method for UAVs targets assignment. A bilevel
optimization theory was proposed to describe the UCAV
2.1 Problem Description
cooperative mission assignment problem in [3], where the
strike effect and the shortest fly path was treated as the goal In this paper, we consider the assignment of M UCAVs
of the upper and lower level, and the intuitionistic fuzzy to N targets, and decide the number of UCAVs being
bilevel programming theory was adopted to solve the launched from each launch point and which target each
problem. A GA based anytime algorithm was proposed in UCAV hits. For the sake of simplify, we make the following
[4], to effectively assign weapons with time constraints. assumptions:
Huma [5] proposed an improved stable marriage algorithm to z The attributes of UCAVs from the same launch point are
solve threat evaluation and weapon assignment problems, identical, while the attributes of UCAVs from different
which can obtain near optimal solution. Cluster algorithm [6] launch points may be different;
z The time interval of two consecutive UCAVs launching
from the same point should satisfy the time limit;
*
This work is supported by National Natural Science Foundation of
z The flying speed of each UCAV is constant;
China (NSFC) under Grant 71172071 and the Specialized Research Fund
for the Doctoral Program of Higher Education, China (No.
20120031110036).

7573
z One UCAV can hit at most one target, while more than several methods for obtaining the route, such as direct line [7]
one UCAVs may be required by one target to guarantee and Voronoi graph [8]. The former is convenient in
the kill probability; calculation, but having large deviation from actual value;
z The hitting time of one target must be located in a time while the latter is more accurate, but with higher complexity
window, otherwise, the UCAV is of no use for the target; in calculation.
z The coordination of multiple UCAVs that hit the same In this paper, we employed sparse A* algorithm for the
target is not considered; calculation of the routes, which can obtain more precise
The notations are listed as follows: result with less time, and is convenient to deal with the
U i : the capacity of launch point i , that is, the maximum physical constraints of UCAV and threat regions in
number of UCAVs can be launched by point i ; battlefield.
Vi : the speed of UCAV from launch point i ; 2.4 Problem Formulation
Di : the distance limit of UCAV from launch point i ; Cooperative targets assignment of multi-UCAVs is
' i : the limit of time interval of two consecutive UCAVs formulated as follows:
M
launched from point i ;
min ¦ yi (1)
pij : the kill probability of one UCAV from launch point i 1

i hitting target j ;
­° ­ N ½½°
d ij : the distance between launch point i and target j ; min ®max (t ik )  min ®t ik | ¦ xikj 1¾¾ (2)
°̄ i , k i,k
¯ j 1 ¿°¿
L j : the kill probability designated to target j ;
[ A j , B j ] : the time window required by hitting target j ; Subject to:
H : a large constant;
N Ui
Decision variables:
­1 if kth UCAVfromlaunchpointi is assignedto targetj,
¦¦ x
j 1 k 1
ikj d U i , i 1,, M (3)
xikj ®
¯0 otherwise; N

yi
­1 if launch point i takes part in the mission,
®
¦x
j 1
ikj d 1, i 1,  , M , k 1, , U i (4)
¯0 otherwise;
t ik the launch time of kth UCAV from launch point i . Di  (1  xikj ) H t d ij ,
(5)
2.2 Objective Functions and Constraints i 1, , M , k 1, , U i , j 1, , N

When using UCAVs for combat purpose, cost is not the M Ui

most important factor. The kill probability of all targets 1  –– (1  pij xikj ) t L j , j 1,  , N (6)
i 1 k 1
should be satisfied first. In addition, the capacity of launch
points, distance limits of UCAVs, time windows of targets
d ij
should be considered. We should decide the number of A j  (1  xikj ) H d t ik  d B j  (1  xikj ) H ,
UCAVs from each launch point and which target each Vi (7)
UCAV hits. The main objective is to minimize the number i 1, , M , k 1,  , U i , j 1,  , N
of launch points taking part in the mission and the time
range. The objective functions are as follows: t i ( k 1)  t ik t ' i 
z To minimize the number of launch points that take part
§ N · § N ·
M
¨1  ¦ xi ( k 1) j ¸ ˜ H  ¨1  ¦ xikj ¸ ˜ H , (8)
in the mission: min ¦ yi . By centralizing mission on ¨ j 1 1 ¸ ¨ j 1 2¸
i 1
© 1 ¹ © 2 ¹
as least launch points as possible, abundant firepower i 1,  , M , k 1,  , U i  1,
is reserved, which can enhance the ability of coping
N N
with unexpected missions;
z To minimize the launching time range of all launched ¦x
j 1
ikj t ¦ xi ( k 1) j , i 1,  , M , k
j 1
1,  ,U i  1 (9)
­° ­ N ½½°
UCAVs: min ®max (t ik )  min ®t ik | ¦ xikj 1¾¾ . By Ui N
°̄ i , k i,k
¯ j 1 ¿°¿ yi d ¦¦ xikj d U i yi , i 1, , M (10)
compressing the time range, the time cost is reduced, k 1 j 1

and the mission is more flexible and the performance is

N
enhanced.
t ik d H ¦ xikj , i 1,  , M , k 1,  , U i , (11)
2.3 Route j 1

It is essential that the route of UCAV is obtained before xikj  {0,1}, yi  {0,1}, t ik t 0,
the assignment procedure. Route cost may influence the (12)
assignment result significantly. In the literature, there are i 1, , M , k 1,  , U i , j 1,  , N

7574
 Constraint (3) represents the number of UCAVs being programming problem with different dimensions [9, 10]. In
launched from a launch point doesn’t exceed its capacity; the above model, two objectives are the number of launch
constraint (4) represents one UCAV can hit at most one points and the time range, which are incomparable. In this
target; constraint (5) ensures the distance between a UCAV section, an improved efficacy coefficient method is used to
and the target assigned to it does not exceed the UCAV’s deal with the two objective functions.
distance limit; constraint (6) represents the kill probability
3.1 Solution Procedure
limit of the targets; constraint (7) is the time window of the
targets; (8) is the launching time interval constraint of each Suppose X be the solution vector, S be the feasible
launch point; (9) represent the UCAVs sequence launched M

from one launch point; constraint (10) represent that if a domain of X . Two objective functions are f1 ( X ) ¦y
i 1
i

launch point does not take part in the mission, there is no

UCAV being launched from it; constraints (11)-(12) are and f 2 ( X ) P  Q . First, the lower bound of f1 and f 2 are
binary and non-negative constraints, constraint (11) also obtained respectively,
M
sets t ik to 0 if the kth UCAV of launch point i is not
f1 min ¦ yi , f 2 min( P  Q ) (18)
X S X S
launched. i 1

The main contribution of this formulation is the

The upper bound of f1 is usually limited by the number of
introduction of kill probabilities of the targets as constraints
while simultaneously minimizing the number of launch launch points; and the following constraint is required for
points and the launching time range. The kill probabilities of the upper bound of f 2 :
the targets are satisfied by multi-UCAVs hitting identical
target; meanwhile the firepower is reserved for unexpected ­ P d t ik , i, k
missions. °
®Q t t  §¨1  x ·¸ H , i, k
N
(19)
2.5 Formulation Transformation ° ik ¨ ¦ ikj ¸
¯ © j 1 ¹
Above formulation contains several nonlinear items,
which are undesirable to deal with. In this sub section, we Then the upper bound of f1 and f 2 are obtained as
will transform them to linear ones. follows:
Let P be the maximum of all t ik , Q be the minimum of M

­ N ½ f1 max ¦ yi , f 2 max( P  Q ) (20)

X S X S
all t ik , P min ®t ik | ¦ xikj 1¾ , then
max{t ik } , Q i 1
i ,k i , k
¯ j 1 ¿ The efficacy coefficient of f1 and f 2 are obtained as
objective function (2) is transformed to:
follows:
min P  Q (13)
Subject to: f1  f1 f2  f2
d1 , d2 (21)
­ P t t ik f1  f1 f2  f2
°
®0 d Q d t  §¨1  x ·¸ H
N
In the traditional efficacy coefficient method, geometry
° ik ¨ ¦ ikj ¸ (14) average is treated as the synthetic efficacy coefficient,
¯ © j 1 ¹ 2
i 1,  , M , k 1,  , U i , max O –d
j 1
j . Nonlinear issue in objective function is not
Taking the logarithm of both sides of constraint (6), it is
obtained: good for the solving procedure. In this paper, we proposed
M Ui the total weighted efficacy coefficient as an improved
¦¦ ln(1  p
i 1 k 1
ij xikj ) d ln(1  L j ), j 1,  , N (15) synthetic efficacy coefficient, as follows:
2

If xikj 1 , then ln(1  pij xikj ) ln(1  pij ) ; O ¦D di 1

i i (22)
If xikj 0 , then ln(1  pij xikj ) ln 1 0 . where D i is the weight assigned to d i .
So we can get, for either xikj 1 or xikj 0, The objective function is:
2
ln(1  pij xikj ) ln(1  pij ) ˜ xikj (16) max O max ¦ D i d i (23)
So constraint (6) is transformed into (17): i 1

M Ui Then the initial model is transformed to as follows:

¦¦ >ln(1  p @
ij ) ˜ xikj d ln(1  L j ), j 1, , N (17) 2
i 1 k 1 max O
X S
¦D d
i 1
i i
Then the formulation (1)-(12) is transformed to a mixed (24)
integer linear programming (MILP) model. ­(3)  (5), (7)  (12),
s.t.®
3 Improved Efficacy-Coefficient-Method-Based ¯(14), (17)
Multi-objective Approach The optimal solution of this MILP model is the (inferior)
effective solution of the initial model.
Compared with other approaches, efficacy coefficient
method is good at dealing with multi-objective

7575
3.2 Effectiveness and Equivalence of Solution Table 2: Parameters of Targets

Lemma 1 [10]: If every efficacy coefficient d i is the Aj Bj

Index Lj
(strictly) monotone function of the objective function /min /min
f i ( X )(i 1,2, , p) , then the synthetic efficacy coefficient 1 0.98 25.2 25.5
O ( F ( X )) is the (strictly) monotone function of 2 0.98 27.6 27.9
F ( X ) ( f1 ( X ), f 2 ( X ), , f n ( X )) . 3 0.98 27 27.3
Theorem 1: If the synthetic efficacy coefficient 4 0.98 27.6 27.9
O ( F ( X )) is the (strictly) monotone function of F ( X ) , then 5 0.98 24.6 24.9
the optimal solution of (24) is the (inferior) effective
6 0.98 24.6 24.9
solution of the initial model.
Proof omitted. 7 0.98 29.4 29.7
8 0.98 28.2 28.5
4 Numerical Experiments
9 0.98 24.6 24.9
In this section, we obtain the optimal solutions of
10 0.98 27 27.3
multi-objective model (24) using efficacy coefficient
method. All experiment results are obtained by CPLEX with First, the upper and lower bound of each objective
Visual Studio 2010 on PC with Core i3 3.1GHz CPU and function are calculated respectively. For objective function
M M
2G RAM. (1), we get: f1 max ¦ yi 10 , f1 min ¦ yi 4 ; for
X S X S
i 1 i 1
4.1 Efficacy Coefficient Method Result
objective function (2), we get:
In order to verify the effectiveness of efficacy coefficient
f 2 max( P  Q) 7.764 , f 2 min( P  Q) 4.241 .
method, the optimal solution is obtained using efficacy X S X S

coefficient method and then compared with solutions of According to (21), the efficacy coefficient of f1 and f 2 are
single objective models. M

10 launch points and 10 targets are chosen in a obtained as follows: d1 1.667  0.167¦ yi ,
i 1
300km u 300km region, the distances between launch
points and targets are calculated beforehand using sparse A* d2 2.2037  0.2838( P  Q) . The weights in (22) are set to
algorithm. The launch capacities, distance limits, time D 1 D 2 0.5 , then the optimal solution can be obtained by
intervals of launch points and UCAV’s speeds are given in the improved efficacy coefficient method, f1 5 ,
Table 1.
f 2 4.5606 , the corresponding efficacy coefficients are:
Table 1: Parameters of Launch Points d1 0.8333 , d 2 0.9092 , and the synthetic efficacy
coefficient is: O 0.8713 , good performance is achieved
Di 'i Vi for each objective. The kill probabilities of all targets are
Index Ui
/km /s /km/h 0.99, 0.99, 0.988, 0.996, 0.991, 0.994, 0.992, 0.98, 0.985,
1 5 430 31 1000 0.994, which satisfy the constraints. The computation time
is 30.82 seconds.
2 5 450 31 960
Different weights in (22) are set to obtain more optimal
3 6 440 35 990 solutions, which are listed in Table 3. The first and last line
4 6 420 35 950 in Table 3 are the single objective (1) and (2) respectively.
5 6 400 32 1000
Table 3: Optimal Solutions with Different Weights
6 5 410 39 990
7 7 400 37 950 f2
D1 D2 f1
8 7 420 31 960 /min

9 6 420 36 970 1 0 4 7.764

10 5 400 39 950 0.8 0.2 4 5.602

The kill probability limits and time windows of targets are 0.5 0.5 5 4.561
listed in Table 2. The length of time windows of all the 0.2 0.8 7 4.241
targets are set 0.3 minutes (18 seconds). This restricts that 0 1 10 4.241
all the UCAVs hitting identical target arrive in the duration
The Pareto front is shown in Fig.1. It can be seen that, for
of less than half minute, which stands for the time
different weights in (22), both objectives are optimized. For
coordination.
The kill probabilities of UCAVs from different launch weights values of D 1 D 2 0.5 , although neither of the two
points to targets are selected stochastically from four values objectives is achieved optimal value, both of the objectives
of 0.75, 0.8, 0.85 and 0.9. This means that to a specified are considered in the multi-objective model.
target, UCAVs of different types have different kill
probabilities.

7576
 4.2 Parameter Analysis
In this section, we design several instances with different
parameters, such as the number of launch points and targets,
the capacities and time intervals of launch points, the kill
probability limits and the length of time windows of targets,
and so on.
First, we choose different number of launch points and
targets, the other parameters are remaining unchanged. The
capacity of each launch point is 5; it is guaranteed that most
of the distances between launch points and targets are no
more than the distance limits of launch points; the length of
time windows are set to 0.6 minutes. In each instance, the
efficacy coefficient method procedure is implemented and
the optimal solution is obtained. The upper and lower
Fig. 1: Pareto front with different weights bounds of each objective function and the solutions are
shown in Table 4.

Table 4: Results with Different Numbers of Launch Points and Targets

Solution
f2 f2 Time
Index (M , N ) f1 f1 f2
/min /min f1 O /s
/min
1 (6, 4) 4 6 2.58 9.36 4 4.5 0.86 0.33
2 (8, 5) 5 8 3.23 10.04 5 5.7 0.82 7.94
3 (10, 10) 6 10 3.15 10.68 6 4.63 0.9 34.28
4 (10, 10) 5 10 6.51 10.68 5 6.77 0.97 30.82
5 (10, 15) 6 10 2.7 10.56 6 4.14 0.91 55.67
6 (10, 15) 7 10 8.91 12.56 7 10.32 0.81 58.53
7 (10, 20) 6 10 7.32 10.51 7 8.25 0.73 160.91
8 (10, 20) 7 10 8.31 12.75 7 9.27 0.89 120.73
For missions with fewer launch points and targets, of time windows of targets are set; we will study how these
optimal solutions can be obtained in very short time. With parameters affect the solutions.
the number increases, the computation time increases The number of launch points and targets are all set to 10;
correspondingly, but still within reasonable time. The the capacities of launch points are set to 5. The intervals of
computational time of the method with 10 launch points and launch points are set respectively to 30 and 60 seconds; the
20 targets is less than 3 minutes, which shows good time length of time windows are set to 18 and 36 seconds. The
performance to some extent. results are shown in Table 5.
Next, for the same numbers of launch points and targets,
different time intervals of launch points and different length

Table 5: Results with Different Time Parameters

Solution
'i B j  Aj f1
f2 f2 Time
Index f1 f2 /s
/s /s /min /min f1 O
/min
1 18 36 5 10 2.88 8.28 5 3 0.99 91.35
2 18 18 5 10 3.72 7.98 5 4.02 0.96 61.89
3 18 10.8 5 10 4.64 7.86 5 4.64 1 58.41
4 36 36 5 10 3.42 8.28 5 3.84 0.96 62.29
5 36 18 5 10 4.56 7.98 5 4.56 1 47.77
6 36 10.8 5 10 4.64 7.86 5 4.64 1 46.41
7 72 36 5 10 4.32 8.28 5 4.74 0.95 49.96
8 72 18 5 10 4.62 7.98 6 4.8 0.87 26.05
9 72 10.8 5 10 4.74 7.86 6 5.1 0.84 22.88

7577
 The computation time with different intervals of launch 5 Conclusions
points and length of time window of targets is shown in
This paper proposes a multi-objective MIP model of
Fig.2. The value of f 2 in optimal solutions with different
cooperative targets assignment of multi-UCAVs, aiming at
parameters is shown in Fig. 3. The abscissa means the length minimizing the number of launch points and the launch time
of time windows of targets, B j  A j ; and the ordinate means range. In this model, the kill probability of targets, the
the computation time and f 2 in optimal solutions. capacities and launch intervals of launch points, the time
It is shown that, for each identified interval of launch windows of targets are considered. Then an improved
point, if the length of time window of targets increases, the efficacy coefficient method is proposed to transform the
time of solving procedure also increases, and from the value model to a single-objective MILP model, which is much
convenient to be solved by CPLEX. Numerical experiments
of f1 and f 2 , the solution gets better. Conversely, while the
show the influence of parameters on the solving procedure,
length of time window is set to same value, the instances and for the instances with normal scale (no more than 20),
with shorter intervals obtain better solution (smaller values the proposed method can get optimal solution within
of f1 and f 2 ); but the computation time increases. reasonable time.
References
[1] Y. Zhu, T. Zhang, N. Cheng, et al, Research on cooperative
mission planning of multiple UAVs, Journal of system
Simulation, 20(12): 194-199, 2009.
[2] M. Zhao, X. Su, P. Ma, et al, A unified modeling method of
UAVs cooperative target assignment by complex
multi-constraint conditions, Acta Automatica Sinica, 38(12):
2038–2048, 2012.
[3] Y. Liu, W. Li, Q. Xing, et al, Cooperative mission assignment
optimization of unmanned combat aerial vehicles based on
bilevel programming, Systems Engineering and Electronics,
32(3): 579-583, 2010.
[4] L. Wu, H. Wang, F. Lu, et al, An anytime algorithm based on
modified GA for dynamic weapon-target allocation problem,
in 2008 IEEE Congress on Evolutionary Computation, 2008:
Fig. 2: Computation time with different parameters 2020-2025.
[5] N. Huma, M. Asif, An optimal dynamic threat evaluation and
weapon scheduling technique, Knowledge-Based Systems,
2010(23): 337-342.
[6] S. Tang, Z. Qin, J. Xin, Collaborative task assignment
scheme for multi-UAV based on cluster structure, in 2010
Second International Conference on Intelligent
Human-Machine Systems and Cybernetics, 2010: 285-289.
[7] P. Ma, Y. Mao, H. Zhang, et al, Cooperative planning for
multiple trajectories with multiple constraints based on
3DSAS, Systems Engineering and Electronics, 2011, 33(7):
1527-1533.
[8] T. Long, Research on distributed task allocation and
coordination for multiple UCAVs cooperative mission
control, in National University of Defense Technology, China,
2006.
[9] C. Wang, Z. Ping, Efficacy coefficient method for interval
Fig. 3: Launching time range with different parameters multi-objective linear optimization, Journal of Beijing
On the other hand, launch points with less time intervals University of Aeronautics and Astronautics, 2013, 39(7):
make it possible that more UCAVs being launched from a 907-916.
launch point and within less time range; similarly, targets [10] G. Zhou, Efficacy coefficient method for multi-objective
with wider time windows provide more flexibility for optimal problem, Journal of University of Shanghai for
Science and Technology, 1987, 9(3): 103-108.
Multi-UCAVs hitting, which is beneficial for better
solutions, but also enlarge the feasible region of the model
and increase the computation time.

7578