A multi-layer dynamic security boundary solving method and system for an intelligent flight control system

Abstract

The application provides a kind of multi-layer dynamic security boundary solution method and system for intelligent flight control system, it is related to autonomous aircraft control and safety monitoring field, including the following steps: S1: multi-source state perception and parameter decoupling;S2: construct multi-layer security boundary unified decision function;S3: multi-dimensional safety margin dynamic evaluation;S4: control instruction correction based on control barrier function.The application constructs "physical-logic-information" three-layer unified representation model, integrates the original discrete safety index to unified decision space: system safety state;Three-layer decision operator is independent, not coupled, which is convenient for risk positioning;System safety state ensures that only when three layers are satisfied simultaneously, it is judged as safe.Compared with the traditional scheme covering only the physical layer, the application eliminates about 64% of the potential risk blind area.

Description

Technical Field

[0001] This invention relates to the field of autonomous aircraft control and safety monitoring, specifically to a multi-layer dynamic safety boundary calculation method and system for intelligent flight control systems. Background Technology

[0002] Autonomous flight control technology has undergone a profound transformation over the past two decades, evolving from manned assistance to full autonomy. Modern intelligent flight control systems typically include multiple functional modules such as perception, planning, decision-making, and control. The decision-making module generates reference control commands based on sensor-perceived information and mission objectives, while the control module drives the aircraft to perform corresponding actions through control law calculations.

[0003] In this technological context, the risks faced by flight safety monitoring have expanded from traditional physical risks such as engine failure, structural overload, and aerodynamic stall to logical risks such as mission sequence violations and phase switching errors, as well as information risks such as perception drift, planning oscillation, and the transmission of decision uncertainty.

[0004] Traditional flight envelope protection systems employ hard limiting strategies for angle of attack α, airspeed V, and overload. Constraining physical quantities, a typical structure involves forcibly activating a hard-limit loop to pull the aircraft back to a safe state when the constrained physical quantity exceeds a threshold. While this protection measure can simply and intuitively prevent the aircraft from being in a dangerous state, it still has the following systemic drawbacks: (1) Lack of predictive ability: Traditional methods are based solely on the current flight state and cannot predict the risk of exceeding limits that may be caused by dynamic coupling in the future. When the aircraft is in a high-maneuver state, even if the current state is within the envelope, it may cross the boundary due to inertial gliding within 1-3 seconds in the future; (2) Uncertainty-free modeling: Hard constraints may produce frequent "chattering" phenomena in sensor noise environments, leading to a severe degradation of control performance; (3) No information layer perception: It cannot identify the shrinkage of the effective control range caused by the decrease in perception confidence, and may generate false "safe" signals in the environment of perception degradation. Summary of the Invention

[0005] To address the shortcomings of the prior art, the present invention aims to provide a multi-layer dynamic safety boundary calculation method and system for intelligent flight control systems, thereby solving the problem in the prior art of how to model physical layer constraints, logical layer constraints, and information layer constraints in a unified mathematical framework so that the constraints of each layer can be independently determined and comprehensively evaluated.

[0006] Specifically, on the one hand, the present invention provides a multi-layer dynamic safety boundary calculation method for intelligent flight control systems, which includes the following steps: S1: Multi-source state perception and parameter decoupling: Real-time acquisition and decoupling of the flight control system's operating parameters and preprocessing them, dividing them into state vectors of three dimensions: physical layer, logical layer and information layer; S2: Construct a unified decision function for multi-layer security boundaries: Using the state vector obtained from S1 as input, construct a joint decision function that integrates the constraints of the physical layer, logic layer, and information layer. S3: Multi-dimensional dynamic assessment of safety margins: Based on the state vector obtained from S1, and combined with the safety boundaries corresponding to the decision operators at each layer in the joint decision function, the comprehensive safety margin of the system from each layer boundary is calculated in real time. : ; in, The distance between the current flight status and the stall or overload limit. To quantify the compliance level of the current operations and task sequence, To ensure the reliability of the information output by the decision-making module. This is the physical layer state vector, used to characterize the real-time kinematics and dynamics parameters of the aircraft; This is a logic layer state vector used to represent the execution state of the current flight phase and mission sequence constraints; This is the information layer state vector, used to characterize the reliability of perceived information and decision output; the upper right subscript... This represents the transpose operation of a vector, used here to concatenate three subvectors into a unified generalized state vector. ; S4: Control command correction based on control barrier function: When the overall safety margin If the value is not lower than the preset threshold, then the corrected instruction will be output. = ; When the overall safety margin When the value is below a preset threshold, activate instruction correction and generate a correction instruction. .

[0007] Furthermore, correction instructions are generated in S4. The methods include: Construct and solve the following quadratic programming problem: ; Its constraints are: ; ; in, This is the original reference instruction. For the nonlinear dynamics model function of the aircraft, To control the gain coefficient of the barrier function, and These are the lower and upper limits of the actuator output, respectively; Effective safety boundary function Defined as: ; in, This is the estimated distance from the aircraft to the nearest obstacle. The preset basic safety distance threshold, Let f be the confidence decay function that monotonically decreases with perceived confidence. Furthermore: Physical layer state vector Including: airspeed Angle of attack Sideslip angle and overload ; Logic layer state vector Including: Current flight phase instructions and task sequence constraints; Information layer state vector Including: the statistical confidence level of environmental perception results and the output stability index of path planning .

[0008] Furthermore, the methods for obtaining the three-dimensional state vector include: Physical layer state vector xp: Raw sensing data is collected in real time through the airborne sensor network, and then filtered, denoised, and time-synchronized to generate the physical layer state vector xp. Logic layer state vector : Obtain the current flight phase identifier from the mission management system Task sequence constraints Available action set Encoded as a logic layer state vector ; Information layer state vector Based on Bayesian filtering, the perceptual posterior distribution p(z|target) is maintained to determine the detection probability. Representation, obtaining perceptual confidence ; Calculate planning entropy ; in, The total number of candidate paths output by the path planning module. This represents the sequence number of the candidate path. , For the first The probability of a candidate path being selected, and satisfying the following conditions: ; Furthermore: In S2, the joint decision function is: ; in, , is the generalized state vector of the system; , , These represent the decision operators for the physical layer, logic layer, and information layer, respectively. This is an indicator function. It outputs 1 if and only if the logical expression within the parentheses is true, i.e., all layer constraints are satisfied; otherwise, it outputs 0. Represents the logical AND operation.

[0009] Furthermore: In S2, the decision function for the three-layer constraints is: ; in, This is the result of the overall system security status assessment; This represents the intersection operation.

[0010] Furthermore: Physical layer decision operator The calculation methods include: judgment Whether the upper and lower limits of each physical quantity are met, calculate. And determine Comprehensive output (within the envelope) ); in, This represents the current physical state vector of the flight control system; Indicates from the current moment Starting from this point, the prediction time domain length is... The state reachable set; This represents a pre-defined set of safety state constraints. Represents the empty set; Represents reachable set With safety constraint set The result of determining whether the intersection is empty, when The condition is true if true and false otherwise; the area within the envelope represents the physical state vector. All components are within the preset upper and lower limits of physical quantities; Logic layer decision operator The calculation method includes: encoding the current state as an LTL atomic proposition truth value assignment, and assigning each security constraint formula... Perform model validation and output the results. ＝ ; Wherein, LTL represents linear sequential logic; Indicates the first The timing logic formula for each security constraint; The truth value assignment of the atomic proposition corresponding to the current state satisfies the formula. ; This indicates a traversal of all safety constraint formulas; Information layer decision operator The calculation methods include: judgment , , Whether the conditions are met simultaneously, and the overall output. (All conditions are met); in, To perceive confidence level, The safety threshold for perceived confidence; For path planning entropy, The safety threshold for path planning entropy; To predict variance, The safety threshold for predicting variance; To continuously detect the number of failures, The safety threshold for the number of consecutive sensing failures; This indicates an indicator function that outputs a logical true or a binary 1 when the logical condition within the parentheses is true, and outputs a logical false or a binary 0 otherwise.

[0011] Furthermore: In S3, physical layer margin The calculation methods include: Take the minimum normalized distance from the current state to the boundaries of each envelope; Logical layer margin The calculation methods include: = ; in, This represents the total number of security constraint formulas at the logic layer. For the sequence number of the safety constraint formula, ; For the first Each logic layer security constraint formula is described using linear sequential logic. The satisfaction measurement function represents the current logical layer state. For the first Safety constraint formula The degree of satisfaction; Information layer margin The calculation methods include: ; in, The preset positive weighting coefficients satisfy... These are used to adjust the impact of perception confidence deficit, planning entropy, prediction variance, and consecutive failure count on the information layer margin, respectively. The confidence deficit function is defined as follows: This is used to characterize the degree to which the current perceived confidence is insufficient relative to the safety threshold. When the value is 0, The value is positive at time. The safety threshold for perceived confidence; The safety threshold for path planning entropy; The safety threshold for predicting variance; To continuously detect the number of failures; This indicates the reciprocal operation.

[0012] Furthermore: In S4, correction instructions are generated. The method includes the following steps: S41: Establish an effective safety boundary function: Based on the current physical layer state vector and perceived confidence Establish an effective safety boundary function: ; in, This is the estimated distance from the aircraft to the nearest obstacle. The preset basic safety distance threshold, Let f be the confidence decay function that monotonically decreases with perceived confidence. S42: Calculate the partial derivatives of the safety boundary with respect to the state: Calculate the effective safety boundary function For physical layer state vectors partial derivatives ; S43: Constructing control barrier function constraints: Based on the control barrier function theory, the following security constraints are constructed: ; in, For the nonlinear dynamics model function of the aircraft, To control the gain coefficient of the barrier function; this constraint ensures that the system state is always maintained by the gain coefficient of the barrier function. Within the defined security set; S44: Construct and solve the quadratic programming problem: With the objective of minimizing the deviation from the original reference command, and combining the CBF constraint and actuator physical limit constraint from step S4-3, a quadratic programming problem is constructed: S45: Output correction command: Solve the quadratic programming problem and use its optimal solution as the correction instruction. The corrected control commands are output to the flight control actuator.

[0013] This invention also provides a multi-layer dynamic safety boundary calculation system for intelligent flight control systems, used to execute a multi-layer dynamic safety boundary calculation method for intelligent flight control systems. The calculation system includes the following modules: Multi-sensor input layer module: used to receive multi-dimensional data, the sources of which include atmospheric data systems, inertial navigation systems, GNSS, obstacle detection sensors, and mission management systems; Independent decision-making modules: including physical layer decision-maker, logic layer decision-maker and information layer decision-maker; The comprehensive margin evaluator module receives the output of the independent decision module, calculates the normalized margin of each layer, and takes the minimum value to obtain the comprehensive margin. CBF-QP Corrector Module: Used to determine the current safety state based on the minimum safety margin, solve the quadratic programming problem with CBF constraints with the goal of minimizing the deviation, and output correction instructions.

[0014] Furthermore, the modules interact with each other via an onboard bus.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Comprehensive Security Dimensions: This invention integrates previously discrete security indicators into a unified judgment space—system security status—by constructing a unified three-layer representation model of "physical-logical-information". The three decision operators are independent and decoupled, facilitating risk location; system security status. Safety is only determined when all three layers are simultaneously satisfied. Based on 10,000 Monte Carlo simulations, this invention eliminates approximately 64% of potential risk blind spots compared to traditional solutions that only cover the physical layer.

[0016] (2) Refined Safety Margin: This invention achieves continuous quantitative characterization of the safety status through normalized margin indices. (Comprehensive Margin) The design ensures that risks at the weakest level are detected first. Compared to traditional binary judgment, this invention provides an early warning window, supporting a tiered response strategy.

[0017] (3) Explicit modeling of information uncertainty: This invention uses... The formula and confidence-weighted CBF establish, for the first time, an explicit propagation link from perceived uncertainty to the safety boundary. When the perceived confidence... When the threshold is lowered, the safety boundary automatically tightens, reflecting a robust control philosophy with moderate conservatism. Compared to traditional deterministic boundaries, this invention reduces the collision probability by approximately 78% in scenarios with diminishing perception.

[0018] (4) Flexible instruction correction: This invention utilizes the CBF-QP optimization framework to achieve flexible instruction correction. Let be the cost function, and satisfy... Under the premise of generating correction instructions u*, this invention preserves the original control intent to the greatest extent, compared to the hard limitations of existing technologies such as speed lock and attitude freeze. In typical scenarios, the control deviation is controlled within 12%, ensuring safety.

[0019] (5) High interpretability: This invention makes the source of risk traceable through the independent design of three-layer decision operators: physical layer decision operator Indicating physical boundary risk, the logic layer determines the operator. Indicates task logic risk; information layer judgment operator. This indicates a risk of information uncertainty. It provides pilots / monitors with clear warning information and a basis for fault diagnosis, meeting the interpretability requirements of aviation safety systems.

[0020] (6) Real-time performance is guaranteed: This invention adopts different real-time guarantee mechanisms through the design of three-layer decision-making: physical layer envelope decision-making is O(1) table lookup operation; logic layer LTL decision-making is O(1) table lookup operation (offline generation of lookup table); information layer margin calculation is O(1) algebra operation; CBF-QP solver has embedded QP accelerator, corrected delay <1ms, end-to-end processing delay <5ms, meeting the real-time requirement of response time <10ms for military and civilian aircraft. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the three-layer security boundary architecture of the present invention; Figure 2 This is a flowchart of the control instruction modification logic based on CBF of the present invention; Figure 3 This is a trajectory comparison curve of the UAV avoiding obstacles in an uncertain environment in an embodiment of the present invention. Detailed Implementation

[0022] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0023] 1. Formal definition of generalized state space: Generalized state space: Let Let X be the generalized state space of the flight control system, and let X be the generalized state vector. ,in: Physical layer state subspace Typical dimensions Physical layer state vector The physical meaning of each component is shown in the table below:

[0024] Logic layer state subspace Logic layer state ,in For the current stage identifier in the finite set {takeoff, climb, cruise, obstacle avoidance, descent, approach, landing}; For the task sequence constraint set; This represents the set of actions allowed at the current stage. The logic layer state space is essentially a hybrid space (discrete + continuous), which is mathematically encoded as a continuous vector form.

[0025] Information layer state subspace Information layer state vector The physical meaning of each component is shown in the table below:

[0026] 2. Mathematical definition of a three-level decision operator: Physical layer decision operator Determine whether the flight state is within the physical safety envelope and predicted to be safe: output 1 indicates safety, output 0 indicates that the flight has exceeded the limit or is predicted to exceed the limit. The physical layer state vector xp contains 9 components: airspeed V, angle of attack α, sideslip angle β, overload nz, three-axis angular rates p / q / r, and attitude angles φ / θ, covering the core parameters of aircraft dynamics.

[0027] Logic layer decision operator The flight mission is modeled as a finite state machine, and the mission sequence safety constraints are described using Linear Temporal Logic (LTL). The LTL syntax includes: atomic propositions a, always true ┬, always false ⊥, Boolean operators ¬ (NOT), ∧ (AND), ∨ (OR), and temporal operators □ (always satisfied), ◇ (eventually satisfied), U (satisfied until...), and X (satisfied in the next step). The output is 1 if and only if the current logic layer state satisfies all LTL safety constraint formulas; otherwise, it outputs 0.

[0028] Typical LTL safety constraint example: Prohibiting high-G maneuvers while the landing gear is unlocked can be modeled as φ1=□(¬ → ¬(nz> During the obstacle avoidance phase, the vehicle must ultimately exit and must not enter the restricted area; this can be modeled as φ2=□( → ◇¬ ).

[0029] Information layer decision operator Determine if the quality of decision information meets the safety threshold: Output 1 if and only if the perceived confidence level Planning entropy Prediction variance And the number of consecutive failures detected Established simultaneously. Each threshold is configured according to the system security level.

[0030] 3. Forward reachability analysis and physical layer determination: Forward reachability analysis predicts the set of reachable states R of the aircraft within a future time window T, determining whether a collision with terrain obstacles or structural strength envelope will occur. If the intersection of R and the collision set C is empty, the forward reachability determination is successful. (Safety), otherwise =0 (dangerous).

[0031] The reachable set is calculated using the polyhedral approximation method for linear time-varying systems. (where w is a bounded perturbation), the reachable set is recursively given by ,in , For the discretized system matrix, Let ⊕ be the perturbation accumulation set, and ⊕ be the Minkowski sum operation. This is achieved by setting an upper limit on the number of recursion steps. To control for approximation error, the error bound increases linearly with the number of steps. Collision detection is transformed into a linear programming problem with a solution complexity of O(n log n). On an airborne DSP, this can be completed within 100μs.

[0032] Physical layer comprehensive judgment (within the envelope) That is, the current state satisfies the dynamic envelope constraint ( , , The output is safe when the forward reachable set does not overlap with an obstacle.

[0033] 4. Comprehensive safety determination function: The results of the three-layer decision operators are fused into a unified security decision function through logical AND operations: ; The system security state is achieved if and only if the physical layer, logical layer, and information layer simultaneously satisfy the security constraints. =1 indicates that the system is in a safe state; otherwise =0 indicates that at least one level of decision has exceeded the limit. This function integrates the originally scattered risk indicators into a unified decision space, eliminating the risk blind spots caused by existing technologies that only cover physical layer constraints.

[0034] 5. Mathematical definition of multidimensional safety margin: Each level of margin extends the binary decision into a continuous quantity, enabling a refined representation of the safe state: Physical layer margin : Take the minimum normalized distance of each physical quantity (airspeed, angle of attack, overload, etc.) to the envelope boundary; =1 indicates that it is in the safe center. =0 indicates that the physical boundary has been reached.

[0035] Logical layer margin Satisfaction measurement based on LTL constraints, taking the satisfaction level of each constraint. The average value. When the atomic proposition is true, ρ = 1, and when it is false, ρ = 0; the ¬ operation ρ(¬a) = 1 - ρ(a); the ∧ operation ρ(a∧b) = min(ρ(a), ρ(b)).

[0036] Information layer margin It comprehensively reflects the perceived confidence deficit, planning entropy, prediction variance, and number of consecutive failures, thus quantifying the reliability of decision-making information.

[0037] The overall safety margin M_total(X, t) is the minimum value of the three-layer margins: ; This design ensures that the risks of the weakest layer are detected first. When = 0, at least one layer has reached the boundary. When the value is 1, all three layers are in the center of the security domain. It provides continuous quantitative early warning capabilities and supports tiered response strategies.

[0038] 6. Control barrier function and security correction: Safety boundary function For a control barrier function (CBF), its validity condition is: there exists a function α∈K_∞ such that And secure collection It is a nonempty closed set. When α takes the linear function α(h) = k·h (k>0), it degenerates into a nonempty closed set. This is the constraint form used in the CBF-QP of this invention.

[0039] The candidate CBF for obstacle avoidance scenarios is taken as ,in This is the estimated distance from the aircraft to the nearest obstacle. Based on the basic safety margin distance.

[0040] The confidence-weighted effective safety boundary function is: ; in, For the confidence decay function, it is recommended to take g(C) = 1 / C: when When the value decreases, g(C) increases. As it decreases, the equivalent safety boundary automatically tightens. When When the value is reduced from 1.0 to 0.4, the equivalent safety boundary expands by approximately 2.5 times, and the correction force is correspondingly strengthened. This mechanism embodies the robustness principle that "the more uncertain the perception, the more conservative the safety boundary."

[0041] when When <δ, a correction instruction is generated via the CBF-QP corrector to minimize For the objective, CBF constraint Given the condition, a quadratic programming problem with safety constraints is solved. This method preserves the original control intention to the greatest extent while ensuring safety, avoiding the drastic degradation of control performance caused by traditional hard constraints (velocity lock, attitude freeze).

[0042] This invention proposes a multi-layer dynamic safety boundary calculation system for intelligent flight control systems, such as... Figure 1 As shown, its overall architecture includes the following core components: Multi-source sensor input layer module: used to receive multi-dimensional data from atmospheric data systems, inertial navigation systems, GNSS, obstacle detection sensors, mission management systems, etc.

[0043] Three parallel independent decision modules: Physical layer determiner: It includes envelope boundary checks and forward reachability analysis.

[0044] Logic layer determiner: It contains a finite state machine and an LTL model checker.

[0045] Information layer decision maker: It contains Bayesian confidence estimation and entropy calculation.

[0046] The comprehensive margin estimator receives the outputs of the three decision modules, calculates the normalized margin of each layer, and takes the minimum value to obtain the comprehensive margin. .

[0047] CBF-QP Modifier Module: When the overall safety margin threshold When the time is right, the safety filter is activated, and the quadratic programming problem with CBF constraints is solved with the goal of minimizing the deviation, and the correction command is output.

[0048] The modules exchange data via an onboard bus.

[0049] This invention also provides a method for solving multi-layer dynamic safety boundaries for intelligent flight control systems, combined with Figure 1 and Figure 2 As shown: Figure 1 This is a schematic diagram of a three-layer safety boundary architecture. The leftmost part is the multi-source sensor input of the aircraft; the middle part consists of three parallel decision modules, which correspond to dynamic envelope constraints, mission timing rule constraints, and decision information quality constraints, respectively; the three converge in a unified decision space at the center, and finally output the decision result through a comprehensive margin solver.

[0050] Figure 2 The logic flowchart for CBF-based control instructions has been revised, with reference to the control instructions. Connect to the quadratic programming (QP) optimizer; the external feedback path introduces the current physical state vector xp into the safety boundary function h(x) calculation module; information layer confidence. The data is simultaneously fed into the confidence-weighted module to generate an effective safety boundary function. ; The constraints generated by the module are applied to the QP optimizer in real time, causing the optimizer to output correction instructions u* that satisfy the boundary conditions.

[0051] The mathematical description of this process is as follows: ; ; in, Here, k represents the nonlinear dynamics model of the aircraft, and k is the CBF gain coefficient (usually k∈[1,6], increasing in emergency mode and decreasing in normal mode) to ensure the positive invariance of the safety set. The QP optimizer solves the above optimization problem in each control cycle (typically 10ms) and outputs the correction instruction u*.

[0052] The technical solution of the present invention will be described in detail below with reference to specific embodiments and accompanying drawings.

[0053] Example 1 This embodiment selects a typical scenario where a drone encounters strong light interference during autonomous low-altitude flight in mountainous areas, causing drastic fluctuations in ranging information. This scenario can fully demonstrate the unique advantages of this invention in information layer risk management.

[0054] Scenario Description: The drone is performing a low-altitude cargo delivery mission, patrolling over mountainous areas. Flight environmental parameters are as follows: Initial altitude... =200m, cruising airspeed =28m / s, heading angle (Due north); There is a mountain obstacle 150m ahead, with a height protruding 80m from the terrain, forming a semi-cone restricted area; At t=5s, strong sunlight shines directly on the airborne camera, causing the range sensor to output a confidence level. The confidence level dropped sharply from 0.95 to 0.40, and returned to normal after 3 seconds. During the period of reduced confidence, the original control command still required high-speed flight based on inaccurate ranging information, which posed a risk of crashing into a mountain.

[0055] The flight envelope parameters are shown in the table below:

[0056] Based on this, the present invention provides a multi-layer dynamic safety boundary calculation method for intelligent flight control systems: S1. Input: The following parameters are fed back in real time by airborne sensors: Physical layer -- Airspeed V = 28 m / s, Angle of attack α = 4°, Sideslip angle β = 0.5°, Normal overload =1.2g, triaxial angular velocities p = 0.1rad / s, q = 0.05rad / s, r = 0.2rad / s, roll angle φ = 5°, pitch angle θ = 3°.

[0057] Logic layer -- Current flight phase =Cruising, mission sequence constraints =Obstacle avoidance has the highest priority; currently allowed action set = {Normal cruising, slight turning, deceleration}.

[0058] Information layer -- Confidence level of ranging information The path planning output entropy decreases from 0.95 to 0.4 (input signal quality deteriorates). =0.8 bits (the planning result shows slight fluctuations), prediction variance (Exceeding the normal threshold by 0.1), number of consecutive failures =1.

[0059] S2, Decision: Call the unified decision function. Perform judgments at each level: Physical layer decision operator Detection: The current state satisfies V∈[15, 35]m / s, α∈[-10°, 25°], Envelope constraints; Forward reachability analysis (time window T = 5s, upper limit of recursion steps) The predicted reachable set does not overlap with terrain obstacles; the overall judgment is Φ_{phys} = True (safe).

[0060] The logic layer decision operator Φ_{logic} checks: The system is currently in the cruise phase, obstacle avoidance priority is set correctly, and the finite state machine state transitions conform to preset safety sequence constraints (such as the LTL constraint φ=□(¬)). Comprehensive judgment =True (Safe).

[0061] Information layer decision operator Detection: =0.4 is below the safety threshold =0.6 (violation) Exceeding the normal threshold (Minor violation) =0.15 exceeds the fluctuation threshold (Violation); Comprehensive Judgment (Unsafe).

[0062] Final unified judgment result: (The system as a whole is insecure).

[0063] S3. Margin Calculation: Calculate the normalized margin for each layer: Physical layer margin (The current flight status still has a margin from the physical envelope boundary, and the value is taken as 0.75 after considering the forward reachability margin coefficient).

[0064] Logical layer margin (Based on the LTL satisfaction measurement method, the current state's satisfaction with the safety constraint φ is 0.9).

[0065] Information layer margin (Significantly reduced due to a sharp decrease in perceived confidence). The calculation formula is: Substitute Calculated Considering the penalty effect of a sharp drop in perceived confidence, the correction value is set to m_{info} = 0.3.

[0066] Overall safety margin: Since 0.3 < δ = 0.5 (the preset threshold, usually 0.5, which can be adjusted according to the security level), the system determines that intervention is necessary.

[0067] S4, Output: Original Reference Instructions Requirement: Maintain propulsion power at 80%. Roll angle command +15° (maintain current heading and perform minor lateral avoidance). Safety filter establishes effective safety boundary function. ;in Estimate the distance to the obstacle. The current perceived confidence level, This is the basic safety margin distance. Substituting it into the calculation yields... (Due to the decreased confidence level, the equivalent safety boundary expands significantly outward from the nominal value of 140m to 365m). Establish CBF constraints and solve the QP problem: Where k = 2 is the CBF gain coefficient (increased to 4-6 in emergency mode, decreased to 1 in normal mode). Solve this quadratic programming problem and output the correction command u*: propulsion power reduced from 80%. Reduced to 55% (Active deceleration), the roll angle command is adjusted from +15° to +8° (increasing the turning radius).

[0068] The flight performance comparison is shown in the table below:

[0069] like Figure 3 The figure shows a comparison of obstacle avoidance trajectory curves for a UAV under uncertain conditions. The figure includes the obstacle area, a reference trajectory generated by a traditional flight control system, and a corrected trajectory generated by this invention. The comparison curves demonstrate how this invention achieves smoother and safer obstacle avoidance by pre-decelerating and adjusting the turning radius when perception confidence decreases.

[0070] In the diagram, the horizontal axis represents flight time, and the vertical axis represents the approach distance between the aircraft and the obstacle. The solid line represents the traditional flight control reference trajectory, which rapidly approaches the obstacle area within 5-8 seconds after the strong light interference occurs, with the closest distance being only 12m. The dashed line represents the corrected trajectory of this invention, which maintains a safe distance of more than 35m by actively decelerating and increasing the turning radius in the early stages of interference.

[0071] Monte Carlo robustness verification: To verify the robustness of this invention under parameter uncertainties and environmental disturbances, N = 10,000 Monte Carlo simulations were performed. Simulation variables included: initial velocity disturbance. ±2m / s; Gust disturbance ∈[-0.5, 0.5]m / s 2 The perceived confidence truth value is ∈ [0.35, 0.45] (simulating strong light interference); the obstacle distance deviation is ±10m.

[0072] Simulation results: The traditional flight control system experienced 87 collisions (collision probability 0.87%), while only the physical CBF (body flow factor) experienced 3 collisions (collision probability 0.03%). The flight control system using the solution method of this invention experienced 0 collisions (collision probability <0.01%). The upper confidence bound (99% confidence level) of the collision probability of the flight control system using the solution method of this invention is less than 0.01%, meeting the safety target for Class A software (catastrophic failure) in the DO-178C standard (<10). -9 per flight hour).

[0073] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for solving multi-layer dynamic safety boundaries for intelligent flight control systems, characterized in that, It includes the following steps: S1: Multi-source state perception and parameter decoupling: Real-time acquisition and decoupling of the flight control system's operating parameters and preprocessing them, dividing them into state vectors of three dimensions: physical layer, logical layer and information layer; S2: Construct a unified decision function for multi-layer security boundaries: Using the state vector obtained from S1 as input, construct a joint decision function that integrates the constraints of the physical layer, logic layer, and information layer; S3: Multi-dimensional dynamic assessment of safety margins: Based on the state vector obtained from S1, and combined with the safety boundaries corresponding to the decision operators at each layer in the joint decision function, the comprehensive safety margin of the system from each layer boundary is calculated in real time. : ； in, The distance between the current flight status and the stall or overload limit. To quantify the compliance level of the current operations and task sequence, To ensure the reliability of the information output by the decision-making module. This is the physical layer state vector; This is the logic layer state vector; This represents the information layer state vector; S4: Control command correction based on control barrier function: When the overall safety margin If the value is not lower than the preset threshold, then the corrected instruction will be output. = ; When the overall safety margin When the value is below a preset threshold, activate instruction correction and generate a correction instruction. .

2. The multi-layer dynamic safety boundary calculation method for intelligent flight control systems as described in claim 1, characterized in that: In S4, the correction instructions are generated. The methods include: constructing and solving the following quadratic programming problem: ； Its constraints are: ； ； in, This is the original reference instruction. For the nonlinear dynamics model function of the aircraft, To control the gain coefficient of the barrier function, and These are the lower and upper limits of the actuator output, respectively; Effective safety boundary function Defined as: ； in, This is the estimated distance from the aircraft to the nearest obstacle. The preset basic safety distance threshold, It is a confidence decay function that monotonically decreases with perceived confidence.

3. The multi-layer dynamic safety boundary calculation method for intelligent flight control systems as described in claim 1, characterized in that: Methods for obtaining a three-dimensional state vector include: Physical layer state vector xp: includes airspeed Angle of attack Sideslip angle and overload The system collects raw sensing data in real time through an airborne sensor network, and performs filtering, noise reduction, and time synchronization processing to generate a physical layer state vector xp. Logic layer state vector Including instructions for the current flight phase. And task sequence constraints, obtain the current flight phase identifier from the task management system. Task sequence constraints Available action set Encoded as a logic layer state vector ; Information layer state vector : Including the statistical confidence level of environmental perception results and the output stability index of path planning ; Based on Bayesian filtering, the perceptual posterior distribution p(z|target) is maintained to determine the detection probability. Representation, obtaining perceptual confidence ; Calculate planning entropy ; in, The total number of candidate paths output by the path planning module. This represents the sequence number of the candidate path. , For the first The probability of a candidate path being selected, and satisfying the following conditions: .

4. The multi-layer dynamic safety boundary calculation method for intelligent flight control systems as described in claim 3, characterized in that: In S2, the joint decision function is: ； in, , is the generalized state vector of the system; , , These represent the decision operators for the physical layer, logic layer, and information layer, respectively. This is an indicator function that outputs 1 if and only if the logical expression within the parentheses is true, otherwise it outputs 0. Represents the logical AND operation.

5. The multi-layer dynamic safety boundary calculation method for intelligent flight control systems as described in claim 1, characterized in that: In S2, the decision function for the three-layer constraint is: ； in, This is the result of the overall system security status assessment; This represents the intersection operation.

6. The multi-layer dynamic safety boundary calculation method for intelligent flight control systems as described in claim 5, characterized in that: Physical layer decision operator The calculation methods include: judgment Whether the upper and lower limits of each physical quantity are met, calculate. And determine Comprehensive output (within the envelope) ); in, This represents the current physical state vector of the flight control system; Indicates from the current moment Starting from this point, the prediction time domain length is... The state reachable set; This represents a pre-defined set of safety state constraints. Represents the empty set; Represents reachable set With safety constraint set The result of determining whether the intersection is empty, when The condition is true if true and false otherwise; the area within the envelope represents the physical state vector. All components are within the preset upper and lower limits of physical quantities; Logic layer decision operator The calculation method includes: encoding the current state as an LTL atomic proposition truth value assignment, and assigning each security constraint formula... Perform model validation and output the results. ＝ ; Wherein, LTL represents linear sequential logic; Indicates the first The timing logic formula for each security constraint; The truth value assignment of the atomic proposition corresponding to the current state satisfies the formula. ; This indicates a traversal of all safety constraint formulas; Information layer decision operator The calculation methods include: judgment , , Whether the conditions are met simultaneously, and the overall output. (All conditions are met); in, To perceive confidence level, The safety threshold for perceived confidence; For path planning entropy, The safety threshold for path planning entropy; To predict variance, The safety threshold for predicting variance; To continuously detect the number of failures, The safety threshold for the number of consecutive sensing failures; This indicates an indicator function that outputs logical truth when the logical condition within the parentheses is true, and logical false otherwise.

7. The multi-layer dynamic safety boundary calculation method for intelligent flight control systems as described in claim 1, characterized in that: In S3, physical layer margin The calculation methods include: Take the minimum normalized distance from the current state to the boundaries of each envelope; Logical layer margin The calculation methods include: = ; in, This represents the total number of security constraint formulas at the logic layer. For the sequence number of the safety constraint formula, ; For the first Each logic layer security constraint formula is described using linear sequential logic. The satisfaction measurement function represents the current logical layer state. For the first Safety constraint formula The degree of satisfaction; Information layer margin The calculation methods include: ; in, The preset positive weighting coefficients satisfy... These are used to adjust the impact of perception confidence deficit, planning entropy, prediction variance, and consecutive failure count on the information layer margin, respectively. The confidence deficit function is defined as follows: This is used to characterize the degree to which the current perceived confidence is insufficient relative to the safety threshold. When the value is 0, The value is positive at time. The safety threshold for perceived confidence; The safety threshold for path planning entropy; The safety threshold for predicting variance; To continuously detect the number of failures; This indicates the reciprocal operation.

8. The multi-layer dynamic safety boundary calculation method for intelligent flight control systems as described in claim 2, characterized in that: In S4, the correction instructions are generated. The method includes the following steps: S41: Establish an effective safety boundary function: Based on the current physical layer state vector and perceived confidence Establish an effective safety boundary function: ; in, This is the estimated distance from the aircraft to the nearest obstacle. The preset basic safety distance threshold, Let f be the confidence decay function that monotonically decreases with perceived confidence. S42: Calculate the partial derivatives of the safety boundary with respect to the state: Calculate the effective safety boundary function For physical layer state vectors partial derivatives ; S43: Constructing control barrier function constraints: Based on the control barrier function theory, the following security constraints are constructed: ； in, For the nonlinear dynamics model function of the aircraft, To control the gain coefficient of the barrier function; this constraint ensures that the system state is always maintained by the gain coefficient of the barrier function. Within the defined security set; S44: Construct and solve the quadratic programming problem: With the goal of minimizing the deviation from the original reference command, a quadratic programming problem is constructed by combining the CBF constraint and the actuator physical limit constraint from step S4-3. S45: Output correction command: Solve the quadratic programming problem in S44 and use its optimal solution as the correction instruction. The corrected control commands are output to the flight control actuator.

9. A multi-layer dynamic safety boundary calculation system for intelligent flight control systems, characterized in that: The solution system for performing the solution method according to any one of claims 1-8 includes the following modules: Multi-sensor input layer module: used to receive multi-dimensional data, the sources of which include atmospheric data systems, inertial navigation systems, GNSS, obstacle detection sensors, and mission management systems; Independent decision-making modules: including physical layer decision-maker, logic layer decision-maker and information layer decision-maker; The comprehensive margin evaluator module receives the output of the independent decision module, calculates the normalized margin of each layer, and takes the minimum value to obtain the comprehensive margin. CBF-QP Corrector Module: Used to determine the current safety state based on the minimum safety margin, solve the quadratic programming problem with CBF constraints with the goal of minimizing the deviation, and output correction instructions.

10. The multi-layer dynamic safety boundary calculation system for intelligent flight control systems as described in claim 9, characterized in that: The modules exchange data via an onboard bus.

Images