Professional Documents
Culture Documents
Evasive Maneuver Strategy For UCAV in Beyond-Visua
Evasive Maneuver Strategy For UCAV in Beyond-Visua
Evasive Maneuver Strategy For UCAV in Beyond-Visua
fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2017.DOI
ABSTRACT This study deals with the autonomous evasive maneuver strategy of unmanned combat
air vehicle (UCAV), which is threatened by a high-performance beyond-visual-range (BVR) air-to-air
missile (AAM). Considering tactical demands of achieving self-conflicting evasive objectives in actual
air combat, including higher miss distance, less energy consumption and longer guidance support time,
the evasive maneuver problem in BVR air combat is defined and reformulated into a multi-objective
optimization problem. Effective maneuvers of UCAV used in different evasion phases are modeled in three-
dimensional space. Then the three-level decision space structure is established according to qualitative
evasive tactical planning. A hierarchical multi-objective evolutionary algorithm (HMOEA) is designed
to find the approximate Pareto-optimal solutions of the problem. The approach combines qualitative
tactical experience and quantitative maneuver decision optimization method effectively. Simulations are
used to demonstrate the feasibility and effectiveness of the approach. The results show that the obtained
set of decision variables constitutes nondominated solutions, which can meet different evasive tactical
requirements of UCAV while ensuring successful evasion.
INDEX TERMS BVR air combat, evasive maneuver, hierarchical evolutionary algorithm, multi-objective
optimization, UCAV.
VOLUME 4, 2016 1
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
Imado et al. [22], [23] made a comparative study of several optimization algorithm, respectively. There were also some
evasive maneuvers against the proportional navigation mis- artificial intelligence methods such as neural networks [28]
sile. It was shown that each maneuver had its advantageous and reinforcement learning [29] applied to solve the prob-
region and the evasion strategy should change depending on lem. In addition, a more innovative approach involves decoy
relative geometry relationship. Some features of the high- deployment [30] and protecting fighter by using a defender
g barrel (HGB) roll and split-S maneuvers, and the effects missile (see Ref. [31], [32]).
of the parameters on the miss distance (MD) were studied The aforementioned literature mostly deals with a sit-
through mathematical simulation [24], [25]. On this basis, uation, in which the fighter is already located within the
Akdag et al. [26], [27] analyzed the evasive effects between missile’s guaranteed kinetic-capture zone (GKZ), which was
the maneuver combinations of Immelmann followed by a called "endgame evasion" in [33]. The above approaches
HGB and split-S followed by a HGB. have achieved satisfactory evasion effects in this situation,
To obtain the analytic solution of optimal maneuver, the but they are not fully applicable to evade BVR AAM, which
problem could be formulated as a two-sided optimization often lanches as the opponent is far from the GKZ. The latter
called a zero-sum two-person differential game [3], [6]. It studies focus more on recommending to the fighter how to
was solved by semidirect collocation with nonlinear pro- apply tactical planning rather than dynamic advantages.
gramming in [3]. Ryan et al. [4] obtained an initial guess The experience and knowledge summarized from numer-
for the physically unintuitive costates through solving a one- ous air combat cases and widely accepted among pilots
sided optimal control problem. This study was extended in are very effective for this problem, but they were rarely
[5] to a family of local solutions, and Alkaher et al. [6] also reflected in traditional methods. What’s more important is
solved this game through an optimal controller in a closed- that BVR evasion usually has multiple tactical objectives
loop form, which resulted in a safe-navigation strategy for in actual air combat, not just MD. For example, prolong-
the fighter. ing the guidance support time (GST) means increasing the
Optimal control theory is also an effective solution in the probability of hitting the target. Maximizing the MD means
evasion problem [7]–[11]. Imado et al. [7], [8] proposed increasing the survival probability [34]. Besides, decreasing
some practical approaches to the construction of suboptimal energy consumption (EC) means UCAV has more energy
strategies of the behavior of players for different data on for subsequent combat after a successful evasion. So this
the choice of controls by their opponents. This study was problem that involves three self-conflicting objectives has no
extended in [9] to evade two missiles simultaneously, which unique optimal solution but rather a set of Pareto-optimal
based on the steepest ascent method. Ong et al. [10], [11] strategies [35], and strategy choice should be decided by
treated the evasion problem as an approximated parameter UCAV’s mission commands or current battlefield situation.
optimization problem, and sequential quadratic programming To solve the aforementioned problem, a hierarchical multi-
was used to solve it. objective evolutionary algorithm (HMOEA) is designed in
The closed-loop method based on optimal control is af- this paper, which combines qualitative tactical experience
fected by the disaster phenomenon of dimensionality. In and quantitative maneuver decision optimization method.
general, it is only suitable for highly simplified problems, Firstly, three-dimensional maneuver models are presented for
and this motivates other studied approaches. Karelathi et different intentions in different BVR evasion phases. Then
al. [12] presented a receding horizon control scheme for the three-level decision space structure is established accord-
obtaining near-optimal controls in a feedback form. Further- ing to qualitative evasive tactical planning, which includes
more, online identification method of the guidance law of timing, type, and parameters of evasive maneuvers. Next,
the missile for the control scheme was addressed in [13]. the optimization model of the evasive maneuver strategy for
A similar problem was solved in [14], [15] by nonlinear UCAV in BVR air combat is established based on the three
model predictive control, which was another strategy for objective function models and corresponding constraints.
computing near-optimal feedback control based on current Finally, a redesigned multi-objective evolutionary algorithm
state information. based on decomposition (MOEA/D) [36] is proposed to solve
Some scholars have tried to use the intelligent optimiza- the model through the design of normalized decomposition
tion algorithm to solve the global optimal solution of the and hierarchical evolutionary mechanism. The simulation
problem, because it opens up the possibility to search for an results demonstrate the strength and practicability of the
optimal solution with the presence of nonlinearity, parameter proposed method. It also effectively overcome defects that
discontinuity, and discrete input [16]. Moore et al. [17], the decision-making period is difficult to determine and the
[18] described the implementation of a genetic programming result is difficult to explain in traditional methods. To the best
system that evolved optimized solutions to the pursuer/evader of our knowledge, this is the first time the problem of evasive
problems. Nusyirwan et al. [16], [19] proposed a technique strategy in BVR air combat is addressed in the open literature
using a parallel evolutionary algorithm to search for optimal using the HMOEA and corresponding mathematics models.
control for an evasive fighter. A similar scheme was applied The rest of this paper is organized as follows. In Section II,
in [20] and [21], which used a co-evolutionary augmented the evasive maneuver problem in BVR air combat is defined
lagrangian method and quantum-behaved particle swarm and reformulated into the multi-objective optimization prob-
2 VOLUME 4, 2016
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
lem. The details of the HMOEA are presented in Section III. phase Pg (i.e., GST) and reducing extrapolation time of the
The proposed models and algorithm are demonstrated with BM can effectively decrease θl , which thereby increase the
simulation experiments in Section IV, followed by conclu- hit probability of the BM. Ideally, BM’s range to xr (t2 ) is
sion and future work in Section V. less than dl at t1 . Then BM does not need to estimate the
RA’s information, so the hit probability is relatively high.
II. PROBLEM FORMULATION But it also means reducing the time spent in the phase Pe ,
Dogfight emphasizes the game of space angle. However, which will results in significantly increasing the threat of RM
BVR air combat usually starts from a head-on geometry with and reducing the survival probability of UCAV. Conversely, if
a large distance. The condition of space angle for launching UCAV starts the phase Pe early, which will results in higher
missile is easier to meet, and there is more of an emphasis survival probability and lower hit probability. Therefore, the
on the distance game. As we all know, principles of first choice of t1 reflects the different tactical tendencies of UCAV.
detection, first shot, first hit and first leave are the key to Besides, during the phase Pe , when and what evasive maneu-
improve both the air superiority and survival probability in ver is adopted, as well as the size of maneuver parameters
BVR air combat. Therefore, the problem of evasive strategy are also the core of evasive strategy. It will affect the MD
in BVR air combat not only need to keep the aircrafte from and also lead to different levels of EC. Less EC means
being hit, but also need to consider the whole battle efficiency more energy is available for subsequent air combat after a
and tactical superiority. successful evasion, thereby creating more tactical superiority
for UCAV.
A. PROBLEM ANALYSIS Accoring to the aforementioned analysis, the evasive ma-
This paper considers a common scenario of one-on-one BVR neuver problem in BVR air combat is defined and refor-
air combat as shown in Fig. 1. The UCAV launches the blue mulated into a multi-objective optimization problem in this
BVR AAM (abbreviated as BM) toward the red enemy air- paper, which includes multi-dimensional decision variables.
craft (abbreviated as RA) at tb0 . Note that RA also launches The flowchart of the solution approach is shown in Fig. 2.
the red BVR AAM (abbreviated as RM) at tr0 . It is assumed It includes the evasion model (as illustrated in the left-side
that both sides launch missile at approximately the same panel of Fig. 2) and the optimization model (as illustrated in
time, namely tb0 ≈ tr0 . UCAV starts the phase of guidance the right-side panel of Fig. 2). The evasion model implements
support Pg immediately after the launch. The BM obtains the following steps: establishing dynamics and constraints of
RA’s information through an uplink from the UCAV which UCAV and missile, according to the qualitative evasive tac-
tracks the RA. At t1 , UCAV finishes supporting BM and tical planning to design the initial decision space. The opti-
starts the evasive phase Pe to evade RM. Then, the BM stops mization model implements the following steps: hierarchical
receiving RA’s information. Therefore, BM extrapolates the initialization of the decision space, generating quantitative
RA’s trajectory with the latest available RA’s information by evasive tactical planning, solving the objective space (i.e.,
assuming that the RA flies at constant velocity and direction. MD, EC, and GST) by digital simulation of evasive behavior,
At t2 , the BM reaches the maximal lock-on distance dl to then a multi-objective hierarchical evolutionary algorithm is
the extrapolated position of the RA, whereupon it switches designed to find the approximate Pareto-optimal strategy set
on its own radar with an attempt to lock on the RA. At that of evasive maneuver.
moment, there is an angle of tracking error θl between the Before modeling, for the scenario in Fig. 1, several as-
true and estimated line-of-sight from BM to the real xr (t2 ) sumptions that simplify the problem to a certain extent with-
and estimated position x0 r (t2 ) of RA (see Fig. 1). θl is an out losing practicability are listed:
important factor that affects the intercept and hit probability 1. Both sides can obtain real-time status information of each
of BVR AAM [34]. During the phase Pe , UCAV tries to other and missiles they launch.
maximize the MD at the terminal time tf by using a series 2. Both sides launch the missile when the opponent is far
of evasive maneuvers. MD is the key to ensuring the survival from the respective GKZ.
probability of UCAV. 3. During the phase of guidance support, the UCAV close to
the RA using initial velocity and direction.
tf MD tf 4. There is no consideration given to the case of adopting
Enemy Red Pe active or passive jamming measures.
aircraft missile
(red) t1 Pg
t1 B. "MISSILE-UCAV" MODELING
tr 0 tb 0
1) Relative Motion Model
UCAV
t1 l Blue
Firstly, the relative motion relationship between the missile
t2 (blue)
missile
x r (t2 ) dl and UCAV is established in three-dimensional space, as
xr (t2 )
shown in Fig. 3.
FIGURE 1: Scenario of one-on-one BVR air combat In Fig. 3, Oxg yg zg , O0 x0g yg0 zg0 , O00 x00g yg00 zg00 , O0 xh yh zh ,
and O0 xl yl zl refer to the geographic coordinate system, the
As can be seen from the above process, prolonging the concomitant inertial coordinate system of the missile, the
VOLUME 4, 2016 3
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
y (km)
set, a simplified model represented by constrained overload 6
0
is used to solve the evasive trajectory of UCAV in this paper. 4 x (km)
The motion of the UCAV is described by -5
20
22.5
ẋu = Vu cos θu cos φu -10
z (km) 25
ẏ = V sin θ
u u u
ż = V cos θ sin φ
u u u u FIGURE 4: Maneuver trajectories of turning type
(10)
V̇ u = (n x − sin θu )g
θ̇ = (ny − cosθu )g/Vu To reduce the closing velocity with the incoming missile
u
ϕ̇u = nz g/(Vu cos θu ) and improve the turning efficiency, UCAV usually turns with
deceleration which is controlled by
where nx , ny and nz are control vectors of the UCAV, which
are the overload component of the corresponding axis of the
−nxd , Vu > Vud
UCAV in the trajectory coordinate system. The overload rate nxc = (12)
sin θu , Vu ≤ Vud
are |ṅx | =a1 and |ṅy | = |ṅz | = a2 , where a1 and a2 are
constant. where Vud , nxc , and nxd are velocity command, actual over-
Equation (10) is constrainted by load control variable in the tangential direction, and overload
command in the tangential direction, respectively. Note that
|nx | ≤ nx max
p 2 nxd ∈ [0 , nx max ] and nxc switches with a rate of change
ny + nz 2 ≤ nn max as nxc =nx ± a1 · ∆t. Similarly, actual control variable of
(11)
y ≤ yu ≤ yu max
u min overload in the pitch and yaw channels, which are denoted
Vu min ≤ Vu ≤ Vu max as nyc and nzc , respectively, switch with the overload rate a2
where nx max and nn max are the maximum available over- (similarly hereinafter).
load of the UCAV in the tangential and normal direction, The out maneuver refers to the horizontal turn with a con-
respectively. yu min and yu max are safe flight altitude bound- stant or varying radius of curvature, which turning efficiency
ary of the UCAV, Vu min and Vu max are safe flight velocity is relatively low but EC is preferable. Its control method in
boundary of the UCAV. the yaw channel is given as
4) Evasive Maneuver Models nzd , |φu − φu0 | < φud
nzc = (13)
Evasive maneuver models of the UCAV are established ac- 0, |φu − φu0 | ≥ φud
cording to the characteristics of BVR air combat in this where nzd , φu0 , and φud are overload command in the yaw
section. Refer to effective qualitative tactics summarized channel, initial heading angle, and turning angle command,
from a large number of air combat cases, evasive maneuvers respectively. Besides, the UCAV keeps nyc = 1 and θu = 0
commonly used in BVR air combat boil down to four types during the turn.
in this paper, include turning, altitude, period, and terminal. The abort maneuver keeps turning back to the tail with
1. Turning type an overload in the normal direction, which aims to reduce
Turning type maneuvers are generally used in the initial eva- the turning radius and improve turning efficiency. So it will
sive phase with a head-on situation, of which tactical purpose reduce the altitude and velocity of the UCAV. The control
is minimizing the closing velocity with the incoming missile method of the maneuver in the yaw channel is the same as
and changing the interception geometry from a head-on to a (13). To realize horizontal flight after finishing the command
VOLUME 4, 2016 5
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
φud , the control method of the abort maneuver in the pitch It’s worth noting that if the incoming missile is relatively
channel is designed as far away, the UCAV can evade the missile through a horizon-
−nyd +cosθu , |φu − φu0 | < φud /2 tal flight with a constant or increased velocity, i.e., altitude
nyc = nyd + cosθu , |φu − φu0 | ≥ φud /2 ∧ θu < 0 hold. In this case, UCAV just needs to keep nyc = 1 and
1, |φu − φu0 | ≥ φud /2 ∧ θu ≥ 0 nzc = 0.
(14) The L-Dive maneuver refers to dive with a constant flight
where nyd is overload command in the pitch channel, and path angle. Firstly, The UCAV starts the dive from horizontal
cosθu is a gravity compensation term. To reduce the com- flight by control the nyc . When the preset flight path angle
plexity of the following optimization algorithm, the com- θud is reached, the UCAV starts the straight dive phase, then
mand is given as nyd =nzd . pull-out in advance before reaching the desired altitude Hd .
The split-S maneuver consists of a half-roll followed by Finally, the L-Dive maneuver ends up with a horizontal flight.
a tight vertical loop downwards, which begins with rolling The control method of the maneuver in the pitch channel is
180deg, once upside down, pulling back on the stick to designed as
execute a vertical U-turn [26]. That can minimize closing
velocity with the incoming missile to the maximum extent
−nyd + cos θu , H > Hd + ∆H ∧ |θu − θu0 | < θud
cos θu , H > Hd + ∆H ∧ |θu − θu0 | ≥ θud
possible and quickly achieve the tail-chase intention. The nyc =
control method of the split-S maneuver in the pitch channel
nyd + cos θu , H ≤ Hd + ∆H ∧ θu < 0
1, H ≤ Hd + ∆H ∧ θu ≥ 0
is designed as
(17)
−nyd +cosθu , θu > −π where θu0 is the initial flight path angle.∆H = (1 − cos θd ) ·
nyc = (15)
1, θu ≤ −π Rd is the altitude loss during pull-out, where Rd is the radius
Besides, the UCAV keeps nzc = 0 during the maneuver. of curvature of pull-out. Rd is simplified to a constant in this
2. Altitude type paper, i.e., Rd = Vu 2 /g · nyd . Besides, the UCAV keeps
Altitude type maneuvers refer to descend the flight altitude nzc = 0 during the maneuver.
by diving under a safe altitude condition, of which tactical The S-Dive maneuver is the combination of dive and
purpose is to make the incoming missile to dive to the horizontal S-shaped maneuvers, that is, changing the above
ground and increasing Dm through greater air density in straight dive mode to S-shaped dive. In the control method,
lower altitudes. Thereby consuming the missile’s energy let nzc =nzd · cos(ωd1 · t), where ωd1 is the angular velocity
more effectively. Moreover, the missile’s seeker would also in the S-Dive maneuver. nzc is still equal to zero in the other
be disturbed by ground clutter to some extent in this way [34]. phase.
However, this also means the reduction of energy and tactical 3. Period type
superiority in BVR air combat. Altitude type maneuvers
include altitude hold, straight dive (abbreviated as L-Dive), Period type maneuvers refer to generating the horizontal
and S-shaped dive (abbreviated as S-Dive) in this paper, S-shaped (abbreviated as S-Plane) or barrel roll trajectory
corresponding maneuver trajectories are demonstrated in Fig. through changing the overload direction periodically, corre-
5. sponding maneuver trajectories are demonstrated in Fig. 6.
Such maneuvers force the incoming missile to constantly
maneuver in different directions. Because of the missile’s
Altitude Hold
L-Dive pursuit mode, it is necessary to re-establish the lead pursuit
S-Dive every time when the target changes course, which greatly in-
creases the missile’s flight distance and consumes its kinetic
10
energy. Moreover, that will accumulate the missile’s delay
0
error continuously, thus generating a larger dynamic error
8
and increasing the MD.
y (km)
-5
6 Similar to the situation of altitude hold, if there is enough
-10
4 -15 x (km) distance to evade the incoming missile, the constant overload
20 -20 maneuvers could be chosen to decrease EC, i.e., flying with
21 constant velocity (CV) or constant acceleration (CA). The
22 -25
z (km)
velocity control method is similar to (16). Besides, let nyc = 1
FIGURE 5: Maneuver trajectories of altitude type and nzc = 0.
The S-Plane maneuver is a periodic turning maneuver with
The motion with constant or increased velocity is gener- the S-shaped trajectory in a horizontal plane. The control
ally used in altitude type maneuvers, which velocity control method of the maneuver in the yaw channel is given as
method is similar to (12) nzc =nzd · cos(ωd2 · t), where ωd2 is the angular velocity in
the S-Plane maneuver. The UCAV keeps nyc = 1 and θu = 0
nxd , Vu < Vud
nxc = (16) during the maneuver.
sin θu , Vu ≥ Vud
6 VOLUME 4, 2016
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
8.5
6 types of maneuvers because of the different initial situations.
8 4 It is selected by a certain priority. In this paper, the tactical
18 2 maneuver priority is set according to the tactical require-
20 x (km) ments and maneuver characteristics. The highest priority is
0
z (km) 22 the terminal type, then followed by the turning type, the
23 -2
altitude type, and the period type in turn. On the other hand,
FIGURE 7: Maneuver trajectories of terminal type the intensity of evasive maneuver depends on the threat level
of the missile (determined by the relative distance and mis-
The control method of the last-ditch maneuver in the yaw sile’s residual energy), and the EC of UCAV is also directly
channel is designed as proportional to the intensity. When the threat level of the
nzc = sgn(ẋu · zr − żu · xr ) · nzd (18) missile is low, the UCAV can choose a relatively moderate
maneuver from each type of maneuver in different evasive
Equation (18) is to turn the UCAV towards the side of the phases, such as the out and the altitude hold maneuver.
incoming missile. Meanwhile, let nyc = 1 during the turn. Therefore, the decision of timing, type, and parameters of
In addition, if the velocity of the incoming missile is low maneuver needs to balance the relationship between MD, EC,
in the final evasive phase, the UCAV can evade the missile and GST. Based on the above qualitative planning of evasive
VOLUME 4, 2016 7
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
tactics, the three-level decision space structure is designed as other objective functions, the objective function of the MD is
Fig. 8. defined as
As shown in Fig. 8, the complex evasive decision problem
1/r(tf ), r(tf ) > dmax
is decomposed hierarchically. The decision space structure J1 = (19)
P, r(tf ) ≤ dmax
generates the evasive timing, maneuver chain, and trajectory
of the UCAV in turn. Moreover, all variables in decision where P is a large constant that represents the punishment
space can be optimized in parallel according to the evasive for failed evasion of the UCAV. Besides, the terminal time tf
objectives and effects. satisfies the following condition:
At the decision level of maneuver timing, when the dis- ṙ(tf )= 0 (20)
tance between the UCAV and the missile is S1 ·r0 , the UCAV
starts to perform turning type maneuvers. That is the moment 2. EC
t1 in Fig. 1, i.e., the UCAV finishes supporting BM and starts In the normal direction, the EC of UCAV is proportional
the phase Pe . The value of S1 represents the length of the to the control overload, i.e., nyc and nzc . In the tangential
UCAV’s guidance support phase, where S1 ∈ [0, 1] and r0 direction, Vu directly reflects the EC of UCAV. Therefore,
is the initial distance when the enemy missile is launched. in this paper, the objective function of EC is modeled as the
Similarly, the desired altitude is S2 · H0 when performing following form:
altitude type maneuvers, where H0 is the initial altitude of the Z tf q
2
UCAV. The period and terminal type maneuvers of UCAV are J2 = (Vu /Vu min ) + nyc 2 + nzc 2 dt (21)
triggered respectively when the distance between the UCAV t0
and the missile are S3 · r0 and S4 · r0 . The value range of S2 , where t0 is the initial time.
S3 , and S4 are the same with that of S1 . 3. GST
At the decision level of maneuver type, K1 , K2 , K3 , and The GST is the flight time of the UCAV from the launch
K4 refer to the sequence number of the maneuver selected of the enemy missile to the distance between them equals
for performing in each maneuver type. For example, K1 = 1 S1 · r0 . In this phase (i.e., the phase Pg in Fig. 1), the UCAV
denotes performing the out maneuver in turning type, K2 = 2 provides guidance information for the missile launched by it-
denotes performing the L-Dive maneuver in altitude type, and self. Therefore, the objective function of GST can be directly
so on. defined as
The decision level of maneuver parameters includes con- 1/ts , ts > 0
J3 = (22)
trol parameters and characteristic parameters. It should be 2/T, ts =0
noted that different maneuver requires different parameters,
where T is the time step of the simulation. Besides, the
and some parameters can be the default, such as default states
interrupt guidance time ts satisfies the following condition:
of nzd and ωd in the L-dive maneuver, default states of nxd ,
nyd , and θud in the S-Plane maneuver, and so on. r(ts ) = S1 · r0 (23)
Moreover, to generate an executable and sequential eva-
sive maneuver chain, it is necessary to ensure the smooth The distance between UCAV and the incoming missile is
transitions between maneuvers. Therefore, the constraints of calculated according to (4). It can be seen that S1 is not
switching maneuvers are given by nyc = 1 and θu = 0, that only the decision variable but also the influence factor of the
is, each maneuver is switched to the next one through a objective space.
horizontal flight. Combined with the above mentioned multi-objective eva-
sive tactical requirements and the objective functions, the
optimization model of evasive maneuver strategy for UCAV
D. OBJECTIVE SPACE
in BVR air combat is defined as
1) Objective function model
In the scenario of Fig. 1, the multi-objective of evasive min J(X) = (J1 (X), J2 (X), J3 (X))
maneuver strategy include: maximizing the MD as much as X= (Si , Ki ,nxd_ij , nyd_ij , nzd_ij , Vud_ij , θud_ij , ωd_ij )
i = 1, 2, ...4; j = 1, 2, 3
possible to increase the survival probability, prolonging the
(24)
GST as much as possible to increase the probability of hitting
where X is the decision vector. To reduce the value range
the target, and decreasing the EC as much as possible for
in decision space and improve optimization efficiency, each
tactical superiority in subsequent combat.
decision variable can be set a reasonable value range based
1. MD on the empirical value. The boundary of the timing and type
The MD is the closest distance between the UCAV and the variable selected for the ith maneuver timing are denoted
missile, namely, the distance when the closing velocity is by Si ∈ [Si min , Si max ] and Ki ∈ {1, 2, 3}, respectively.
zero. For the evasive UCAV, the most basic objective is to Besides, K4 ∈ {1, 2}. The composition and value range
make the MD greater than the missile’s damage radius. The of maneuver parameter variables are determined by maneu-
damage radius of the missile is simplified to a sphere with a ver type variable Ki , so the dimension of X is changing
radius of dmax in this paper. To ensure the consistency with dynamically. nxd_ij is the overload of the UCAV in the
8 VOLUME 4, 2016
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
Feedback
Initial
situations Maneuver
Timing Trajectory
Decision
Decision of
of Decision
Decision of
of chain Decision
Decision of
of
maneuver
maneuver timing
timing maneuver
maneuver type
type maneuver
maneuver parameters
parameters
Tactical
requirements
Decision space
tangential direction, which belongs to the jth maneuver in decision variables. It is difficult to achieve individual evolu-
the ith maneuver timing. The value range of nxd_ij can tion through the traditional strategy of offspring generating.
be preset as nxd_ij ∈ [nxd_ij min , nxd_ij max ] based on Therefore, the MOEA/D is redesigned in this section, a new
experience, other maneuver parameters variables are simi- HMOEA is proposed to solve the problem in this paper.
larly, such as nyd _ij ∈ [nyd _ij min , nyd _ij max ], Vud_ij ∈
[Vud_ij min , Vud_ij max ], and so on. A. NORMALIZED DECOMPOSITION MECHANISM
In the MOEA/D, the number of the subproblems is de-
2) End conditions of evasion let λ = (λ1 , λ2 , ..., λN ) be a weight
noted by PNN , then
According to the established missile and UCAV model, if vector( i=1 λ = 1, λi ≥ 0, i = 1, ..., N ), where λi =
i
given the initial situation information of the missile and the (λi1 , λi2 , λi3 ). According to the decomposition mechanism in
UCAV, whether the UCAV can successfully evade the mis- the MOEA/D, the multi-objective optimization problem is
sile can be determined through simulation. Evasive results decomposed into a series of subproblems min g i (X)(i =
include the following possible scenarios: 1, 2, ..., N ) through the above weight vector. The objec-
1. Failed evasion tive function of each subproblem is the aggregate function
In the simulation, if the situation satisfies r ≤ dmax and other of all objective components. The commonly used aggrega-
conditions (i.e., the missile velocity and working time of the tion function construction approaches include the weighted
missile energy), the missile is believed to have hit the aircraft sum approach, the boundary intersection approach, and the
(i.e., a failed evasion). Besides, if the flight trajectory given tchebycheff approach. The weight coefficient and the penalty
by the evasive strategy causes UCAV to exceed the height coefficient of the first two approaches need to be preset
constraint in (11), the evasion is also failed. through experience or experiment. On this basis, the aggre-
2. Successful evasion gation function is constructed by the tchebycheff approach in
There are also several possible scenarios for UCAV to evade this paper, then the ith subproblem could be defined as
the missile successfully. a) The simulation time t exceeds
min g te (X|λi , z ∗ ) = max {λij |Jj (X) − zj∗ |} (25)
the maximum working time of the missile energy tmax . b) 1≤j≤3
The missile velocity Vm is less than the minimum controlled
where z ∗ = (z1∗ , z2∗ , z3∗ ) is the reference point, i.e., zj∗ =
velocity Vmmin , which will lead to the self-destruction of the
min Jj (X)(j = 1, 2, 3).
missile as out of control. c) In the terminal phase, if r > dmax
In order to reduce the error caused by the dimension differ-
when ṙ = 0. All these scenarios will lead to successful
ence and value range difference of objective functions in this
evasion.
paper. The value of objective function needs to be performed
III. SOLUTION ALGORITHM DESIGN
normalization and mapped to [0, 1], which is calculated by
It is a complex multi-objective optimization problem with the following equation:
(Jj (X) − zj∗ )/(zjmax − zj∗ ), if zjmax 6= zj∗
a multi-modal combination to generate the approximate
Pareto-optimal evasive maneuver strategy set. Studies have J¯j (X)=
1, otherwise
shown that the MOEA/D is effective in dealing with most (26)
multi-objective optimization problems [37]. However, the where zjmax = max Jj (X)(j = 1, 2, 3). To reduce computa-
problem in this paper shows the characteristics of a multi- tional complexity, zj and z̄jmax are used to approximate zj∗
modal combination caused by different physical meanings of and zjmax , respectively, where zj = min {Jj (X)|X ∈ pop}
VOLUME 4, 2016 9
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
and z̄jmax = max {Jj (X)|X ∈ pop}. pop denotes the pop- the following linear transformation is used for the continuous
ulation of current generation. Therefore, (25) could be de- decision variables in pop1 and pop3 :
signed as
xi = xli + u(0, 1)(xui − xli ) (28)
( )
i Jj (X) − zj The following linear transformation is used for the discrete
te i ∗
min g (X|λ , z ) = max λj max (27)
z̄j − zj integer decision variables in pop2 :
1≤j≤3
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
pop1 S1 S2 S3 S4 pop2 K1 K2 K3 K4
Ind1_1 0.6 0.8 0.5 0.2 Ind1_ 2 3 1 2 1
crossover c1 crossover c 2
Ind 2 _1 0.8 0.4 0.4 0.1 Ind 2 _ 2 1 2 3 2
variation p1 p1 ... ... variation p2 p 2 ... ...
Ind N _1 0.7 0.3 0.5 0.1 Ind N _ 2 2 3 1 1
correction correction
When the code of decision variables in a new individual Generate the weight vector
exceed their corresponding boundaries in the process of ge- Start and the neighborhood list
Variables in (31) have the same meanings as that in (28). New Hierarchical Hierarchical crossover operation
evolutionary
subindividuals (i.e., Indc_1 , Indc_2 , and Indc_3 ) created by
such genetic manipulation as crossover, mutation, and modi- Hierarchical mutation operation
Evaluate the
fication make up a complete offspring individual together.
new individual
Hierarchical modification operation
C. ALGORITHM FLOWCHART
Update Update the optimal solutions
The aforementioned normalized decomposition and hierar- mechanism
chical evolution mechanisms combined with the traditional
Update the neighborhood solutions
MOEA/D update mechanisms, such as the optimal solution based on normalized decomposition
No All subproblems
update, the neighborhood solution update, and the external have updated ?
population update [36] constitute the HMOEA framework for Update the external population
solving the evasive maneuver strategy of UCAV in this paper. Yes
The algorithm flowchart is shown in Fig. 10. No Reach the max Yes Output the external population
generation ? (i.e., Pareto solution set)
IV. SIMULATION AND ANALYSIS
To validate the accuracy of models and the effectiveness of End
the approach in this paper, three simulation experiments are FIGURE 10: The flowchart of the HMOEA
performed in this section, and the results are analyzed based
on the process of air combat.
Firstly, all initial parameters used in simulations are given
effectiveness of the proposed approach are further validated
in the following Section IV-A. In Section IV-B, the satisfac-
through several simulations under different initial scenarios.
tory initial scenario is designed, then the rationality of the
scenario and the accuracy of models are tested by simulation.
Then, under the above initial scenario, an ideal strategy set A. SIMULATION PARAMETERS
of evasive maneuvers is obtained by the proposed approach The model parameters of the UCAV and the missile, algo-
in Section IV-C. Finally, in Section IV-D, the feasibility and rithm parameters, and operating environment parameters are
VOLUME 4, 2016 11
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
y (km)
Nm =4, tmax = 120s, nmax = 40, τm = 0.2s, and 9
Vmmin = 400m/s. To fully guarantee the security of the 20
UCAV, the damage radius of the missile is set to a relatively 8 0 x (km)
16
large value, let dmax = 50m.
18 -20
z (km) 20
Parameters of the algorithm are set as following: the max -40
generation, the number of the subproblem, the neighborhood
list size, and the external population size are 2000, 500, FIGURE 11: Three-dimensional trajectories of the UCAV
40, and 200, respectively. Different genetic parameters are and the missile with typical evasive strategy
designed according to the actual characteristics of different
populations, let µc1 =µc3 =1, µc2 =2, p1 =p3 =0.2, p2 =0.1, 1 1200
Vm
µp1 =µp3 =6, and µp2 =4.
θu (rad)
1000 Vu
0
All simulation experiments were carried out in MATLAB
-1 800
0 20 40 60 80 100 120
V (m/s)
R2012a environment on a PC with i7-2.5GHz CPU and 4GB t (s)
memory. 0 600
φu (rad)
-2 400
evasion", in this research, the UCAV is located outside FIGURE 12: Time history of FIGURE 13: Time history of
the missile’s GKZ. Meanwhile, not exceed the missile’s the UCAV’s track angle velocity of vehicles
dynamic-escape zone (DEZ) [33]. In this case, the UCAV
cannot successfully evade the missile through a conventional
single maneuver. Therefore, according to the missile perfor- explosion point (the UCAV is hit. t > tmax or Vm < Vmmin
mance and research needs, initial states of the UCAV and will also triggers the explosion of the missile, similarly
the missile are set as shown in Table 1. The meanings of six hereinafter).
variables in Table 1 are similar to (5) and (10), just delete the As showed in Fig. 11 and Fig. 12, θu is always zero in
subscript. the whole process, φu is changing from 0 to −180deg in
less than 20s during the evasion. Meanwhile, the intercep-
TABLE 1: Initial states of the UCAV and the missile
tion geometry is changing from a head-on to a tail-chase
x (km) z (km) y (km) V (m/s) θ (rad) φ (rad) geometry. It’s noted from Fig. 13, the time history of the
UCAV 0 20 9 250 0 -pi
UCAV’s velocity is consistent with the above strategy. The
Missile 50 20 9 300 0 0 missile is in the passive deceleration state after the active
acceleration of tp . End condition of tmax or Vmmin is not
States in Table 1 form a typical initial scenario of BVR triggered. According to Fig. 14, the normal overload of the
air combat, and the scenario is set for a head-on starting missile does not change much, especially it equals 1 in the
geometry, in which both the aspect angle and the flight path tail-chase phase. Moreover, the relative distance r gradually
angle of the UCAV and missile are set to zero. The geometry decreases from r0 = 50km to r ∈ [0, dmax ] (see Fig. 15).
is also set with a relative distance of 50km. As regards the above three evasive tactical objectives, the
In this simulation experiment, a typical evasive strategy
used for the UCAV is set as follows: As the enemy missile is 2.5 50
launched, the UCAV immediately performs the out maneuver
40
with a constant initial velocity, and let nzd =4, φu0 =0◦ , and 2
φud = 180◦ in (13). Then it accelerates to Vu max by using 30
r (km)
m
n
12 VOLUME 4, 2016
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
J3
research foundation for the subsequent algorithm verification 0.4
through the test of models and the initial scenario. On the 0.2
other hand, it illustrates that the typical evasive strategy
cannot accomplish the evasive task, let alone multiple tactical 0
0
objectives.
0.2
0.4
C. SIMULATION EXPERIMENT 2
J1 0.6
As a comparative study, the initial scenario in this simula-
0.8 1
tion experiment is the same with that of Section IV-B. The 0.6 0.8
1 0.2 0.4
main difference is that the evasive strategy is given by the 0 J2
optimization model and algorithm proposed in this paper.
An approximate Pareto front is obtained through digital FIGURE 17: Fitted surface of the approximate Pareto front
simulation (see Fig. 16), it shows a set of nondominated
solutions which reflect the different evasive tactical require-
Furthermore, the distribution characteristics of the deci-
ments of UCAV while ensuring successful evasion. Each
sion variables in the obtained nondominanted strategy set can
nondominanted solution in Fig. 16 represents a strategy with
be seen in Fig. 18 and Fig. 19.
a structure as shown in Fig. 8. It also represents a feasible eva-
sive flight path. The values of the three objective functions in 70
1 Ki=1
Fig. 16 are normalized, and the boundary values of the MD, 60
Ki=2
EC, and GST are [0.0531, 13.4111]km, [272.651, 603.245], 0.8 50 Ki=3
proportion (%)
and [0, 15.20]s, respectively. To observe the changing trend 0.6 40
of the approximate Pareto front more intuitively, through the 0.4
30
Curve Fitting Tool in Matlab2012a, the data in Fig. 16 is 20
0.2
generated into the fitted surface as shown in Fig. 17. 10
0 0
S1
S1 S2
S2 S3
S3 S4
S4 K1
K K2
K K3
K KK4
2000
1 2 3 4
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
1) The optimal MD Besides, the deceleration effect of the missile is more obvious
According to the calculation results in Fig. 16, it can be than that in simulation experiment 1, the major reason is the
obtained that the optimal MD in the nondominated solution larger drag force of the missile in lower altitude.
set is 13.4111km, of which decision set are shown in Table As regards the tactical objectives, the objective function
2. value of the EC and the GST is 603.245 and 0, respectively.
Based on the initial scenario in Table 1 and the decision set Therefore, the evasive trajectory of optimal MD can guaran-
in Table 2, the simulation results are shown in Fig. 20-Fig. 22. tee the maximum survival probability of UCAV, but it also
means that the probability of hitting the target is relatively
low. Furthermore, the large EC also leads to relatively less
tactical superiority in subsequent air combat.
Missile
UCAV 2) The optimal EC
According to the calculation results in Fig. 16, it can be
10 60 obtained that the optimal EC in the nondominated solution
40 set is 272.651, of which decision set are shown in Table 3.
y (km)
5
20 Based on the initial scenario in Table 1 and the decision set
0 0 x (km)
in Table 3, the simulation results are shown in Fig. 23-Fig. 25.
15
20 -20
z (km) 25 -40
10 60
8 40
y (km)
2 1200
Vm
0
6 20
θu (rad)
1000 Vu
-2
0 x (km)
-4 800
16
-20
V (m/s)
0 20 40 60 80 100 120
t (s)
600
18
z (km) 20 -40
0
φu (rad)
-2 400
-4 200
FIGURE 23: Three-dimensional trajectories of the UCAV
0 20 40 60 80 100 120 0 20 40 60 80 100 120
t (s) t (s) and the missile when EC is optimal
FIGURE 21: Time history of FIGURE 22: Time history of
the UCAV’s track angle velocity of vehicles 1 1200
Vm
θu (rad)
0 1000 Vu
Simulation results show that the missile triggers a self-
800
destruction caused by Vm < Vmmin at 104.24s and the MD -1
0 20 40 60 80 100 120
V (m/s)
t (s) 600
is 13.4111km. 0
400
It can be seen from Fig. 20 and Fig. 21 that the UCAV
φu (rad)
-2
200
performs a split-S maneuver with high overload immediately
after the missile is launched, and the tail-chase geometry -4 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
t (s) t (s)
quickly formed within 8.2s. The turning efficiency is much
higher than that in simulation experiment 1. The jump of FIGURE 24: Time history of FIGURE 25: Time history of
the UCAV’s track angle between 0rad and −3.14rad at the UCAV’s track angle velocity of vehicles
8.2s is caused by the defined range of track angle, it’s
actually continuous (similarly hereinafter). Then it performs Simulation results show that the missile triggers a self-
the L-Dive maneuver with θud = −0.72rad, followed by a destruction caused by t > tmax and the MD is 0.0531km.
horizontal flight with the velocity of Vu max at the altitude It can be seen from Fig. 23 and Fig. 24 that the UCAV
of r0 · S2 = 3.87km. Until the relative distance r is performs the out maneuver with light overload immediately
r0 · S3 = 22.5km, the UCAV starts the barrel roll maneuver. after the missile is launched, the turning takes nearly 20s
The periodic change of the UCAV’s track angle can be seen and then the L-Dive maneuver starts with θud = −0.78rad,
in Fig. 21. As the MD is greater than the trigger distance of followed by a horizontal flight at the altitude of r0 · S2 =
the terminal type maneuvers in the decision set, the terminal 5.49km. Afterwards, UCAV accelerates to 360.2m/s by the
maneuver is not performed. It’s noted from Fig. 22 that the CA maneuver in the period type maneuver. In the terminal
UCAV turns with deceleration and then accelerates to Vu max . phase, UCAV continues the horizontal flight state of the
14 VOLUME 4, 2016
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
2 1200
previous phase. It’s noted from Fig. 25, the UCAV never Vm
0
θu (rad)
Vu
use the maximum velocity during the whole process, and -2
1000
V (m/s)
deceleration effect of the missile is relatively slow, which t (s)
1 600
results in a corresponding increase in the evasive time. 0
φu (rad)
400
As regards the tactical objectives, compared with the strat- -2
egy of the optimal MD, this strategy saves more than half of -4
0 20 40 60 80 100 120
200
0 20 40 60 80 100 120
t (s) t (s)
the energy for the UCAV, and fully guarantees the tactical
superiority in subsequent air combat. However, the zero GST FIGURE 27: Time history of FIGURE 28: Time history of
and the small MD also lead to relatively low probability both the UCAV’s track angle velocity of vehicles
of survival and hitting the target.
3) The optimal GST load periodically change along with the altitude reduction,
According to the calculation results in Fig. 16, it can be thereby consuming the missile’s energy. Besides, the desired
obtained that the optimal GST in the nondominated solution altitude is r0 · S2 = 3.24km. The period type maneuvers
set is 15.20s, of which decision set are shown in Table 4. are triggered during the process of altitude reduction, so the
Based on the initial scenario in Table 1 and the decision set UCAV performs the barrel roll maneuver immediately after
in Table 4, the simulation results are shown in Fig. 26-Fig. 28. the S-Dive maneuver. Finally, the UCAV starts the last-ditch
maneuver when the relative distance r is r0 · S4 = 3.4km.
As regards the tactical objectives, this strategy provides
longer GST for UCAV, which ensures the probability of
Missile
hitting the target to a large extent. However, the large EC (i.e.,
UCAV 602.18) and the small MD also bring adverse effects to the
survival probability and the tactical superiority in subsequent
10 60 air combat, respectively.
The simulation result of the missile’s normal overload and
40
y (km)
z (km) 25
nmJ1
-30 35
nmJ2 40
rJ2 0.15
30 rJ3
nmJ3 0.1
25 30
FIGURE 26: Three-dimensional trajectories of the UCAV 0.05
r (km)
20
m
0
and the missile when GST is optimal 103105 110 115 120
n
20
15
10
10
Simulation results show that the missile triggers a self- 5
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
Fig. 29 that both nmJ1 and nmJ3 are higher than nmJ2 in further validate the both feasibility and effectiveness of the
the middle and later periods. This means that the missile’s proposed approach. Furthermore, it is of guiding significance
energy decreases rapidly, and the periodic change of the to the research the decision of attacking maneuver for UCAV
missile’s normal overload caused by the periodic maneuver in BVR air combat. Moveover, it also helps pilots perform
of UCAV also accelerates this decrease. It’s worth noting that the evasive tactical maneuvers preferably in the actual air
the last-ditch maneuver leads to the missile’s normal overload combat.
surge to the limit nmax in nmJ3 . According to Fig. 30, rJ3
declines rapidly in the early evasive phase due to the phase of V. CONCLUSION AND FUTURE WORK
guidance support. Besides, the MD of rJ1 is the largest, the Ensuring tactical superiority and survival probability in BVR
GST of rJ2 is the longest. air combat are critical to UCAV. This paper studies the
The results and analysis of this simulation experiment autonomous evasive maneuver strategy of the UCAV to evade
validate the both feasibility and effectiveness of the proposed BVR AAM. The proposed novel approach combines qualita-
approach for solving the evasive maneuver problem in BVR tive tactical experience and quantitative maneuver decision
air combat. The obtained nondominanted solutions all are optimization method effectively. The amalgamative tactical
feasible evasive strategies, and different evasive strategies demands of achieving self-conflicting evasive objectives in
have different tactical superiority and tendencies. The strat- actual BVR air combat are also taken into account. The
egy should be determined according to the current air combat objectives include the MD, the EC and the GST. They re-
scenario and tactical requirements, so as to achieve the de- sponds directly to the survival probability, the superiority
sired tactical objectives, as well as ensure air superiority and in subsequent air combat and the probability of hitting the
the whole combat efficiency while evasive successfully. target, respectively. In this approach, effective maneuvers of
UCAV used in different evasion phases are modeled in three-
D. SIMULATION EXPERIMENT 3 dimensional space, and the three-level decision space struc-
In order to further validate the proposed approach, several ture is established according to qualitative evasive tactical
simulations under different initial scenarios are carried out in planning. On this basis, the HMOEA is designed to solve
this section, which the UCAV is attacked by the missile from the aforementioned problem. Using scenario-based simula-
the same altitude, same distance and different directions. The tions, a set of nondominated evasive strategies is derived,
simulation results are shown in Table 5. which satisifies a variety of evasive tactical requirements of
The heading angle of the missile varies from −30deg to UCAV while ensuring successful evasion. Besides, because
+30deg with the interval of 15deg. Because the situation is the obtained results in this paper are easy to interpret, they
the same in symmetric scenario, only the scenario of turning also have practical guiding significance for pilots to perform
right is considered for the UCAV. The heading angle of the the evasive tactical maneuvers preferably in the actual air
UCAV varies from 0deg to 30deg with the interval of 10deg. combat. In general, the proposed models and algorithm are
These are common initial scenarios of BVR air combat, feasible and effective for solving the problem of evasive
and the initial scenario in simulation experiment 1 and 2 maneuver strategy for UCAV in BVR air combat.
correspond to the head-on geometry where the heading angle It is necessary to note that, the simulation experiment 2
of both the missile and UCAV are 0deg. takes 2.3 hours based on the operating environment parame-
z1 , z2 , and z3 in Table 5 represent the optimal value of J1 , ters given in Section IV-A, so it is difficult to meet the real-
J2 , and J3 in the obtained nondominanted solutions. It can time requirements of online application through the computer
be seen from Table 5, the value of each objective function is configuration in Section IV-A. Therefore, the proposed al-
the largest in the scenario where the heading angle of both gorithm can be used for off-line calculation in a variety of
the missile and the UCAV are 0deg, which is the worst- scenarios, and the results could be loaded into the UCAV
case scenario for evasion. As the heading angle of both sides computer. It can select the corresponding decision set to
deviates from the head-on state, the value of each objective perform evasive maneuver online according to the current
function decreases gradually. When the heading angle of scenario and tactical requirements. Furthermore, the results
both the missile and the UCAV are 30deg, the value of each could be applied for the UCAV or the manned fighter in
objective function reaches the minimum, that is, it is the most the daily BVR air combat evasive tactics training to improve
favorable case to evade. It’s worth noting that the variation combat efficiency.
of the heading angle of the UCAV has a larger effect on the Moreover, considering the uncertainty of information and
evasive results than that of the heading angle of the missile. building more accurate models are also the focus and direc-
The results and analysis in this simulation experiment tion of future research.
16 VOLUME 4, 2016
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2978883, IEEE Access
Z. Yang et al.: Evasive Maneuver Strategy for UCAV in BVR Air Combat
VOLUME 4, 2016 17
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.