Unmanned aerial vehicle evasive tracking object method and system based on particle swarm algorithm

Abstract

The application provides a UAV evading tracking object method and system based on a particle swarm algorithm, relates to the technical field of parameter optimization of UAV evading tracking objects, and comprises the following steps: constructing a dynamic model according to the position, speed, acceleration and guidance law of a tracking object; constructing a UAV flight simulation model according to aerodynamic parameters, control surface response and environmental interference factors; constructing a UAV evading action model, wherein the evading action model comprises multiple evading schemes with adjustable parameters; performing simulation of the UAV evading the tracking object based on the dynamic model, the UAV flight simulation model and the evading action model, solving the constructed objective function through a particle swarm algorithm to obtain globally optimal evading tracking object parameters; and performing offline training based on the optimal evading tracking object parameters to generate an evading strategy library. The application has the advantages of improving the reliability, adaptability and acquisition efficiency of UAV evading parameters.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of parameter optimization of unmanned aerial vehicle (UAV) evading tracking objects, in particular to a method and system for UAV evading tracking objects based on a particle swarm algorithm. BACKGROUND

[0002] With the wide application of UAVs in the fields of logistics transportation, disaster relief, etc., the security threats faced by UAVs are increasingly complex, especially the threats from dynamic tracking objects.

[0003] Existing evasion techniques are mostly based on fixed parameters or empirical rules, that is, the existing evasion action parameters (such as overload and maneuvering time) are mostly statically set and cannot be dynamically adjusted according to real-time threats. If optimization methods based on genetic algorithms or dynamic programming are used, the convergence speed will be slow, which cannot meet the real-time requirements. Therefore, the existing UAV evasion techniques cannot adapt to the complex behaviors and environmental conditions of tracking objects.

[0004] Therefore, it is urgent to study the parameter optimization method for UAV evading tracking objects to improve the reliability, adaptability and acquisition efficiency of UAV evasion parameters. SUMMARY

[0005] The present application relates to the field of parameter optimization of unmanned aerial vehicle (UAV) evading tracking objects, in particular to a method and system for UAV evading tracking objects based on a particle swarm algorithm.

[0006] The present application is achieved by the following technical solutions:

[0007] The method for UAV evading tracking objects based on a particle swarm algorithm comprises the following steps:

[0008] A dynamic model is constructed according to the position, speed, acceleration and guidance law of the tracking object, and the dynamic model is used to simulate the real-time behavior of the tracking object;

[0009] A UAV flight simulation model is constructed according to the aerodynamic parameters, control surface response and environmental disturbance factors, and the UAV flight simulation model is used to simulate the flight state of the UAV;

[0010] An evasion action model of the UAV is constructed, and the evasion action model comprises multiple evasion schemes with adjustable parameters;

[0011] Based on the dynamic model, the UAV flight simulation model and the evasion action model, a simulation of the UAV evading tracking objects is performed, and a global optimal evasion parameter for tracking objects is obtained by solving the constructed objective function through a particle swarm algorithm;

[0012] Based on the optimal evasion parameter for tracking objects, an offline training is performed to generate an evasion strategy library.

[0013] Preferably, the method for constructing the dynamic model is:

[0014] ;

[0015] ;

[0016] ;

[0017] wherein, is a proportional coefficient, is an overload command in the pitch direction of the tracking object, is an overload command in the yaw direction of the tracking object, is an overload limiting value, is the flight height of the tracking object, is the relative closing speed, is the relative angular velocity of the UAV in the horizontal direction, is the relative angular velocity of the UAV in the yaw plane, is the gravitational acceleration;

[0018] If , the values of and are transformed as follows:

[0019] ;

[0020] .

[0021] Preferably, the method for constructing the UAV flight simulation model is:

[0022] ;

[0023] ;

[0024] ;

[0025] ;

[0026] ;

[0027] ;

[0028] wherein, , and are the positions of the UAV in the x-axis, y-axis and z-axis of the ground coordinate system at time t, and the letters appearing at the top of the dots represent the derivation with respect to time, , and respectively the horizontal, the pitch and the yaw of the UAV, is the longitudinal control overload of the UAV, is the yaw direction turning control overload of the UAV, is the pitch direction turning control overload of the UAV, is the gravity acceleration.

[0029] Preferably, the evasive scheme comprises:

[0030] Tail setting maneuver: the direction of the horizontal overload of the UAV is the side with the larger angle with the vector from the UAV to the tracked object, the pitch direction overload controls the UAV to maintain the vertical speed, the yaw direction overload and the vertical speed are the adjustable parameters;

[0031] Down high maneuver: the UAV makes accelerated straight line motion, the longitudinal size is the maximum value under the limit adjustment, the vertical speed is the adjustable parameter;

[0032] Terminal maneuver: the direction of the horizontal overload of the UAV is the side with the larger angle with the vector from the tracked object to the UAV, the pitch direction overload is 0, the longitudinal overload takes the maximum value, and the yaw direction overload is the adjustable parameter.

[0033] Preferably, the model construction method of the tail setting maneuver is:

[0034] ;

[0035] ;

[0036] ;

[0037] wherein, , and respectively the horizontal, the pitch and the yaw of the UAV, is the longitudinal maximum overload of the UAV, represents the vertical speed proportion coefficient and is a fixed coefficient set in advance, is the adjustment target of the vertical speed as the adjustable parameter, is the current actual vertical speed, is the angle between the horizontal projection of the UAV speed vector and the vector from the UAV to the tracked object, is the maximum overload that the UAV can bear, is the adjustment target of the yaw direction overload as the adjustable parameter.

[0038] Preferably, the construction method of the down high maneuver is:

[0039] ;

[0040] ;

[0041] ;

[0042] wherein, is a proportional coefficient.

[0043] Preferably, the terminal maneuvering model construction method is:

[0044] ;

[0045] ;

[0046] .

[0047] Preferably, the objective function comprises:

[0048] minimizing the hit penalty :

[0049] ;

[0050] wherein, represents the miss distance of the end time of the avoidance action, is a preset miss distance threshold, and P is a penalty value; minimizing the energy consumption

[0051] :

[0052] ; wherein,

[0053] represents the actual flight speed of the UAV, represents the minimum flight speed of the UAV, and respectively represent the overload size of the UAV in the y and z directions, is the start time of the avoidance action; When a particle swarm algorithm is used for solving, the individual speed of the i-th particle

[0054] is updated using the following formula:

[0055] ; wherein,

[0056] is an inertia coefficient, is an individual extremum coefficient, is a group extremum coefficient, is a random coefficient, ​generating a random number between 0 and 1, generating a d-dimensional independent Gaussian distribution vector, and the mean of each dimension is 0 and the variance is 1, representing the upper limit of the value of the parameter vector, representing the lower limit of the value of the parameter vector, representing the current position of the example, representing the individual extreme value, representing the population extreme value.

[0057] Preferably, the generation method of the avoidance strategy library is:

[0058] Collecting the corresponding optimal avoidance tracking object parameters calculated under different speed, direction and altitude of the tracking object and the unmanned aerial vehicle;

[0059] Using a neural network for offline training to memorize the optimal avoidance tracking object parameters under all scenarios.

[0060] The present application also provides an unmanned aerial vehicle avoidance tracking object system based on a particle swarm algorithm, which is applied to the unmanned aerial vehicle avoidance tracking object method based on the particle swarm algorithm described above, and comprises:

[0061] A tracking object simulation module is configured to construct a dynamic model according to the position, speed, acceleration and guidance law of the tracking object, and the dynamic model is configured to simulate the real-time behavior of the tracking object;

[0062] An unmanned aerial vehicle flight simulation module is configured to construct an unmanned aerial vehicle flight simulation model according to the aerodynamic parameters, control surface response and environmental disturbance factors, and the unmanned aerial vehicle flight simulation model is configured to realize the simulation of the flight state of the unmanned aerial vehicle;

[0063] An avoidance action module is configured to construct an avoidance action model of the unmanned aerial vehicle, and the avoidance action model comprises a plurality of avoidance schemes with adjustable parameters;

[0064] A random particle swarm optimization module is configured to simulate the avoidance of the tracking object by the unmanned aerial vehicle based on the dynamic model, the unmanned aerial vehicle flight simulation model and the avoidance action model, to solve the constructed objective function by the particle swarm algorithm, and to obtain the globally optimal avoidance tracking object parameters;

[0065] A neural network learning module is configured to perform offline training based on the optimal avoidance tracking object parameters, and to generate an avoidance strategy library.

[0066] The technical scheme of the present application has at least the following advantages and beneficial effects:

[0067] Compared with the fixed parameters or empirical rules used in the existing evasion technology, the present application can realize more reliable and targeted adjustment of the evasion parameters by constructing multiple evasion action models with adjustable parameters and combining the dynamic behavior of the tracked object and environmental information, and improve the adaptability of the unmanned aerial vehicle to complex threat environment.

[0068] The present application introduces a particle swarm algorithm to solve the constructed objective function, has the characteristics of fast convergence speed and strong global search ability, and can quickly obtain high-quality evasion strategies in a limited time to meet the real-time requirements in practical applications.

[0069] The present application performs offline training based on the optimization results of the particle swarm algorithm, generates an evasion strategy library covering multiple environments, and can directly call the content of the strategy library in practical applications, without the need for real-time complex operation to make high-quality evasion decisions, significantly improving the response speed and decision efficiency of the unmanned aerial vehicle when encountering dynamic threats.

[0070] The present application combines dynamic modeling, simulation deduction and intelligent optimization, can automatically adjust the evasion parameters according to the different behavior characteristics of the tracked object, form diversified and optimized response strategies, and greatly enhance the autonomous survival ability and task completion rate of the unmanned aerial vehicle in various working fields. BRIEF DESCRIPTION OF DRAWINGS

[0071] Figure 1 The flowchart of the unmanned aerial vehicle evasion method based on the particle swarm algorithm provided in Embodiment 1 of the present application is shown.

[0072] Figure 2 The principle diagram of the unmanned aerial vehicle evasion system based on the particle swarm algorithm provided in Embodiment 2 of the present application is shown. DETAILED DESCRIPTION

[0073] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0074] Embodiment 1

[0075] The present embodiment provides an unmanned aerial vehicle evasion method based on a particle swarm algorithm, referring to Figure 1 , comprising the following steps:

[0076] Step S1: Construct a dynamic model according to the position, speed, acceleration and guidance law of the tracked object, and the dynamic model is used to simulate the real-time behavior of the tracked object.

[0077] In the embodiment, the dynamic model is constructed based on the proportional guidance principle, and the construction method is as follows:

[0078] ;

[0079] ;

[0080] ;

[0081] wherein, is a proportional coefficient, is an overload instruction in the pitch direction of the tracked object, is an overload instruction in the yaw direction of the tracked object, is an overload limiting value, is the flight height of the tracked object, is the relative closing speed, is the relative angular velocity of the UAV in the horizontal direction, is the relative angular velocity of the UAV in the yaw plane, is the gravity acceleration;

[0082] If , the values of and are transformed as follows:

[0083] ;

[0084] .

[0085] Step S2: constructing a UAV flight simulation model according to the aerodynamic parameters, control surface responses and environmental disturbance factors, wherein the UAV flight simulation model is used to realize UAV flight state simulation.

[0086] As a preferred solution, the construction method of the UAV flight simulation model is as follows:

[0087] ;

[0088] ;

[0089] ;

[0090] ;

[0091] ;

[0092] ;

[0093] wherein, , and respectively are the positions of the UAV in the x-axis, y-axis and z-axis of the ground coordinate system at time t, the letter appearing at the top of the dot represents the derivation with respect to time, 、 and respectively are the velocity, the track pitch angle and the track yaw angle of the UAV, is the longitudinal control overload of the UAV, is the yaw direction turning control overload of the UAV, is the pitch direction turning control overload of the UAV, is the gravity acceleration.

[0094] Step S3: constructing an evasion action model of the UAV, the evasion action model comprising a plurality of evasion schemes with adjustable parameters.

[0095] On this basis, the evasion schemes comprise:

[0096] Tail setting maneuver: the direction of the horizontal direction overload of the UAV is the larger one of the angle between the vector from the UAV to the tracked object, the pitch direction overload controls the aircraft to maintain the vertical direction speed, and the yaw direction overload and the vertical direction speed are the adjustable parameters;

[0097] High-low maneuver: the UAV performs accelerated linear motion, the longitudinal size is the maximum value under the limit adjustment, and the vertical direction speed is the adjustable parameter;

[0098] End maneuver: the direction of the horizontal direction overload of the UAV is the larger one of the angle between the vector from the tracked object to the UAV, the pitch direction overload size is 0, the longitudinal overload takes the maximum value, and the yaw direction overload is the adjustable parameter.

[0099] Specifically, the model construction method of the tail setting maneuver is:

[0100] ;

[0101] ;

[0102] ;

[0103] wherein, 、 and respectively are the horizontal direction overload, the pitch direction overload and the yaw direction overload of the UAV, is the maximum longitudinal overload of the UAV, represents the vertical speed proportion coefficient and is a fixed coefficient set in advance, is the adjustment target of the vertical direction speed as the adjustable parameter, is the current actual vertical direction speed, Let be the angle between the UAV's velocity vector and the horizontal projection of the UAV onto the tracked object vector. To ensure the maximum overload that the drone can withstand, The target for adjusting yaw overload, which is an adjustable parameter.

[0104] Furthermore, the method for constructing the lower high maneuver is as follows:

[0105] ;

[0106] ;

[0107] ;

[0108] in, This is the proportionality coefficient.

[0109] On the other hand, the model construction method for the end-effector maneuver is as follows:

[0110] ;

[0111] ;

[0112] .

[0113] This step breaks down the drone's evasive maneuvers into three stages: tail placement, altitude drop, and terminal maneuver, and optimizes the parameters for each stage.

[0114] Step S4: Simulate the UAV's avoidance of tracked objects based on the dynamic model, UAV flight simulation model, and avoidance action model. Solve the constructed objective function using the particle swarm optimization algorithm to obtain the global optimal avoidance parameters.

[0115] The objective functions include:

[0116] Minimize hit penalty :

[0117] ;

[0118] in, Represents the end time of the evasion maneuver. Off-target amount, P is the preset off-target threshold, and P is the penalty value.

[0119] Minimize hit penalty In the process, if the tracked object hits the drone, J1 is equal to a large penalty value P; if the tracked object misses, J1 is equal to the reciprocal of the miss amount.

[0120] Minimize energy consumption :

[0121] ;

[0122] wherein, represents the actual flight speed of the UAV, represents the minimum flight speed of the UAV, and respectively represent the overload size of the UAV in the y and z directions, is the start time of the evasion action.

[0123] The target function includes maximizing the survival probability and minimizing the energy consumption. The maximizing of the survival probability ensures that the UAV gives priority to evading risks and avoids being hit or interfered when facing the threat of the tracking object. The minimizing of the energy consumption ensures that the energy consumption is reduced in the evasion process, the endurance time is prolonged, and the resource waste caused by excessive maneuvering is avoided. The combination of the two can realize efficient flight under the premise of ensuring safety.

[0124] The random particle swarm algorithm balances the maneuvering performance and energy consumption, reduces the energy consumption to provide more operation space for subsequent maneuvering, and can also avoid the individual from falling into local optimum, thereby increasing the reliability and accuracy of the solution.

[0125] As a further optimization scheme, when the particle swarm algorithm is used for solving, the individual speed of the i-th particle is updated using the following formula:

[0126] ;

[0127] wherein, is an inertia coefficient, is an individual extreme value coefficient, is a population extreme value coefficient, is a random coefficient, generates a random number between 0 and 1, generates a d-dimensional independent Gaussian distribution vector, and the mean value of each dimension is 0 and the variance is 1, represents the upper limit of the value of the parameter vector, represents the lower limit of the value of the parameter vector, represents the current position of the example, represents the individual extreme value, represents the population extreme value.

[0128] Compared with the traditional particle swarm algorithm which only relies on the individual and global optimum to guide the search, the embodiment introduces a Gaussian random disturbance term A certain random disturbance can be injected in the search process, so as to effectively avoid the algorithm from falling into a local optimal solution and improve the global optimization capability. Meanwhile, the introduction of the Gaussian disturbance enables the particle update to have a directional random jumping feature, and the boundary interval The disturbance amplitude can be dynamically adjusted according to the scale of the search space, the search diversity is maintained, and the search domain is not excessively divergent, so that the entire search domain is more efficiently covered. When solving a high-dimensional nonlinear problem involving multiple optimization objectives, i.e., maximizing the survival probability and minimizing the energy consumption, the traditional particle swarm algorithm is prone to premature convergence. The improved scheme of the embodiment introduces an inertia coefficient and three control coefficients - jointly regulate the search balance, realizes the dynamic balance between exploration and utilization, and helps to improve the convergence accuracy and robustness of the algorithm in a complex environment.

[0129] Step S5: Based on the optimal evasive tracking object parameters, offline training is performed to generate an evasion strategy library.

[0130] Based on the scheme, the generation method of the evasion strategy library is as follows:

[0131] The corresponding optimal evasion tracking object parameters calculated under different speeds, directions and altitudes of the tracking object and the unmanned aerial vehicle are collected;

[0132] The neural network is used for offline training to memorize the optimal evasion tracking object parameters in all scenarios.

[0133] That is, based on the offline training of the neural network, when the unmanned aerial vehicle needs to evade the tracking object, the speed, direction and altitude information of the tracking object and itself can be input to quickly obtain a good evasion scheme, meeting the rapid decision-making demand of dynamic threats.

[0134] Based on the scheme of the embodiment, through multi-objective optimization, the survival rate of the unmanned aerial vehicle in a complex threat scenario is improved.

[0135] Embodiment 2

[0136] The application further provides an unmanned aerial vehicle evasion tracking object system based on a particle swarm algorithm, which is applied to the unmanned aerial vehicle evasion tracking object method based on the particle swarm algorithm and is described with reference to Figure 2 , and comprises:

[0137] A tracking object simulation module is configured to construct a dynamic model according to the position, speed, acceleration and guidance law of the tracking object, and the dynamic model is configured to simulate real-time behavior of the tracking object.

[0138] The UAV flight simulation module is used for constructing a UAV flight simulation model according to the aerodynamic parameters, the control rudder response and the environmental interference factors, and the UAV flight simulation model is used for realizing the UAV flight state simulation.

[0139] The evasion action module is used for constructing an evasion action model of the UAV, and the evasion action model comprises a plurality of evasion schemes with adjustable parameters.

[0140] The random particle swarm optimization module is used for simulating the UAV evading the tracking object based on the dynamic model, the UAV flight simulation model and the evasion action model, solving the constructed target function through the particle swarm algorithm, and obtaining the globally optimal evasion tracking object parameters.

[0141] The neural network learning module is used for performing offline training based on the optimal evasion tracking object parameters, and generating an evasion strategy library.

[0142] The above only is the preferred embodiment of the present application, and is not used for limiting the present application, for the person skilled in the art, the present application can have various changes and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for UAV evading tracking object based on particle swarm optimization algorithm, characterized in that, The method comprises the following steps: a dynamic model is constructed according to the position, speed, acceleration and guidance law of the tracking object, and the dynamic model is used to simulate the real-time behavior of the tracking object; an unmanned aerial vehicle flight simulation model is constructed according to aerodynamic parameters, control surface response and environmental interference factors, and the unmanned aerial vehicle flight simulation model is used to realize simulation of the flight state of the unmanned aerial vehicle; an evasion action model of the unmanned aerial vehicle is constructed, and the evasion action model comprises multiple evasion schemes with adjustable parameters; simulation of evasion of the unmanned aerial vehicle from the tracking object is performed based on the dynamic model, the unmanned aerial vehicle flight simulation model and the evasion action model, a global optimal evasion tracking object parameter is obtained by solving a constructed objective function through a particle swarm algorithm; the objective function comprises: Minimizing hit penalties : ； wherein, represents the end time of the cutoff avoidance action is an off-target amount, is a preset off-target threshold value, and P is a penalty value. Minimizing energy consumption : ； wherein, represents the actual flight speed of the drone, represents the minimum flight speed of the drone, and represent the magnitude of the overloading of the drone in the y and z directions, respectively, is the start time of the evasion maneuver; When solving by using the particle swarm algorithm, the individual velocity of the i-th particle The following formula is used for updating: ； wherein, is an inertia coefficient, is an individual extremum coefficient, is a population extremum coefficient, is a random coefficient, generates a random number between 0 and 1, generates a d-dimensional independent Gaussian distributed vector with mean 0 and variance 1 in each dimension, denotes an upper limit of the value of the parameter vector, denotes a lower limit of the value of the parameter vector, denotes the current position of the example, denotes the individual extremum, denotes the population extremum; offline training is performed based on the optimal evasion tracking object parameter, and an evasion strategy library is generated.

2. The method of claim 1, wherein, The method for constructing the dynamic model is: ； ； ； wherein, is a proportional coefficient, is an overload command in the pitch direction of the tracking object, is an overload command in the yaw direction of the tracking object, is an overload limit value, is a flight altitude of the tracking object, is a relative closing speed, is a relative angular velocity of the UAV in the horizontal direction, is a relative angular velocity of the UAV in the yaw plane, is a gravitational acceleration; If then transform the values of and ​ ； 。 3. The method of claim 1, wherein, The method for constructing the unmanned aerial vehicle flight simulation model is: ； ； ； ； ； ； wherein, , and are the positions of the UAV in the x-axis, y-axis and z-axis of the ground coordinate system at time t, respectively, and the letters appearing at the top of the dots represent the derivation with respect to time, , and are the velocity, the path pitch angle and the path yaw angle of the UAV, respectively, is the longitudinal control overload of the UAV, is the yaw direction turning control overload of the UAV, is the pitch direction turning control overload of the UAV, is the gravitational acceleration.

4. The method of claim 1, wherein, The evasion scheme comprises: a tail setting maneuver: the direction of the horizontal direction overload of the unmanned aerial vehicle is the side with a larger included angle with the vector from the unmanned aerial vehicle to the tracking object, the pitch direction overload controls the unmanned aerial vehicle to maintain the vertical direction speed, the yaw direction overload and the vertical direction speed are the adjustable parameters; a low-to-high maneuver: the unmanned aerial vehicle performs accelerated linear motion, the longitudinal size is the maximum value under the limitation adjustment, and the vertical direction speed is the adjustable parameter; a terminal maneuver: the direction of the horizontal direction overload of the unmanned aerial vehicle is the side with a larger included angle with the vector from the tracking object to the unmanned aerial vehicle, the pitch direction overload size is 0, the longitudinal overload takes the maximum value, and the yaw direction overload is the adjustable parameter.

5. The method of claim 4, wherein, The method for constructing the model of the tail setting maneuver is: ； ； ； wherein, , and are respectively horizontal, pitch and yaw direction overload of the UAV, is the longitudinal maximum overload of the UAV, represents the vertical speed proportionality coefficient and is a fixed coefficient set, is the vertical direction speed adjustment target as an adjustable parameter, is the current actual vertical direction speed, is the angle between the horizontal direction projection of the UAV speed vector and the UAV-to-tracked object vector, is the maximum overload that the UAV can bear, is the yaw direction overload adjustment target as an adjustable parameter.

6. The method of claim 5, wherein, The method for constructing the low-to-high maneuver is: ； ； ； wherein is a proportionality factor.

7. The method of claim 6, wherein, The method for constructing the model of the terminal maneuver is: ； ； 。 8. The method of claim 1, wherein, The method for generating the evasion strategy library is: corresponding optimal evasion tracking object parameters calculated under different speeds, directions and altitudes of the tracking object and the unmanned aerial vehicle are collected; a neural network is used for offline training to memorize the optimal evasion tracking object parameters under all scenes.

9. The UAV evading tracker system based on particle swarm optimization algorithm, applied to the UAV evading tracker method based on particle swarm optimization algorithm in any one of claims 1-8, characterized in that, The method comprises the following steps: a tracking object simulation module is configured to construct a dynamic model according to the position, speed, acceleration and guidance law of the tracking object, and the dynamic model is used to simulate the real-time behavior of the tracking object; an unmanned aerial vehicle flight simulation module is configured to construct an unmanned aerial vehicle flight simulation model according to aerodynamic parameters, control surface response and environmental interference factors, and the unmanned aerial vehicle flight simulation model is used to realize simulation of the flight state of the unmanned aerial vehicle; an evasion action module is configured to construct an evasion action model of the unmanned aerial vehicle, and the evasion action model comprises multiple evasion schemes with adjustable parameters; a random particle swarm optimization module is configured to perform simulation of evasion of the unmanned aerial vehicle from the tracking object based on the dynamic model, the unmanned aerial vehicle flight simulation model and the evasion action model, and a global optimal evasion tracking object parameter is obtained by solving a constructed objective function through a particle swarm algorithm; a neural network learning module is configured to perform offline training based on the optimal evasion tracking object parameter, and generate an evasion strategy library.

Images