Maneuvering Decision Making Based on Cloud Modeling Algorithm for UAV Evasion–Pursuit Game

Abstract

When facing problems in the aerial pursuit game, most of the current unmanned aerial vehicles (UAVs) have good maneuverability performance, but it is difficult to utilize the overload maneuverability of UAVs properly; further, UAVs tend to be more costly, and it is often difficult to effectively prevent the enemy from reaching the tailgating position behind the UAV in the aerial pursuit game. Therefore, there is a pressing need for a maneuvering algorithm that can effectively allow a UAV to quickly protect itself in a disadvantageous position, stably and effectively select a maneuver with the maneuvering algorithm, and stably and effectively establish an advantage by moving to an advantageous position. Therefore, this paper establishes a cloud model-based UAV-maneuvering aerial pursuit decision-making model based on pursuit-and-evasion game positions. Based on the evaluation of the latter, when the UAV is at a disadvantage, we use the constructed defensive maneuver expert pool to abandon the disadvantageous position. When the UAV is at an advantage, we use cloud model-based pursuit-and-evasion game maneuvering decision making to establish an advantageous position. According to the results of the simulation examples, the maneuvering decision-making method designed in this paper confirms that the UAV can quickly abandon its position and establish an advantage in case of parity or disadvantage and that it can also stably establish a tail-chasing position in case of advantage.

1. Introduction

The unmanned aerial vehicle (UAV) aerial fugitive game problem is a key issue in current research in the field of UAVs. Pursuit-and-evasion game confrontation among UAVs has attracted the attention of various countries and scholars. Current research on UAV pursuit-and-evasion confrontation strategies has been mainly conducted via the differential countermeasure method [1], the expert system method [2], the influence diagram decision-making method, etc., but the shortcomings of these methods include the level of difficulty in obtaining analytical solutions, lack of flexibility, and limited applicability. Thus, studying tailgating game UAV-maneuvering decision making has become very important [1,2].

However, in the increasingly complex aerial pursuit mission environment, autonomous maneuvering decision making requires that UAVs be able to rationally generate pursuit game maneuvering control commands in different aerial pursuit positions. Regarding the current aerial pursuit maneuvers, although UAVs can perform more aerial maneuvers than manned aircraft, there are still major deficiencies and shortcomings in their ability to assess the pursuit situation and to make correct pursuit game maneuvering decisions [3,4]. One instance in the literature [5] used fuzzy inference to select maneuvers from a standard maneuver library. Another study [6] proposed an autonomous maneuvering decision-making method based on intelligent differential countermeasures. Yet another study [7] optimized control using multilevel influence diagrams for rolling control. One study [8] proposed to apply an adaptive pseudo-parallel genetic algorithm to solve the maneuvering decision-making problem. However, there is a big gap between the decision-making methods in the above studies and actual aerial pursuit, and it is difficult to perform situational assessment in high-speed changing aerial pursuit situations.

How to make the UAV ensure that the enemy aircraft cannot reach a tailing position when the UAV is in a disadvantageous environment and how to have it make more correct maneuver decisions when the enemy maneuvers into an advantageous position, thus reaching a tailing position, are the current key problems in maneuvering decision making in pursuit games.

The main idea behind pursuit game UAV-maneuvering decision making is based on previous successful maneuvering decision-making experiences in air flight confrontation exercises, as well as actual aerial pursuit examples taken from active pilots [9]. In this context, decision making relates to two aspects: how to get rid of an inferior aerial pursuit position and how to establish a superior aerial pursuit position [10]. In this paper, firstly, through the study of the constructed cloud model-based position assessment algorithm, which is used to carry out the position analysis of the performance parameters of the two aircraft and the pursuit battle position information in a modern UAV air environment, we establish the correlation model for the pursuit battle position assessment based on the positions of the enemy and the UAV, the radar position, and the aerial pursuit performance. Afterwards, based on the assessment of the flight positions of both aircraft by the aforementioned algorithm, maneuvering decision making is performed in two steps: decision making based on the specialization-modified defensive expert maneuver library, which is used to abandon the disadvantageous position, and decision making based on the MAX–MIN cloud model, which is used to steadily improve the advantageous position. The UAV-maneuvering simulation model is created by setting up a UAV simulation model in Matlab/Simulink. The UAV maneuver instruction library is also established in Matlab/Simulink, and the enemy maneuvers range from simple, straight-line flight to random-maneuver flight, while our UAV adopts the decision-making method presented in this paper. The innovations of this paper mainly consist of the following:

I. A cloud model-based aerial pursuit position assessment model is designed;

II. A maneuvering decision-making method based on MAX–MIN cloud reasoning is established.

2. UAV Evasion–Pursuit Modeling

2.1. Mathematical Modeling of Fugitive-Tracing Dynamics

The prerequisite for a UAV to make aerial pursuit maneuvering decisions is for it to be able to perform an assessment of the initial pursuit situation. The first step of such assessment is to obtain information on the situation [8]. Then, the UAV obtains the pursuit game state information through sensing sensors and the information interaction network; then, it obtains the pursuit game state information space, which is composed of the pursuit game state information set after filtering. Let the information at moment t in the state space be denoted by $s_i^t$, where $i$ denotes the $i^{th}$ message in the state space [9] and $i=1,2,3\dots n$. The state space of the decision information set at moment $t$ is

$$S^t=[s_1^t,s_2^t,s_3^t,s_4^t\dots s_n^t]$$

(1)

Due to the large amount of information related to real air flight and the pursuit game, to be able to better establish a model describing a pursuit game situation, this paper hereby simplifies the state space of the decision-making information of the latter to a certain extent, and according to the main method of decision making adopted, the state space is divided into the following:

(1)

UAV location: $s_1^t=(x_u^t,y_u^t,z_u^t)$. It indicates the position of the UAV in the inertial coordinate system;

(2)

Approaching rate: $s_2^t$. It indicates the difference between the velocity vectors of the two airplanes projected on the target line-of-sight angle formed by the line of sight between the UAV and the target; when it is greater than 0, it means that the two airplanes are approaching each other, and when it is less than 0, it means that the two airplanes are moving away from each other;

(3)

Relative positions of the enemy and UAV: $s_3^t=(R^t,{\phi }_h^t,{\phi }_v^t)$, where $R^t$ denotes the relative distance between the enemy and the UAV at moment t, ${\phi }_h^t$ denotes the azimuth of the enemy aircraft on the horizontal plane at moment t, and ${\phi }_v^t$ denotes the azimuth of the enemy aircraft in the lead hammer plane at moment t. The relative positions of the enemy aircraft and the UAV are determined based on these three data;

(4)

Enemy aircraft heading direction: $s_4^t=({\theta }_1,{\theta }_2)$, as shown in Figure 1, where ${\theta }_1$ denotes [10,11,12,13,14] the angle between the UAV velocity vector and the relative distance vector between the enemy and the UAV and ${\theta }_2$ denotes the angle between the velocity vector of the enemy aircraft and the UAV velocity vector [3];

(5)

Enemy radar system status: $s_5^t=\{radar,ECM\}$, where $rader$, for enemy radar locking status, indicates whether the aircraft is locked by the enemy radar irradiation, where 0 indicates not locked and 1 indicates locked; $ECM$ indicates the electronic countermeasures, where 0 indicates that the enemy has not implemented electronic interference and 1 indicates that the enemy has implemented electronic interference. Therefore, the mathematical model of UAV decision making can be expressed as follows:

$$a_k^t=f(s_1^t,s_2^t,s_3^t,s_4^t,s_5^t)$$

(2)

As shown in Figure 2, the aspect angle (AA) and the antenna train angle (ATA) are a pair of angular data that are used to describe the situation of airborne pursuit. When the situation is even or our UAV is at an advantage in the pursuit game and there are no time constraints or weak time constraints, we can use cloud reasoning for decision making to gain a greater advantage, thus achieving a dominant position and winning the game faster.

2.2. Cloud Modeling-Based Assessment of UAV Aerial Pursuit Position

In the decision-making process of pursuit game maneuvering, the first step is to assess the position. This assessment involves extracting the feature attributes of the pursuit game position. To achieve this, this paper proposes a cloud model-based system for assessing pursuit-and-evasion game positions. The system conceptualizes the feature attributes of the information of such positions in a cloud and constructs a feature attribute cloud model.

The cloud model, proposed by Deyi Li [15], an academician of the Chinese Academy of Engineering, falls under the category of uncertainty in artificial intelligence. It is an uncertainty transformation model that deals with qualitative concepts and quantitative descriptions. The model uses uniform information values to portray the large number of random and fuzzy elements in the system and links the fuzzy qualitative concepts in the system with their corresponding numerical values.

The cloud of the model consists of many cloud droplets, which derive from the quantization of the qualitative concepts in the fuzzy system and represent their projection in the number field space. The cloud is represented by the following: $Ex$—expectation; $En$—entropy; and $He$—hypertrophy, which represent the mathematical properties of the fuzzy concepts. $Ex$ denotes the most representative value of a fuzzy concept after quantization; $En$ denotes the size of the range of values accepted by the concept; and $He$ denotes the degree of uncertainty. Thus, a cloud can be represented as $Cloud(Ex,En,He)$. The determination of the fuzzy qualitative concept is achieved by using a large number of cloud drops; then, for the aerial pursuit position assessment, the affiliation degree of the current position is obtained by cloud inference to infer the current aerial pursuit position [16].

Cloud models include various types of clouds, such as normal clouds, trapezoidal clouds, semi-lifting clouds, semi-descending clouds, exponential clouds, and others. These can be further classified into one-dimensional, two-dimensional, and multi-dimensional clouds based on the number of dimensions. Figure 3 illustrates the different types of clouds.

Many natural stochastic phenomena follow a normal distribution, which is why scholars often use normal clouds in cloud models, as we do in this paper. Cloud generators are used to convert qualitative concepts into quantitative values for aerial pursuit positions. The cloud generator comprises a forward component and an inverse component. The forward generator is necessary to convert imprecise qualitative concepts into quantitative value mapping [17].

The one-dimensional normal cloud droplet generation method is shown in Figure 4.

When constructing the pursuit game position information of maneuvering rules, different sub-domains are created according to different pursuit game feature attributes; for instance, the enemy and UAV altitude difference information is divided into three sub-domains, i.e., high, medium, and low. Then, the conceptual cloud family of pursuit game position features is constructed according to the specific position language information in the thesis domain, as shown in Figure 5.

As shown in Figure 5, the figure shows the model cloud of the proximity rate. The set of blue points represents the affiliation cloud when the proximity rate of both sides is too low. The set of red points represents the affiliation cloud when the proximity rate of both sides is moderate. The set of yellow points represents the affiliation cloud in the case where the proximity rate of both sides is too high. The situational assessment system uses the family of proximity rate clouds to determine the proximity rate and derive the proximity rate affiliation function.

Figure 6 displays the cloud model of heights. The blue point set represents the affiliation cloud when the UAV’s height is too low. The set of red dots represents the affiliation cloud when the UAV has an advantageous altitude. The yellow point set represents the affiliation cloud when the UAV’s altitude is too high. The altitude cloud family is used by the situational assessment system to determine the altitude situation and derive the altitude affiliation function.

As shown in Figure 7, the yellow point set shows the affiliation cloud when the attacking aircraft’s angle of advance deviates to the right. The blue point set shows the affiliation cloud when the attacker’s angle of advance deviates to the left. The red point set shows the affiliation cloud when the attacking aircraft’s angle of advance is centered. The model represents the angular cloud of the attacking aircraft. The system for situational assessment uses the angular cloud family to determine the advance angle of the attacking aircraft and derive the angular affiliation function.

Figure 8 shows the modeled distance cloud. The blue point set represents the affiliation cloud when the two aircraft are too close. The red point set represents the affiliation cloud when the distance between the two aircraft is moderate. The yellow point set represents the affiliation cloud when both parties are far away from each other. The system for situational assessment uses the distance cloud family to determine the distance between the two parties and derive the distance affiliation function.

The sub-domains and numerical features of different eigenvalues are reasonably set based on the operational performance of the UAV. Cloud families are constructed for each eigenvalue attribute. The family of cloud-related terms, including $[s_1^t,s_2^t,s_3^t,s_4^t]$, i.e., relative altitude, proximity, distance, and ATA, is established. For $s_5^t$, i.e., features with weak or non-ambiguous ambiguity and randomness, such as whether the UAV is locked or not, no cloud model is built. Instead, position assessment matching is carried out directly in the decision-making process.

3. Cloud Model-Based Aerial Pursuit Maneuvering Decision Making

UAV Modeling

Before determining the maneuvering decision-making method, the UAV used in this paper is first modeled. The model parameters are shown in Figure 9.

The mass point model is

$$\{\begin{array}{l}\dot{x}=V\cos \gamma \sin \psi \\ \dot{y}=V\cos \gamma \cos \psi \\ \dot{z}=V\sin \gamma \end{array}$$

(3)

where x, y, and z represent the position of the UAV in the inertial coordinate system; $\dot{x}$, $\dot{y}$, and $\dot{z}$ are the components of the velocity vector on the three coordinate axes; $\gamma$ denotes the track angle, i.e., the angle between the velocity vector and the horizontal plane [18,19,20]; and $\psi$ indicates the heading angle, which is the angle between the horizontal projection of the velocity vector and the direction of the axis. The derivative expressions of $V$, $\gamma$, and $\psi$ are

$$\dot{V}=g\left(n_x-\sin \gamma \right)$$

(4)

$$\dot{\gamma }=\frac{g}{V}\left(n_z\cos \varphi -\cos \gamma \right)$$

(5)

$$\dot{\psi }=\frac{g{n}_z\sin \varphi }{V\cos \gamma }$$

(6)

The prime model utilizes three main control quantities: $n_x$, $n_z$, and $\varphi$. This paper assumes that the velocity vector is consistent with the axial direction of the airframe. $n_x$ indicates that the UAV platform is overloaded along the velocity direction, mainly for UAV thrust. $n_z$ is normal overload, indicating overload along the pitch direction. $\varphi$ indicates the velocity roll angle, and this item also characterizes the platform roll control vector method based on MAX–MIN cloud inference.

In the process of the UAV evasion–pursuit game described in this paper, the UAV first acquires both its own current fuzzy position and that of the enemy. It then uses the cloud model established in a previous paper to make fuzzy information judgments and position assessments. Finally, it scores and derives the current position model.

Figure 10 shows that after the position assessment using the cloud model, when the aircraft is at a disadvantage in the pursuit game, we perform a time constraint judgment and set the current position as having strong time constraints. This requires rapid decision making to overcome the disadvantage. This paper proposes the use of expert pool-based decision making specialized in defensive aerial pursuit maneuvering. When time is limited, decision making based on the defense expert pool is used to quickly determine the appropriate pursuit game maneuver for the current situation. When there are no time constraints, MAX–MIN cloud inference decision making is used to make maneuvering decisions when the UAV is in a favorable or even position in the pursuit game.

The universal maneuvering decision-making expert bank system is mainly composed of a knowledge base, a maneuver bank, and a reasoning machine. When the UAV enters the expert library decision-making system, the acquired situational assessment information is passed onto the expert system reasoning machine, which matches it with each information condition in the knowledge base and finds a flight maneuver in the maneuver library that matches the current situation. The maneuver library built in this paper focuses on defensive improvements to the main maneuvers in the traditional aerial maneuver expert library to ensure that UAVs can quickly get out of a disadvantageous situation when making decisions.

In the literature [21] typical tactical maneuvers have been designed: (1) straight flight, (2) constant circling, (3) forward tracking, (4) dive and speed increase, (5) diagonal pull-up, (6) sharp pull-up, (7) pure tracking, (8) half-jacket, (9) leap and half roll, (10) high-overload up-roll, (11) high-overload down-roll, (12) evasive turn, (13) sharp evasion, (14) speed increase turn, (15) snake maneuver, (16) barrel roll, (17) sharp disc descent, (18) large slope turn, (19) slew, (20) sharp descent turn, (22) declining inversion, (23) half-roll inversion, (24) high-speed yoyo, and (25) low-speed yoyo.

These 25 maneuvers are mainly formed by the synthesis of a variety of basic movements.

Table 1. Basic maneuvers.

	Left turn  $d\psi >0$
$V$	Acceleration  $dv>0$
$n$	$n_x>\sin \gamma$
$dy$	$dy>0$ (Rapid climb)	$dy=0$ (Constant-velocity flight)			$dy<0$ (Fast dive)
		$\gamma <0$ (Swooping down)	$\gamma >0$ (Climb)	$\gamma \approx 0$ (Horizontal plane)
$n_z$	$n_z\cos \varphi >\cos \gamma$	$n_z\cos \varphi =\cos \gamma$		$n_z\cos \varphi =1$	$n_z\cos \varphi <\cos \gamma$
$\varphi$	$-\pi /2\leftarrow 0$ taper	$\varphi =\mathrm{constant}(-\pi /2,0)$			$-\pi /2\leftarrow 0$ taper
	1 Left turn, acceleration, and rapid climb	2 Left turn, acceleration, and constant dive	3 Left turn, acceleration, and constant climb	4 Horizontal acceleration and left turn	5 Left turn, acceleration, and fast dive

shows some of the basic maneuvers established in this paper.

The rules of the tactical maneuvering reasoning base adopted to establish an expert system that can maximize the success of maintaining our aircraft in an inferior position are mainly to put the opponent in the worst attack position and to deny them the required launch parameters, which corresponds to the $s_5^t=\{radar,ECM\}$ items in the position information [22,23,24,25,26,27,28]. When we are targeted by enemy radar or use ECM jamming, defensive maneuvering decisions are made according to a strong time constraint. The defensive improvements in the rule base are detailed below.

(1) When enemy aircraft achieve a tailgating position from the rear hemisphere, they tend to reduce altitude and increase speed in order to reduce the range from the rear hemisphere.

(2) The tendency to emergency dive based on the expert pool should be increased when the altitude allows for it, increasing the overload by emergency dive descent using gravity and also forcing the enemy aircraft to look down and change the AA because of the reduced altitude.

(3) When enemy planes reach a tailing position with respect to our planes, a defensive emergency turn is taken to urgently turn toward clouds, the sun, and other strong interferers on the one hand and to reach a lateral orientation to achieve maximum angular velocity of aiming line movement on the other.

In maneuvering decision making in the pursuit game, most of the maneuvering decision rules based on pilots’ experience are fuzzy rules based on the fuzzy correlation of “if A then B”. For example, “if the conditions of the enemy aircraft being located on the right side of our UAV and the ATA being less than the maximum off-axis angle are satisfied and if the position information reaches the launch condition, then our aircraft will arrive at the tailing condition”. In the decision-making cloud model established in this paper, the cloud model is first used to cloud the fuzzy concept feature attributes contained in the qualitative rules of tactical maneuvering decision making; then, the feature attribute cloud model is constructed.

In this paper, we use the previously constructed situational assessment information to cloud all the information required for maneuvering decisions and transform the “if A then B” aerial pursuit decision into a cloud inference situational rule set as follows:

If $S_1^t(v_1),S_2^t(v_1),S_3^t(v_1),S_4^t(v_1),S_5^t(v_1)$, then $stat{e}_1$ and $actio{n}_{1\dots n}$.

If $S_2^t(v_2),S_2^t(v_2),S_3^t(v_2),S_4^t(v_2),S_5^t(v_2)$, then $stat{e}_2$ and $actio{n}_{1\dots n}$.

If $S_1^t(v_m),S_2^t(v_m),S_3^t(v_m),S_4^t(v_m),S_5^t(v_m)$, then $stat{e}_m$ and $actio{n}_{1\dots n}$.

In the above equation, v in rule m denotes the linguistic value ordinal number of state volume S at moment t; state denotes the current state type inferred through cloud inference; $actio{n}_{1\dots n}$ indicates the preferred 1-to-n tactical actions in the tactical action scheme according to the current rule.

1. Input the current state information feature values into the cloud generator; evaluate the input information as a cloud drop(i); and obtain the affiliation and concept evaluation corresponding to the current state.

$$Te{p}_i^t=[(mu{A}_i^t,sign{A}_i^t),(mu{V}_i^t,sign{V}_i^t),(mu{H}_i^t,sign{H}_i^t),(mu{D}_i^t,sign{D}_i^t)]$$

(7)

2. Input the situational information obtained from the cloud inference model into the maneuvering decision-making cloud model, and score the current maneuver library using the cloud inference rule base.

3. Analyze the current highest-scoring maneuvers using simulations to predict the next position.

Convert the predicted situational information into cloud droplets and input into the cloud generator, and evaluate the situation after the predicted maneuvering decision.

$$G_{a_i}^t\prime =\alpha G_{a_i}^t+\beta G_{a_i}^{t+1}$$

(8)

4. Repeat processes 2–4 to review the maneuvers whose scores are higher than the set scoring threshold in the current rule base, and select the maneuvers with the highest scores. As is shown in Figure 11.

The MAX–MIN cloud inference approach utilizes the cloud family definition to construct a cloud family of all feature attributes for the natural language rule set of maneuvering operation. To ensure a reasonable set of numerical features for different language values, common sense of modern aerial pursuit is taken into account.

Firstly, according to the antecedent cloud generator of each feature language value, information set (cloud droplets) $q_1$ is used as the input to the generator to find the deterministic degree (${\mu }_{T_1}(q_1)$) of $q_1$ belonging to feature $T_1$, which satisfies

$${\mu }_{T_i}\left(q_i\right)=\max \left\{{\mu }_{T_i({\nu }_1)}(q_i),\cdots ,{\mu }_{T_i({\nu }_j)}\left(q_i\right),\cdots \right\},j=1,2,\cdots ,N_i$$

(9)

where ${\mu }_{T_1}(q_1)$ is the certainty of each language value of feature $T_1$. After determining the value, the position of the corresponding linguistic value of the feature in the rule set is saved.

Then, the selected rule is finalized based on the determinism of the cloud droplets of each feature, and the determinism of the rule is obtained as ${\mu }^{*}$. ${\mu }^{*}$ satisfies

$${\mu }^{\ast }=\min \left\{{\mu }_{T_i(v_i)}(q_1),\cdots ,{\mu }_{T_i(v_i)}(q_i),\cdots \right\},i=1,2,\cdots m$$

(10)

where the subscripts $a,b,c$ denote the linguistic value sequence number of each feature. After determining ${\mu }^{*}$, the selected rule for inference is determined based on the corresponding linguistic value, and the selected rule corresponds to the maneuvering decision.

4. Simulation Example

This section compares and simulates the UAV aerial pursuit maneuvering decision model constructed in this paper to verify the performance of the proposed method. The description of the dominant situation in the membership function of the situation parameters is revised. Several typical situations, such as non-maneuvering flight and maneuvering flight of enemy aircraft, are simulated and analyzed separately. The simulation demonstration is mainly presented through 3D position maps, altitude difference curves, velocity difference curves, and fitness curves. The fitness curve represents the total evaluation score of the UAV’s position.

The fitness curve represents the tailing weapon launch indicator derived from combining the four positions $[muA,muV,muH,muD]$. In the relative altitude plot, the horizontal axis represents the simulation time, and the vertical axis represents the difference in altitude between the UAV and the enemy aircraft, with the UAV being positive at the top. In the relative distance graph, the horizontal axis represents the simulation time, and the vertical axis represents the difference in distance between the UAV and the enemy aircraft, with the enemy aircraft at the front being positive.

4.1. Simulate Scenario 1

Figure 12, Figure 13, Figure 14 and Figure 15 show the enemy’s maneuver, represented by the red line, as a straight-line maneuver, while the UAV, represented by the blue line, uses the maneuvers selected with the model designed in this paper.

The enemy aircraft’s initial flight status consists of the altitude of (3000, 3000, 3000) m and the speed of 204 m/s. The UAV’s initial flight status consists of the position of (0, 0, 2700) m and the speed of 250 m/s. It is assumed that the enemy aircraft does not carry out any attack operations and only maintains steady straight flight. The UAV′s front hemisphere faces the enemy, and the enemy aircraft flies in a straight line.

Based on the simulation results, it can be concluded that the head-on flight of two planes can enable our aircraft to achieve a more favorable position and ultimately reach a tail-following position. This position can achieve a score of more than 0.7 in the integrated affiliation fitting function.

4.2. Simulation Scenario 2

As shown in Figure 16, Figure 17, Figure 18 and Figure 19, the enemy’s maneuver, represented by the red line, is set as an S-turn maneuver, while the UAV, represented by the blue line, uses the maneuvers selected with the model designed in this paper.

The enemy aircraft’s initial flight status consists of the position of (3000, 3000, 3000) m and the speed of 204 m/s. The UAV’s initial flight status consists of the position of (0, 0, 2700) m and the speed of 250 m/s. It is assumed that the enemy performs turning maneuvers.

Based on the simulation results, it can be concluded that the proposed strategy allows the UAV to escape from an inferior position when the enemy executes a turning maneuver. This enables the UAV to achieve an advantageous tailgating position faster while maintaining the relative distance and height required to comply with the launch standard. The integrated affiliation fitting function can reach a score of more than 0.6.

4.3. Simulation Scenario 3

As shown in Figure 20, Figure 21, Figure 22, Figure 23, Figure 24, Figure 25, Figure 26, Figure 27, Figure 28, Figure 29, Figure 30 and Figure 31, the enemy’s maneuver, represented by the red line, is set as a straight-line maneuver, and the UAV, represented by the blue line, uses the maneuvers selected with the model designed in this paper.

The enemy aircraft’s initial flight status consists of the position of (3000, 3000, 3000) m and the speed of 204 m/s. The UAV’s initial flight status consists of the position of (0, 0, 2700) m and the speed of 250 m/s. It is assumed that the enemy performs turning maneuvers. The UAV’s rear hemisphere faces the enemy, who flies in a straight line. Our aircraft is at a disadvantage and being tailed.

Based on the simulation results, it can be concluded that the strategy enables our UAV to quickly recover from its being disadvantaged and establish an advantage by achieving the required relative distance and height in line with launch standards. The integrated affiliation fitting function can achieve a score of over 0.9.

4.4. Simulation Scenario 3

As shown in Figure 24, the enemy’s maneuver, represented by the red line, is set to a free maneuver, and the UAV, represented by the blue line, uses the maneuvers selected with the model designed in this paper.

The enemy aircraft’s initial flight status consists of the position of (3000, 3000, 3000) m, the speed of 204 meters per second, the track inclination of 0 degrees, and the course angle of −135 degrees. The UAV’s initial flight status consists of the position of (0, 0, 2700) m, the speed of 250 meters per second, the track inclination of 0 degrees, and the course angle of 45 degrees. It is assumed that the enemy performs random maneuvers.

4.5. Simulation Scenario 4

As shown in Figure 28, the enemy’s maneuver, represented by the red line, is set to a free maneuver, and the UAV, represented by the blue line, uses the maneuvers selected with the model designed in this paper.

The enemy aircraft’s initial flight status consists of the position of (3000, 3000, 3000) m, the speed of 204 meters per second, the track inclination of 0 degrees, and the course angle of 45 degrees. The UAV’s initial flight status consists of the position of (0, 0, 2700) m, the speed of 250 meters per second, the track inclination of 0 degrees, and the course angle of 45 degrees. It is assumed that the enemy performs random maneuvers.

Based on the simulation results, it can be concluded that if the enemy employs random maneuvering, the UAV can escape from disadvantageous positions and reach the advantageous tailgating position faster, meeting the launch criteria at relative distance and height with a score greater than 0.7 in the integrated affiliation fitting function.

Based on the simulation results of the several scenarios above, it can be concluded that the UAV maneuvering decision-making method proposed in this section based on a cloud model–expert database combination allows UAVs to abandon disadvantageous situations as soon as possible, prolonging their survival time as much as possible, shortening the occupying time during the attack process, and quickly seizing the attack opportunity. The survival rate of UAVs is greatly improved in disadvantageous situations, and they achieve dominant positions much faster in advantageous scenarios.

5. Conclusions

According to the simulation scenarios set in the previous section for analysis, it can be seen from the results that the expert pool–cloud model maneuvering decision-making model proposed in this paper can achieve the following: when the UAV is in a disadvantaged position, it can make a quick decision in a limited time to abandon this position; when the UAV is in an advantageous aerial pursuit position, it can effectively increase the advantage (continuously improve the comprehensive calculation of the fitness function in the cloud model) and assume a tailgating position. From the simulation results, it can be seen that the decision-making method based on the defensive extended expert pool–cloud model proposed in this paper has good robustness and optimization ability in the game of chasing a fleeing enemy.