Performance boundary measurement method for intelligent decision-making system of unmanned aerial vehicle

Abstract

The invention discloses an unmanned aerial vehicle intelligent decision-making system performance boundary measurement method, and belongs to the technical field of unmanned aerial vehicle performance verification and artificial intelligence safety evaluation. According to the method, uncertain factors of an action space and a state space of the intelligent decision-making system of the unmanned aerial vehicle are identified, then a closed-loop simulation test device is constructed, multiple tests are carried out, the error injection magnitude is gradually increased, simulation results such as an event-driven evaluation value and a hit rate are analyzed and processed, and the performance boundary of the intelligent decision-making system of the unmanned aerial vehicle is obtained. The problem that the unmanned aerial vehicle is mistakenly used for the unadaptive scene due to the fact that the intelligent algorithm performance boundary is not clear is solved. The simulation system is divided into the simulation environment and the intelligent decision-making module which are deployed in the upper computer and the industrial personal computer respectively, communication between the upper computer and the industrial personal computer is established based on an HTTP network protocol, the closed-loop simulation function is achieved, the simulation environment and a tested object are separated in a distributed deployment mode, and the simulation environment and the intelligent decision-making module are arranged in the upper computer and the industrial personal computer respectively. And the algorithm model leakage risk is avoided.

Description

Technical Field

[0001] This invention belongs to the field of UAV performance verification and artificial intelligence safety assessment technology, and relates to a method for measuring the performance boundary of a UAV intelligent decision-making system based on the injection of uncertainty factors. Background Technology

[0002] With the continuous development of artificial intelligence technologies such as deep reinforcement learning, intelligent drones, with their high degree of automation and rapid decision-making capabilities, are gradually being applied to complex mission scenarios such as autonomous air combat. However, these data-driven intelligent drones often face multiple uncertainties in real mission environments, such as sensor noise and actuator errors. These factors may cause unexpected behavior or even loss of control of the aircraft platform, seriously threatening flight safety. These problems often stem from significant differences between the training scenarios of intelligent algorithms and real-world environments. When the training scenarios are not covered, the intelligent decision-making algorithms lack sufficient resilience to uncertainties, and disturbances exceeding the performance limits of the intelligent decision-making algorithm lead to erroneous output commands. Therefore, it is urgent to measure the performance limits of intelligent decision-making systems, clarify the degree of resilience of intelligent decision-making algorithms to uncertainties, and thus restrict the application scenarios and flight conditions of intelligent drones to ensure aircraft platform safety.

[0003] Currently, performance verification methods for unmanned aerial vehicle (UAV) systems mainly focus on simulation-based functional testing and traditional fault injection testing. The former is mostly conducted in normal or controlled environments, making it difficult to effectively verify the system's robustness in complex and uncontrollable environments. The latter primarily targets traditional control systems based on explicit physical models and mathematical formulas, failing to address the uncertainties in the action and state spaces of intelligent decision-making algorithms. Its pre-defined fault modes cannot adequately cover the "black box" characteristics of intelligent decision-making algorithms, and traditional fault injection methods cannot quantitatively and progressively explore the performance degradation boundaries and safety thresholds of intelligent decision-making algorithms under the influence of uncertainty. Furthermore, current closed-loop simulation systems deploy the test object and the simulation environment on the same computer, raising concerns among users about algorithm model leakage when delivering the test object. Existing methods cannot meet users' intellectual property protection needs for the test object.

[0004] In summary, existing technologies lack a method to systematically and quantitatively evaluate the performance boundaries of intelligent decision-making algorithms under multiple uncertainties, making it difficult to accurately predict their safe operating limits before the algorithm is used. This constitutes a technical problem of insufficient reliability and prominent safety hazards in the practical application of intelligent drones. Summary of the Invention

[0005] To address the issue of drones being misused in unsuitable scenarios due to unclear performance boundaries of intelligent algorithms, which could lead to aircraft loss of control and safety problems, this invention aims to provide a method for measuring the performance boundaries of a drone intelligent decision-making system. By building a distributed measurement device consisting of a host computer and an industrial control computer, the method can determine the task-qualified performance boundaries and safe flight boundaries for the intelligent decision-making system, thereby improving drone flight safety.

[0006] The objective of this invention is achieved through the following technical solution:

[0007] This invention discloses a method for measuring the performance boundary of an unmanned aerial vehicle (UAV) intelligent decision-making system, comprising the following steps:

[0008] Step 1: Construct an uncertainty factor analysis and error model to inject uncertainties of different magnitudes into the intelligent decision-making system in a simulation environment.

[0009] 1.1: Analyze the state space and motion space parameters of the object under test to obtain parameters with uncertainties.

[0010] The model's state space and action space are obtained. The state space includes the position of our aircraft, the position of the enemy aircraft, and their relative distance. The action space includes rudder deflection commands and throttle commands. The position information of our aircraft comes from the integrated navigation system, and the position information of the enemy aircraft comes from the detection system; both contain uncertainties introduced by sensor measurements. Furthermore, the rudder deflection commands output to the actuators are subject to uncertainties such as response errors.

[0011] 1.2: For the parameters in 1.1 that have uncertainties, construct the corresponding error model.

[0012] For the position parameters of our aircraft, an error model is established:

[0013] x t_out =x t_real +ΔR xz *cos(θ)

[0014]

[0015] z t_out =z t_real +ΔR xz *sin(θ)

[0016] In the formula, x t_real y t_real z t_real This indicates the actual location of our aircraft; x t_out y t_out z t_outThe position output of the added error model; θ is the error direction angle randomly generated within a single simulation, θ∈[-π,π); n1 is a positive integer randomly generated within a single simulation; ΔR xz ΔR is the horizontal positioning error randomly generated for a single simulation step size. xz ~N(μ) Rxz ,σ Rxz );ΔR y The vertical positioning error, ΔR, is generated randomly for a single simulation step size. y ~N(μ) Ry ,σ Ry );

[0017] For the enemy aircraft's position parameters, an error model is established:

[0018]

[0019] In the formula, θ 1t_real θ 2t_real R t_real Actual detection angle and distance; θ 1t_out θ 2t_out R t_out To add the angle and distance output results of the error model; Δθ1, Δθ2, and ΔR are the angle measurement error and distance measurement error randomly generated for a single simulation step, Δθ1~N(μ θ1 ,σ θ1 ), Δθ2~N(μ θ2 ,σ θ2 ), ΔR~N(μ R ,σ R n2, n3, and n4 are positive integers randomly generated within a single simulation.

[0020] For the rudder deflection angle parameter, an error model is established:

[0021]

[0022] In the formula, D Aileron_real D Elevator_real D Rudder_real The actual rudder deflection angle command output by the model; D Aileron_out D Elevator_out D Rudder_out To add the rudder deflection angle output of the error model; ΔD A ΔD E ΔD R ΔD is the error randomly generated within a single simulation step. A ~N(μ) DA ,σ DA ),ΔD E ~N(μ) DE ,σDE ),ΔD R ~N(μ) DR ,σ DR n5, n6, and n7 are positive integers randomly generated within a single simulation.

[0023] Step 2: Set up the experimental setup, deploy the simulation environment and the object under test on the host computer and industrial control computer respectively, and transmit the motion space parameters and state space parameters through the HTTP network protocol to achieve closed-loop simulation.

[0024] 2.1: In the Python environment, the JSBim module, perception module, execution module, and intelligent decision system are combined and data communication relationships are established. The error model constructed in 1.2 is written into the perception module and execution module to build a closed-loop simulation system for UAVs. The JSBim module is used to simulate the dynamics and kinematics model of the aircraft, and the intelligent decision module is used to deploy the object under test and perform inference.

[0025] 2.2: The simulation system is divided into two parts: a simulation environment and an intelligent decision-making module. The simulation environment consists of a JSBsim module, a perception module, and an execution module. The simulation environment is deployed on a host computer, and the object under test is deployed on an industrial control computer. Communication between the host computer and the industrial control computer is established based on the HTTP network protocol to realize the transmission of motion space parameters and state space parameters.

[0026] 2.3: Record the aircraft state at the end of the closed-loop simulation run in Section 2.2. Initialize the crash, shotdown, and missile_success parameters to 0. If the aircraft state is crashed, then crash = -200. If the aircraft state is shot down, then shotdown = -200. If the aircraft state is a missile hit, then missile_success = +200.

[0027] Step 3: Inject the error model into the simulation environment based on the fault injection module; conduct simulation experiments, inject uncertainties of different magnitudes, and analyze the task qualification performance boundary and safe flight boundary of the intelligent decision-making system based on the results output by the experimental device.

[0028] 3.1: Conduct M simulation experiments in the experimental setup built in step two. During the experiments, based on the error model in section 1.2, set the initial error injection values. The initial error value of our aircraft's position parameters is: μ Rxz0 σ Rxz0 μ Ry0 σ Ry0 Initial value of enemy aircraft position parameter error: μ θ10 σ θ10 μ θ20 σθ20 μ R0 μ R0 Initial value of rudder deflection angle parameter error: μ DA0 σ DA0 μ DE0 σ DE0 μ DR0 σ DR0 Furthermore, the simulation number i, i∈[1,2,...,M] is used as the amplification factor to amplify the magnitude of the error injection:

[0029] μ Rxz =i*μ Rxz0 ,σ Rxz =i*σ Rxz0

[0030] μ Ry =i*μ Ry0 ,σ Ry =i*σ Ry0

[0031] μ θ1 =i*μ θ10 ,σ θ1 =i*σ θ10

[0032] μ θ2 =i*μ θ20 ,σ θ2 =i*σ θ20

[0033] μ R =i*μ R0 ,σ R =i*σ R0

[0034] μ DA =i*μ DA0 ,σ DA =i*σ DA0

[0035] μ DE =i*μ DE0 ,σ DE =i*σ DE0

[0036] μ DR =i*μ DR0 ,σ DR =i*σ DR0

[0037] 3.2: For the M simulation experiments conducted in 3.1, based on the device output in Section 2.1, the crash of the i-th simulation... i shotdown imissile_success i Parameters are used to calculate the event-driven evaluation value for this simulation. i :

[0038] Evaluation i =crash i +shotdown i +missile_success i

[0039] 3.3 Evaluation based on the M event-driven evaluation values ​​in 3.2 i For i ∈ [1,2,...,M], the evaluation of the first i simulations is statistically analyzed. i The number of times n is greater than or equal to 200 i Calculate the success rate (Success_rate) of the first i simulations. i :

[0040] Success_rate i =(n i / i)*100%3.4 Based on the M success rates in Section 3.3 Success_rate i For i∈[1,2,...,M], calculate the first Success_rate. i The simulation number j when ≤50% is used, and the error injection parameter of the jth simulation is the acceptable performance boundary of the task.

[0041] 3.5 Evaluation based on the M event-driven evaluation values ​​in 3.2 i For i∈[1,2,...,M], count the first Evaluation. i The number of simulations k when the value is less than or equal to -200 is the safe flight boundary.

[0042] The process also includes step four: based on the mission-compliant performance boundaries and safe flight boundaries determined in step three, the tolerable error ranges of various sensors for the aircraft's intelligent decision-making system are obtained. This allows the system to determine whether the operating conditions are within the allowable limits during actual flight. When the system operates within these limits, the reliability of its decision-making and control is ensured. Furthermore, clear performance boundaries can fundamentally prevent the misuse of UAVs in unsuitable scenarios due to unclear intelligent algorithm capability boundaries, ensuring the effectiveness and safety of UAV flight missions.

[0043] Beneficial effects:

[0044] 1. The present invention discloses a method for measuring the performance boundary of an unmanned aerial vehicle (UAV) intelligent decision-making system. By identifying the uncertainty factors in the action space and state space of the UAV intelligent decision-making system, a closed-loop simulation test device is constructed, multiple tests are carried out, and the error injection level is gradually increased. The simulation results such as event-driven evaluation value and hit rate are analyzed and processed to obtain the performance boundary of the UAV intelligent decision-making system, thereby solving the problem that the UAV is misused in unsuitable scenarios due to the unclear performance boundary of the intelligent algorithm.

[0045] 2. The present invention discloses a method for measuring the performance boundary of an unmanned aerial vehicle (UAV) intelligent decision-making system. This method splits the simulation system into two parts: a simulation environment and an intelligent decision-making module. These parts are deployed on a host computer and an industrial control computer, respectively. Communication between the host computer and the industrial control computer is established based on the HTTP network protocol to achieve closed-loop simulation. By separating the simulation environment from the object under test through distributed deployment, the risk of algorithm model leakage is avoided, and the requirements for measuring the performance boundary of the UAV intelligent decision-making system are met. Attached Figure Description

[0046] Figure 1 This is a flowchart of the performance boundary measurement method for an unmanned aerial vehicle (UAV) intelligent decision-making system according to an embodiment of this application.

[0047] Figure 2 This is a closed-loop simulation system according to an embodiment of this application.

[0048] Figure 3 This describes the communication relationship between the host computer and the industrial control computer in this embodiment of the application.

[0049] Figure 4 The evaluation curves are from 100 simulations.

[0050] Figure 5 The success rate curve is generated from 100 simulations. Detailed Implementation

[0051] To better illustrate the purpose and advantages of the present invention, the invention will be further described below in conjunction with the accompanying drawings and examples.

[0052] Example 1:

[0053] like Figure 1 As shown in the figure, the specific implementation steps of the performance boundary measurement method of the UAV intelligent decision-making system disclosed in this embodiment are as follows:

[0054] Step 1: Construct an uncertainty factor analysis and error model to inject uncertainties of different magnitudes into the intelligent decision-making system in a simulation environment.

[0055] 1.1: Analyze the state space and motion space parameters of the object under test, and identify parameters with uncertainties.

[0056] The model's state space and action space are obtained. The state space includes the position of our aircraft, the position of the enemy aircraft, and their relative distance. The action space includes rudder deflection commands and throttle commands. The position information of our aircraft comes from the integrated navigation system, and the position information of the enemy aircraft comes from the detection system; both contain uncertainties introduced by sensor measurements. Furthermore, the rudder deflection commands output to the actuators are subject to uncertainties such as response errors.

[0057] 1.2: For the parameters in 1.1 that have uncertainties, construct their error models.

[0058] For the position parameters of our aircraft, an error model is established:

[0059] x t_out =x t_real +ΔR xz *cos(θ)

[0060]

[0061] z t_out =z t_real +ΔR xz *sin(θ)

[0062] In the formula, x t_real y t_real z t_real This indicates the actual location of our aircraft; x t_out y t_out z t_out The position output of the added error model; θ is the error direction angle randomly generated within a single simulation, θ∈[-π,π); n1 is a positive integer randomly generated within a single simulation; ΔR xz ΔR is the horizontal positioning error randomly generated for a single simulation step size. xz ~N(μ) Rxz ,σ Rxz );ΔR y The vertical positioning error, ΔR, is generated randomly for a single simulation step size. y ~N(μ) Ry ,σ Ry );

[0063] For the enemy aircraft's position parameters, an error model is established:

[0064]

[0065] In the formula, θ 1t_real θ 2t_real R t_real Actual detection angle and distance; θ 1t_out θ2t_out R t_out To add the angle and distance output results of the error model; Δθ1, Δθ2, and ΔR are the angle measurement error and distance measurement error randomly generated for a single simulation step, Δθ1~N(μ θ1 ,σ θ1 ), Δθ2~N(μ θ2 ,σ θ2 ), ΔR~N(μ R ,σ R n2, n3, and n4 are positive integers randomly generated within a single simulation.

[0066] For the rudder deflection angle parameter, an error model is established:

[0067]

[0068] In the formula, D Aileron_real D Elevator_real D Rudder_real The actual rudder deflection angle command output by the model; D Aileron_out D Elevator_out D Rudder_out To add the rudder deflection angle output of the error model; ΔD A ΔD E ΔD R ΔD is the error randomly generated within a single simulation step. A ~N(μ) DA ,σ DA ),ΔD E ~N(μ) DE ,σ DE ),ΔD R ~N(μ) DR ,σ DR n5, n6, and n7 are positive integers randomly generated within a single simulation.

[0069] Step 2: Set up the experimental setup, deploy the simulation environment and the object under test on the host computer and industrial control computer respectively, and transmit the motion space parameters and state space parameters through the HTTP network protocol to achieve closed-loop simulation.

[0070] 2.1: In the Python environment, the JSBim module, perception module, execution module, and intelligent decision-making system are combined and data communication relationships are established. The error model constructed in Section 1.2 is written into the perception module and execution module to build a closed-loop simulation system for the UAV, such as... Figure 2 As shown, the JSBim module is used to simulate aircraft dynamics and kinematic models, while the intelligent decision-making module is used to deploy the tested object and perform inference.

[0071] 2.2: The simulation system is then divided into two parts: a simulation environment and an intelligent decision-making module. The simulation environment consists of a JSBsim module, a perception module, and an execution module. The simulation environment is deployed on a host computer, and the object under test is deployed on an industrial control computer. Communication between the host computer and the industrial control computer is established based on the HTTP network protocol to realize the transmission of motion space parameters and state space parameters, such as... Figure 3 As shown.

[0072] 2.3: Record the aircraft state at the end of the closed-loop simulation run in Section 2.2. Initialize the crash, shotdown, and missile_success parameters to 0. If the aircraft state is crashed, then crash = -200. If the aircraft state is shot down, then shotdown = -200. If the aircraft state is a missile hit, then missile_success = +200.

[0073] Step 3: Inject the error model into the simulation environment based on the fault injection module; conduct simulation experiments, inject uncertainties of different magnitudes, and analyze the task qualification performance boundary and safe flight boundary of the intelligent decision-making system based on the results output by the experimental device.

[0074] 3.1: Conduct 100 simulation tests in the experimental setup designed in step two. During the tests, based on the error model in section 1.2, set the initial error injection values. The initial error value of our aircraft's position parameters is: μ Rxz0 =50m, σ Rxz0 =0.25m, μ Ry0 =-50m, σ Ry0 =0.5m, initial value of enemy aircraft position parameter error: μ θ10 =0.2°, σ θ10 =0.05°, μ θ20 =0.2°, σ θ20 =0.05°, μ R0 =5m, μ R0 =0.1m, initial value of rudder deflection angle parameter error: μ DA0 =0.005, σ DA0 =0.0005, μ DE0 =0.005, σ DE0 =0.0005, μ DR0 =0.005, σ DR0 =0.0005, and the simulation number i, i∈[1,2,...,100] is used as the amplification factor to amplify the magnitude of the error injection:

[0075] μ Rxz =i*μ Rxz0 ,σ Rxz =i*σRxz0

[0076] μ Ry =i*μ Ry0 ,σ Ry =i*σ Ry0

[0077] μ θ1 =i*μ θ10 ,σ θ1 =i*σ θ10

[0078] μ θ2 =i*μ θ20 ,σ θ2 =i*σ θ20

[0079] μ R =i*μ R0 ,σ R =i*σ R0

[0080] μ DA =i*μ DA0 ,σ DA =i*σ DA0

[0081] μ DE =i*μ DE0 ,σ DE =i*σ DE0

[0082] μ DR =i*μ DR0 ,σ DR =i*σ DR0

[0083] 3.2: For the 100 simulation experiments conducted in 3.1, based on the device output in Section 2.1, the crash of the i-th simulation... i shotdown i missile_success i Parameters are used to calculate the event-driven evaluation value for this simulation. i :

[0084] Evaluation i =crash i +shotdown i +missile_success i

[0085] Table 1. Evaluation of 100 simulations

[0086] Simulation number i <![CDATA[Evaluation i ]]> Simulation number i <![CDATA[Evaluation i ]]> Simulation number i <![CDATA[Evaluation i ]]> Simulation number i <![CDATA[Evaluation i ]]> 1 200 26 0 51 -200 76 0 2 200 27 0 52 200 77 0 3 0 28 0 53 0 78 200 4 200 29 200 54 200 79 200 5 0 30 -200 55 0 80 0 6 0 31 200 56 200 81 0 7 0 32 200 57 0 82 0 8 200 33 200 58 200 83 0 9 200 34 200 59 0 84 0 10 0 35 200 60 0 85 0 11 200 36 200 61 200 86 0 12 200 37 200 62 200 87 0 13 200 38 0 63 200 88 0 14 200 39 200 64 0 89 0 15 200 40 200 65 200 90 -200 16 200 41 200 66 -200 91 0 17 0 42 0 67 0 92 -200 18 0 43 200 68 200 93 0 19 200 44 0 69 0 94 0 20 0 45 200 70 0 95 0 21 200 46 0 71 0 96 -200 22 200 47 0 72 200 97 0 23 200 48 200 73 0 98 0 24 -200 49 0 74 200 99 -200 25 200 50 200 75 0 100 0

[0087] 3.3 Evaluation based on 100 event-driven evaluation values ​​from 3.2 i For i ∈ [1,2,...,100], the evaluation of the first i simulations is recorded. i The number of times n is greater than or equal to 200 i Calculate the success rate (Success_rate) of the first i simulations. i :

[0088] Success_rate i =(n i / i)*100%

[0089] Table 2 Success rate of 100 simulations

[0090]

[0091]

[0092] 3.4 Success_rate based on 100 hit rates from Section 3.3 i For i∈[1,2,...,100], calculate the first Success_rate. i The number of simulations when ≤50% is reached. See Table 2 in 3.3. Figure 5 As shown, in the 6th simulation, Success_rate i The initial threshold is 50%, therefore the following parameters represent the acceptable performance boundary for the task.

[0093] μ Rxz =6*μ Rxz0 =300m,σ Rxz =6*σ Rxz0 =1.5m

[0094] μ Ry =6*μ Ry0 =-300m,σ Ry =6*σ Ry0 =3m

[0095] μ θ1 =6*μ θ10 =1.2°,σ θ1 =6*σ θ10 =0.3°

[0096] μ θ2 =6*μ θ20 =1.2°,σ θ2 =6*σ θ20 =0.3°

[0097] μ R =6*μ R0 =30m,σ R =6*σ R0 =0.6m

[0098] μ DA =6*μ DA0 =0.03,σ DA =6*σ DA0 =0.003

[0099] μ DE =6*μ DE0 =0.03,σ DE =6*σ DE0 =0.003

[0100] μ DR =6*μ DR0 =0.03,σ DR =6*σ DR0 =0.003

[0101] 3.5 Evaluation based on 100 event-driven evaluation values ​​from 3.2 i For i∈[1,2,...,100], count the first Evaluation. i The number of simulations when the value is less than or equal to -200. See Table 1 in section 3.2. Figure 4 As shown, the Evaluation during the 24th simulation. i The initial value is -200, therefore the following is the safe flight boundary.

[0102] μ Rxz =24*μ Rxz0 =1200m,σ Rxz =6*σ Rxz0 =6m

[0103] μ Ry =24*μ Ry0 =-1200m,σ Ry =6*σ Ry0 =12m

[0104] μ θ1 =24*μ θ10 =4.8°, σ θ1 =24*σ θ10 =1.2°

[0105] μ θ2 =24*μ θ20 =4.8°, σ θ2 =24*σ θ20=1.2°

[0106] μ R =24*μ R0 =120m,σ R =24*σ R0 =2.4m

[0107] μ DA =24*μ DA0 =0.12,σ DA =24*σ DA0 =0.012

[0108] μ DE =24*μ DE0 =0.12,σ DE =24*σ DE0 =0.012

[0109] μ DR =24*μ DR0 =0.12,σ DR =24*σ DR0 =0.012

[0110] The above detailed description further illustrates the purpose, technical solution, and beneficial effects of the invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for measuring the performance boundary of an unmanned aerial vehicle (UAV) intelligent decision-making system, characterized in that: Includes the following steps: Step 1: Construct an uncertainty factor analysis and error model to inject uncertainties of different magnitudes into the intelligent decision-making system in a simulation environment; Step 2: Set up the experimental setup, deploy the simulation environment and the object under test on the host computer and industrial control computer respectively, and transmit the motion space parameters and state space parameters through the HTTP network protocol to achieve closed-loop simulation. Step 3: Inject the error model into the simulation environment based on the fault injection module; conduct simulation experiments, inject uncertainties of different magnitudes, and determine the task qualification performance boundary and safe flight boundary of the intelligent decision-making system based on the results output by the experimental device.

2. The method for measuring the performance boundary of an unmanned aerial vehicle (UAV) intelligent decision-making system as described in claim 1, characterized in that: The implementation method for step one is as follows: 1.1: Analyze the state space and motion space parameters of the object under test to obtain parameters with uncertainties; The state space and action space of the model are obtained; the state space includes the position of our aircraft, the position of the enemy aircraft, and the relative distance; the action space includes rudder deflection commands and throttle commands; wherein, the position information of our aircraft comes from the integrated navigation system, and the position information of the enemy comes from the detection system; in addition, the rudder deflection commands are output to the actuators. 1.2: For the parameters with uncertainties in 1.1, construct the corresponding error models; For the position parameters of our aircraft, an error model is established: x t_out =x t_real +ΔR xz *cos(θ) z t_out =z t_real +ΔR xz *sin(θ) In the formula, x t_real y t_real z t_real This indicates the actual location of our aircraft; x t_out y t_out z t_out The position output of the added error model; θ is the error direction angle randomly generated within a single simulation, θ∈[-π,π); n1 is a positive integer randomly generated within a single simulation; ΔR xz ΔR is the horizontal positioning error randomly generated for a single simulation step size. xz ~N(μ) Rxz ,σ Rxz );ΔR y The vertical positioning error, ΔR, is generated randomly for a single simulation step size. y ~N(μ) Ry ,σ Ry ); For the enemy aircraft's position parameters, an error model is established: In the formula, θ 1t_real θ 2t_real R t_real Actual detection angle and distance; θ 1t_out θ 2t_out R t_out To add the angle and distance output results of the error model; Δθ1, Δθ2, and ΔR are the angle measurement error and distance measurement error randomly generated for a single simulation step, Δθ1~N(μ θ1 ,σ θ1 ), Δθ2~N(μ θ2 ,σ θ2 ), ΔR~N(μ R ,σ R n2, n3, and n4 are positive integers randomly generated within a single simulation. For the rudder deflection angle parameter, an error model is established: In the formula, D Aileron_real D Elevator_real D Rudder_real The actual rudder deflection angle command output by the model; D Aileron_out D Elevator_out D Rudder_out To add the rudder deflection angle output of the error model; ΔD A ΔD E ΔD R ΔD is the error randomly generated within a single simulation step. A ~N(μ) DA ,σ DA ),ΔD E ~N(μ) DE ,σ DE ),ΔD R ~N(μ) DR ,σ DR n5, n6, and n7 are positive integers randomly generated within a single simulation.

3. The method for measuring the performance boundary of an unmanned aerial vehicle (UAV) intelligent decision-making system as described in claim 2, characterized in that: The second step is implemented as follows: 2.1: In the Python environment, the JSBim module, perception module, execution module, and intelligent decision-making system are combined and data communication relationships are established. The error model constructed in 1.2 is written into the perception module and execution module to build a closed-loop simulation system for the UAV. The JSBim module is used to simulate aircraft dynamics and kinematic models, while the intelligent decision-making module is used to deploy the object under test and perform reasoning. 2.2: The simulation system is divided into two parts: a simulation environment and an intelligent decision-making module. The simulation environment consists of the JSBsim module, the perception module, and the execution module. The simulation environment is deployed on the host computer, and the object under test is deployed on the industrial control computer. Communication between the host computer and the industrial control computer is established based on the HTTP network protocol to realize the transmission of motion space parameters and state space parameters. 2.3: Record the aircraft state at the end of the closed-loop simulation run in Section 2.

2. Initialize the crash, shotdown, and missile_success parameters to 0. If the aircraft state is crashed, then crash = -200. If the aircraft state is shot down, then shotdown = -200. If the aircraft state is a missile hit, then missile_success = +200.

4. The method for measuring the performance boundary of an unmanned aerial vehicle (UAV) intelligent decision-making system as described in claim 3, characterized in that: 3.1: Conduct M simulation experiments in the experimental setup built in step two. During the experiments, based on the error model in section 1.2, set the initial error injection values. The initial error value of our aircraft's position parameters is: μ Rxz0 σ Rxz0 μ Ry0 σ Ry0 Initial value of enemy aircraft position parameter error: μ θ10 σ θ10 μ θ20 σ θ20 μ R0 μ R0 Initial value of rudder deflection angle parameter error: μ DA0 σ DA0 μ DE0 σ DE0 μ DR0 σ DR0 Furthermore, the simulation number i, i∈[1,2,...,M] is used as the amplification factor to amplify the magnitude of the error injection: m Rxz =i*μ Rxz0 ,s Rxz =i*σ Rxz0 m Ry =i*μ Ry0 ,s Ry =i*σ Ry0 m θ1 =i*μ θ10 ,s θ1 =i*σ θ10 m θ2 =i*μ θ20 ,s θ2 =i*σ θ20 m R =i*μ R0 ,s R =i*σ R0 m DA =i*μ DA0 ,s DA =i*σ DA0 m DE =i*μ DE0 ,s DE =i*σ DE0 m DR =i*μ DR0 ,s DR =i*σ DR0 3.2: For the M simulation experiments conducted in 3.1, based on the device output in Section 2.1, the crash of the i-th simulation... i shotdown i missile_success i Parameters are used to calculate the event-driven evaluation value for this simulation. i : Evaluation i =crash i +shotdown i +missile_success i 3.3 Evaluation based on the M event-driven evaluation values ​​in 3.2 i For i ∈ [1,2,...,M], the evaluation of the first i simulations is statistically analyzed. i The number of times n is greater than or equal to 200 i Calculate the success rate (Success_rate) of the first i simulations. i : Success_rate i =(n i / i)*100% 3.4 Based on the M success rates (Success_rate) in Section 3.3 i For i∈[1,2,...,M], calculate the first Success_rate. i The simulation number j when ≤50% is used; the error injection parameter of the jth simulation is the acceptable performance boundary of the task. 3.5 Evaluation based on the M event-driven evaluation values ​​in 3.2 i For i∈[1,2,...,M], count the first Evaluation. i The number of simulations k when the value is less than or equal to -200 is the safe flight boundary.

5. A method for measuring the performance boundary of an unmanned aerial vehicle (UAV) intelligent decision-making system as described in claim 1, 2, 3, or 4, characterized in that: It also includes step four, which, based on the mission qualification performance boundary and safe flight boundary determined in step three, obtains the range of various sensor errors that the aircraft's intelligent decision-making system can tolerate. This allows the system to determine whether the current operating condition is within the allowable range during actual flight. When the system operates under conditions that do not exceed the limits, it can ensure the reliability of its decision-making and control. Furthermore, clear performance boundaries can fundamentally prevent drones from being misused in unsuitable scenarios due to unclear intelligent algorithm capability boundaries, thus ensuring the effectiveness and safety of drone flight missions.

Images