Safety supervision method and system for flight state of unmanned aerial vehicle

Abstract

The invention discloses a safety supervision method and system for the flight state of an unmanned aerial vehicle, and relates to the technical field of unmanned aerial vehicle safety supervision. Flight behavior characteristics are compared with a global reference generated through federal learning, and when a malicious unmanned aerial vehicle passes authentication through a forged electronic identifier, a positioning verification mechanism is triggered in real time; illegal invasion is effectively identified and intercepted, and the problem of response delay caused by dependence on manual checking in the prior art is solved; meanwhile, the judgment standard is automatically tightened according to the safety level of the no-fly zone, the core zone adopts a stricter threshold amplitude, and the disguise unmanned aerial vehicle is forced to be exposed abnormally; the threshold value is in smooth transition during cross-regional flight, so that the misoperation risk caused by sudden change of a judgment standard in a traditional scheme is avoided; a hardware isolation area operation behavior analysis model of the unmanned aerial vehicle flight control chip is reused, and a calculation module does not need to be newly added; existing positioning base stations are dynamically dispatched to construct a verification network, and the reconstruction cost of supervision infrastructures is remarkably reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of safety supervision of unmanned aerial vehicles (UAVs), and in particular to a safety supervision method and system for the flight status of UAVs. Background Art

[0002] Current drone monitoring systems generally use electronic identification authentication combined with geo-fencing technology to monitor flight status in real time through a monitoring platform. In no-fly zones such as airports and military bases, drones are required to report encrypted IDs and location data, and can only be released after platform verification. Such solutions rely on hardware-level trusted execution environments (TEEs) to ensure ID authenticity. However, the core vulnerabilities of the current solutions lie in signal spoofing and passive response mechanisms. Malicious users can copy legitimate drone ID sequences and simulate signals through software radio to deceive the monitoring platform.

[0003] Although the existing system can trigger abnormal alarms, it requires manual verification of the authenticity of the signal source, and the response delay cannot meet the needs; in addition, the traditional behavior analysis model relies on cloud computing and cannot compare the rationality of the ID and flight trajectory in real time; for example, when a drone with a disguised ID suddenly enters a no-fly zone, the system can only mark it as suspicious but cannot intercept it immediately.

[0004] To address the above issues, some solutions have introduced blockchain-based ID storage or multi-base station triangulation to verify the location of the signal source. However, blockchain only ensures that ID registration cannot be tampered with and cannot prevent the duplication of signals during flight. Positioning verification requires additional infrastructure, which increases costs sharply and has insufficient coverage in mountainous areas. It can be seen that the current solutions still lack the ability to counter malicious camouflage in real time. Therefore, a safety supervision solution for drone flight status is urgently needed to solve such problems. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides a method and system for safely supervising the flight status of unmanned aerial vehicles (UAVs), which solves the problem that existing supervision relies on electronic identification authentication that is easy to forge, and the passive response mechanism is difficult to prevent malicious UAVs from invading no-fly zones in real time.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, an embodiment of the present invention provides a method for safely supervising the flight status of a drone, which includes:

[0009] Step S1: Build a distributed collaborative learning framework, access the flight control module of the registered UAV, and collect its flight behavior time series data in real time;

[0010] Step S2: generating an individual behavior baseline model at a local computing node based on the time series data, wherein the behavior baseline model includes acceleration change entropy, heading angle deviation frequency, and hovering stability index;

[0011] Step S3: uploading the encrypted characteristic parameters of the individual behavior baseline model to the supervision platform, and aggregating them to generate a global behavior baseline through a differential privacy mechanism;

[0012] Step S4: When the monitoring platform detects that the electronic identification authentication of the target drone has been passed but the real-time behavior data deviates from the global behavior benchmark by more than the dynamic threshold, the multi-base station signal source positioning verification is automatically activated;

[0013] Step S5: If the verification fails, an emergency obstacle avoidance command is immediately sent to the target UAV, and the radar tracking system in the adjacent area is triggered.

[0014] As a preferred solution of the method for safely supervising the flight status of a drone according to the present invention, the individual behavior baseline model in step S2 is generated as follows:

[0015] A lightweight isolation forest algorithm is embedded in the UAV's local flight control system to analyze continuous frame flight status data using a sliding window method and output a dynamically updated behavior feature vector.

[0016] The sliding window method is specifically as follows: intercepting flight status data frames at fixed time intervals, and the window length is adaptively adjusted according to the maneuverability strength of the UAV.

[0017] As a preferred solution of the safety supervision method of the drone flight status described in the present invention, the acceleration change entropy value is calculated by extracting the standard deviation sequence of E consecutive acceleration sampling points and quantifying the degree of fluctuation dispersion through the information entropy formula, where E is dynamically determined by the current flight speed of the drone.

[0018] As a preferred solution of the method for safely supervising the flight status of a drone according to the present invention, in step S1, the step of generating an individual behavior baseline model at a local computing node includes:

[0019] The real-time flight state vector s(t) is injected into the adaptive sliding window, and the window length is dynamically adjusted according to the maneuverability intensity. The formula is:

[0020] L w (t) = L0[1-αL m (t)],

[0021] Among them, L w (t) is the sliding window length at time t, L0 is the standard window length, α is the upper limit of the window shortening coefficient, I m(t) is the normalized value of mobility intensity, ranging from 0 to 1, and t is the time index;

[0022] At length L w (t) is collected in the window acceleration sampling points;

[0023] Where E(t) is the number of sampling points, f s is the acceleration sampling frequency, in Hz, is the rounding symbol;

[0024] For acceleration sequence Calculate the local standard deviation σ k , and then use information entropy to quantify the fluctuation discreteness, expressed as:

[0025]

[0026] Among them, H a (t) is the entropy value of acceleration change, in bits, k is the sampling point number, p k is the kth normalized probability density, σ k is the kth local standard deviation, in m / s 2 , j is the sum index;

[0027] Heading angle difference sequence Δψ k =ψ k -ψ k-1 After fast Fourier transform, take the main peak frequency:

[0028]

[0029] Among them, F ψ (t) is the main frequency of heading angle deviation, in Hz, f is the frequency independent variable, is the FFT operator, Δψ k is the kth heading angle increment, ψ k represents the original heading angle of the kth frame, ψ k-1 The heading angle of the previous frame, in radians;

[0030] Hover stability is measured by the ratio of the vertical velocity to the horizontal velocity modulus variance:

[0031]

[0032] Among them, S h (t) is the hovering stability index, Var(·) is the variance operator, is the vertical velocity in m·s -1 , v xy is the horizontal velocity modulus;

[0033] The three features are combined into a behavior vector by column: x(t) = [H a (t),F ψ (t),S h (t)] T ,

[0034] Where x(t) is the three-dimensional behavior feature vector at time t, and T is the transposition symbol;

[0035] After the lightweight isolation forest is trained online, the anomaly score is calculated for x(t) using the following formula:

[0036]

[0037] Among them, S o (t) is the anomaly score, ranging from 0 to 1, h(x(t)) is the sample average path length, c(n) is the theoretical average path length constant, H(n-1) is the n-1th harmonic number, and n is the number of observed samples;

[0038] After gating the anomaly score, the baseline vector is updated in an exponential decay manner. The update method is:

[0039]

[0040] Where B(t) is the baseline vector at time t, λ is the exponential decay coefficient, is the indicator function, when S o When (t)>τ, it takes 1, otherwise it takes 0, τ is the abnormal threshold with a value range of (0–1), and t is the time index;

[0041] The attenuation coefficient is adaptive to random dynamics, λ=λ0[1-βI m (t)],

[0042] Among them, λ is the current attenuation coefficient, λ0 is the static smoothing coefficient, and β is the weight compression coefficient.

[0043] As a preferred solution of the method for safely monitoring the flight status of a drone according to the present invention, the differential privacy mechanism in step S3 specifically includes:

[0044] A parameter perturbation engine is deployed on the regulatory platform, which adds Gaussian noise to the received encrypted feature parameters and performs federated averaging calculation to generate a global behavioral benchmark that is resistant to reverse engineering.

[0045] As a preferred solution of the method for safely monitoring the flight status of a drone according to the present invention, the dynamic threshold is set based on:

[0046] No-fly zone safety level;

[0047] The density of certified drones in the current airspace;

[0048] Frequency of historical malicious behavior incidents

[0049] The safety level of the no-fly zone is divided into three categories according to airspace management regulations: core area, buffer area and monitoring area.

[0050] As a preferred solution of the method for safely monitoring the flight status of a drone according to the present invention, the dynamic threshold value in step S4 is set as follows:

[0051] Dynamically adjust the threshold amplitude based on the no-fly zone geofence level;

[0052] The magnitude of the dynamic threshold is negatively correlated with the safety level of the no-fly zone.

[0053] As a preferred embodiment of the method for safely monitoring the flight status of a drone according to the present invention, in the dynamic threshold setting process of step S4, the threshold amplitude is dynamically adjusted according to the geofence level of the no-fly zone, including:

[0054] Divide the geofence into monitoring areas according to security level L g =1, buffer zone L g =2, core area L g =3; To maintain a negative correlation between the threshold amplitude and the security level, first set the exponential decay coefficient:

[0055] ρ(L g )=exp[-μ(L g -2)],

[0056] Among them, ρ(L g ) indicates level L g The scaling factor for the reference amplitude, μ is the decay rate constant, L g is the geo-fence security level, exp is the natural exponential function;

[0057] Apply the attenuation coefficient to the base amplitude Θ ref , get the real-time threshold amplitude Θ(t) = Θ ref ρ(L g ),

[0058] Among them, Θ(t) is the threshold amplitude at time t, Θ ref is the buffer reference amplitude, t is the time index;

[0059] Introduce first-order recursive smoothing:

[0060] Θ f (t) = γΘ(t) + (1-γ)Θ f (t-1),

[0061] Among them, Θ f (t) is the final threshold amplitude after smoothing, γ is the smoothing weight coefficient (0–1), Θ f (t-1) is the smoothing amplitude at the previous moment.

[0062] In a second aspect, the present invention provides a safety monitoring system for the flight status of a drone, comprising:

[0063] A local behavior analysis module deployed on the drone is used to calculate behavior feature vectors in real time. The local behavior analysis module is integrated into the isolated security zone of the flight control chip and includes a hardware acceleration unit and a protected storage area.

[0064] A federated learning engine in the regulatory platform to aggregate distributed parameters and generate a global behavioral benchmark;

[0065] Dynamic verification gateway, connecting to signal source positioning base station and radar tracking system;

[0066] Command execution interface, sending encrypted control commands to the target drone.

[0067] As a preferred solution of the safety supervision system for the flight status of a drone described in the present invention, the dynamic verification gateway pre-stores an airspace-base station mapping table, and selects the three base station combinations with the highest signal strength according to the latitude and longitude coordinates of the target drone and the base station load status.

[0068] The beneficial effects of the present invention are as follows: by comparing flight behavior characteristics with the global benchmark generated by federated learning, the present invention triggers a real-time positioning verification mechanism when a malicious drone passes authentication using a forged electronic identification, effectively identifying and intercepting illegal intrusions, and resolving the response delay problem caused by the existing technology's reliance on manual verification. At the same time, the present invention automatically tightens the judgment criteria based on the safety level of the no-fly zone, adopting a stricter threshold amplitude in the core area to force disguised drones to reveal abnormalities. The threshold transitions smoothly during cross-zone flight, avoiding the risk of misoperation caused by sudden changes in the judgment criteria in traditional solutions.

[0069] The invention reuses the hardware isolation zone operation behavior analysis model of the UAV flight control chip, without the need for a new computing module; dynamically schedules existing positioning base stations to build a verification network, significantly reducing the cost of regulatory infrastructure transformation; uses encryption aggregation technology to protect individual data, and behavioral characteristics only extract abstract motion indicators (such as acceleration fluctuations and heading change frequency), eliminating the risk of user privacy leakage; and through an adaptive analysis window and weight adjustment mechanism, maintains feature recognition accuracy in scenarios such as strong wind disturbances and low-altitude obstacle interference, reducing misjudgments and missed judgments. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0071] Figure 1 Schematic diagram of the flow of the method for safely supervising the flight status of a drone in Example 1.

[0072] Figure 2 This is a schematic diagram of the framework of the safety supervision system for the flight status of the drone in Example 1. DETAILED DESCRIPTION

[0073] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0074] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0075] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0076] Example 1, with reference to Figure 1 and Figure 2 This embodiment provides a method for safely monitoring the flight status of a drone, including the following steps:

[0077] Step S1: Build a distributed collaborative learning framework, access the flight control module of the registered UAV, and collect its flight behavior time series data in real time;

[0078] Step S2: Based on the time series data, an individual behavior baseline model is generated at the local computing node. The behavior baseline model includes acceleration change entropy, heading angle deviation frequency, and hovering stability index.

[0079] The individual behavior baseline model in step S2 is generated as follows:

[0080] A lightweight isolation forest algorithm is embedded in the UAV's local flight control system to analyze continuous frame flight status data using a sliding window method and output a dynamically updated behavior feature vector.

[0081] The sliding window method is as follows: the flight status data frame is intercepted at a fixed time interval, and the window length is adaptively adjusted according to the maneuverability intensity of the UAV. During high-intensity maneuvers, the window is shortened to 50% of the standard value.

[0082] The calculation method of acceleration change entropy is as follows: extract the standard deviation sequence of E consecutive acceleration sampling points and quantify its fluctuation discreteness through the information entropy formula, where E is dynamically determined by the current flight speed of the UAV;

[0083] In step S1, the steps of generating an individual behavior baseline model on a local computing node include:

[0084] The real-time flight state vector s(t) is injected into the adaptive sliding window, and the window length is dynamically adjusted according to the maneuverability intensity. The formula is:

[0085] L w (t) = L0[1-αL m (t)],

[0086] Among them, L w (t) is the sliding window length at time t, L0 is the standard window length, α is the upper limit of the window shortening coefficient, I m (t) is the normalized value of mobility intensity, ranging from 0 to 1, and t is the time index;

[0087] At length L w (t) is collected in the window acceleration sampling points;

[0088] Where E(t) is the number of sampling points, f s is the acceleration sampling frequency, in Hz, is the rounding symbol;

[0089] For acceleration sequence Calculate the local standard deviation σ k , and then use information entropy to quantify the fluctuation discreteness, expressed as:

[0090]

[0091] Among them, H a (t) is the entropy value of acceleration change, in bits, k is the sampling point number, p k is the kth normalized probability density, σ k is the kth local standard deviation, in m / s 2 , j is the sum index;

[0092] Heading angle difference sequence Δψ k =ψ k -ψk-1 After fast Fourier transform, take the main peak frequency:

[0093]

[0094] Among them, F ψ (t) is the main frequency of heading angle deviation, in Hz, f is the frequency independent variable, is the FFT operator, Δψ k is the kth heading angle increment, ψ k represents the original heading angle of the kth frame, ψ k-1 The heading angle of the previous frame, in radians;

[0095] Hover stability is measured by the ratio of the vertical velocity to the horizontal velocity modulus variance:

[0096]

[0097] Among them, S h (t) is the hovering stability index, Var(·) is the variance operator, is the vertical velocity in m·s -1 , v xy is the horizontal velocity modulus;

[0098] The three features are combined into a behavior vector by column: x(t) = [H a (t),F ψ (t),S h (t)] T ,

[0099] Where x(t) is the three-dimensional behavior feature vector at time t, and T is the transposition symbol;

[0100] After the lightweight isolation forest is trained online, the anomaly score is calculated for x(t) using the following formula:

[0101]

[0102] Among them, S o (t) is the anomaly score, ranging from 0 to 1, h(x(t)) is the sample average path length, c(n) is the theoretical average path length constant, H(n-1) is the n-1th harmonic number, and n is the number of observed samples;

[0103] After gating the anomaly score, the baseline vector is updated in an exponential decay manner. The update method is:

[0104]

[0105] Where B(t) is the baseline vector at time t, λ is the exponential decay coefficient, is the indicator function, when So When (t)>τ, it takes 1, otherwise it takes 0, τ is the abnormal threshold with a value range of (0–1), and t is the time index;

[0106] The attenuation coefficient is adaptive to random dynamics, λ=λ0[1-βI m (t)],

[0107] Among them, λ is the current attenuation coefficient, λ0 is the static smoothing coefficient, and β is the weight compression coefficient;

[0108] Specifically, the process uses maneuverability intensity to drive dual adaptation of windows and weights, and synchronizes the time scale with the convergence speed, which can maintain feature resolution accuracy between the two extreme states of violent maneuvering and stable hovering. Acceleration entropy measures longitudinal attitude fluctuations, the heading angle main frequency captures the plane turning rhythm, and the hovering variance ratio complements the vertical stability evaluation. The three complement each other to weaken the distortion effect of a single feature. The computational complexity of the isolation forest in low-dimensional space increases linearly with the sample size, making it suitable for embedded implementation. Its anomaly score is gated and then exponentially smoothed, which can quickly absorb regular patterns while maintaining inertial suppression of occasional anomalies, reducing the probability of false triggering. All calculations are completed under hardware isolation in the safe zone, eliminating the impact of external communication delays on baseline updates and providing high-confidence local features for subsequent federated aggregation.

[0109] Step S3: Upload the encrypted feature parameters of the individual behavior baseline model to the supervision platform and aggregate them to generate a global behavior baseline through the differential privacy mechanism;

[0110] The differential privacy mechanism in step S3 specifically includes:

[0111] Deploy a parameter perturbation engine on the regulatory platform, add Gaussian noise to the received encrypted feature parameters, and then perform federated averaging calculations to generate a global behavioral benchmark that is resistant to reverse engineering.

[0112] Step S4: When the monitoring platform detects that the electronic identification authentication of the target drone has been passed but the real-time behavior data deviates from the global behavior benchmark by more than the dynamic threshold, the multi-base station signal source positioning verification is automatically activated;

[0113] The basis for setting dynamic thresholds includes:

[0114] No-fly zone safety level;

[0115] The density of certified drones in the current airspace;

[0116] Frequency of historical malicious behavior incidents

[0117] The safety level of no-fly zones is divided into three categories according to airspace management regulations: core zone, buffer zone and monitoring zone;

[0118] The dynamic threshold value in step S4 is set as follows:

[0119] Dynamically adjust the threshold amplitude based on the no-fly zone geofence level;

[0120] The amplitude of the dynamic threshold is negatively correlated with the safety level of the no-fly zone, where the amplitude of the core zone threshold is 60% of the buffer zone;

[0121] During the dynamic threshold setting process of step S4, the threshold amplitude is dynamically adjusted according to the no-fly zone geofence level, including:

[0122] Divide the geofence into monitoring areas according to security level L g =1, buffer zone L g =2, core area L g =3; To maintain a negative correlation between the threshold amplitude and the security level, first set the exponential decay coefficient:

[0123] ρ(L g )=exp[-μ(L g -2)],

[0124] Among them, ρ(L g ) indicates level L g The scaling factor for the reference amplitude, μ is the decay rate constant, L g is the geo-fence security level, exp is the natural exponential function;

[0125] Apply the attenuation coefficient to the base amplitude Θ ref , get the real-time threshold amplitude Θ(t) = Θ ref ρ(L g ),

[0126] Among them, Θ(t) is the threshold amplitude at time t, Θ ref is the buffer reference amplitude, t is the time index;

[0127] In order to avoid drastic changes in the threshold when switching across zones, first-order recursive smoothing is introduced:

[0128] Θ f (t) = γΘ(t) + (1-γ)Θ f (t-1),

[0129] Among them, Θ f (t) is the final threshold amplitude after smoothing, γ is the smoothing weight coefficient (0–1), Θ f (t-1) is the smoothing amplitude at the previous moment;

[0130] When the system detects a change in the geofence level, only L g You can adjust Θ synchronously f(t), thus maintaining the requirement that the core area threshold is 60% of the buffer area, and taking μ = ln(1 / 0.6) can satisfy this proportional relationship;

[0131] Specifically, the exponential decay coefficient makes the threshold amplitude decrease monotonically with the safety level, avoiding linear mapping that is too loose in high-level areas or too strict in low-level areas. The decay rate constant is back-calculated by the 0.6 ratio given by the core and buffer zones to ensure that the value conforms to the zoning specifications. If the standard needs to be adjusted, only μ needs to be recalibrated. Recursive smoothing provides temporal continuity. When flying across regions or when airspace is redrawn, the threshold changes gradually to reduce false positive peaks.

[0132] Step S5: If the verification fails, an emergency obstacle avoidance command is immediately sent to the target UAV, and the radar tracking system in the adjacent area is triggered.

[0133] This embodiment also provides a safety monitoring system for the flight status of a drone, including:

[0134] The local behavior analysis module deployed on the drone is used to calculate the behavior feature vector in real time. The local behavior analysis module is integrated into the isolated security zone of the flight control chip and includes a hardware acceleration unit and a protected storage area.

[0135] A federated learning engine in the regulatory platform to aggregate distributed parameters and generate a global behavioral benchmark;

[0136] Dynamic verification gateway, connecting to signal source positioning base station and radar tracking system;

[0137] Dynamically verify the gateway's pre-stored airspace-base station mapping table and select the three base station combinations with the highest signal strength based on the target drone's latitude and longitude coordinates and base station load status;

[0138] Command execution interface, sending encrypted control commands to the target drone.

[0139] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for safely monitoring the flight status of an unmanned aerial vehicle, characterized in that: include, Step S1: Build a distributed collaborative learning framework, access the flight control module of the registered UAV, and collect its flight behavior time series data in real time; Step S2: generating an individual behavior baseline model at a local computing node based on the time series data, wherein the behavior baseline model includes acceleration change entropy, heading angle deviation frequency, and hovering stability index; Step S3: uploading the encrypted characteristic parameters of the individual behavior baseline model to the supervision platform, and aggregating them to generate a global behavior baseline through a differential privacy mechanism; Step S4: When the monitoring platform detects that the electronic identification authentication of the target drone has been passed but the real-time behavior data deviates from the global behavior benchmark by more than the dynamic threshold, the multi-base station signal source positioning verification is automatically activated; Step S5: If the verification fails, an emergency obstacle avoidance command is immediately sent to the target UAV, and the radar tracking system in the adjacent area is triggered.

2. A method for safely monitoring the flight status of a drone according to claim 1, characterized in that: The individual behavior baseline model in step S2 is generated as follows: A lightweight isolation forest algorithm is embedded in the UAV's local flight control system to analyze continuous frame flight status data using a sliding window method and output a dynamically updated behavior feature vector. The sliding window method is specifically as follows: intercepting flight status data frames at fixed time intervals, and the window length is adaptively adjusted according to the maneuverability strength of the UAV.

3. A method for safely monitoring the flight status of an unmanned aerial vehicle according to claim 2, characterized in that: The acceleration change entropy value is calculated by extracting the standard deviation sequence of E consecutive acceleration sampling points and quantifying the degree of its fluctuation dispersion using the information entropy formula, where E is dynamically determined by the current flight speed of the drone.

4. A method for safely monitoring the flight status of a drone as claimed in claim 3, characterized in that: In step S1, the step of generating an individual behavior baseline model at a local computing node includes: The real-time flight state vector s(t) is injected into the adaptive sliding window, and the window length is dynamically adjusted according to the maneuverability intensity. The formula is: L w (t)=L0[1-αI m (t)], Among them, L w (t) is the sliding window length at time t, L0 is the standard window length, α is the upper limit of the window shortening coefficient, I m (t) is the normalized value of mobility intensity, ranging from 0 to 1, and t is the time index; At length L w (t) is collected in the window acceleration sampling points; Where E(t) is the number of sampling points, f s is the acceleration sampling frequency, in Hz, is the rounding symbol; For acceleration sequence Calculate the local standard deviation σ k , and then use information entropy to quantify the fluctuation discreteness, expressed as: Among them, H a (t) is the entropy value of acceleration change, in bits, k is the sampling point number, p k is the kth normalized probability density, σ k is the kth local standard deviation, in m / s 2 , j is the sum index; Heading angle difference sequence Δψ k =ψ k -ψ k-1 After fast Fourier transform, take the main peak frequency: Among them, F ψ (t) is the main frequency of heading angle deviation, in Hz, f is the frequency independent variable, is the FFT operator, Δψ k is the kth heading angle increment, ψ k represents the original heading angle of the kth frame, ψ k-1 The heading angle of the previous frame, in radians; Hover stability is measured by the ratio of the vertical velocity to the horizontal velocity modulus variance: Among them, S h (t) is the hovering stability index, Var(·) is the variance operator, is the vertical velocity in m·s -1 , v xy is the horizontal velocity modulus; The three features are combined into a behavioral vector by column: Among them, x(t) is the three-dimensional behavior feature vector at time t, is the transpose symbol; After the lightweight isolation forest is trained online, the anomaly score is calculated for x(t) using the following formula: Among them, S o (t) is the anomaly score, ranging from 0 to 1, h(x(t)) is the sample average path length, c(n) is the theoretical average path length constant, H(n-1) is the n-1th harmonic number, and n is the number of observed samples; After gating the anomaly score, the baseline vector is updated in an exponential decay manner. The update method is: Where B(t) is the baseline vector at time t, λ is the exponential decay coefficient, is the indicator function, when S o When (t)>τ, it takes 1, otherwise it takes 0, τ is the abnormal threshold with a value range of (0–1), and t is the time index; The attenuation coefficient is adaptive to random dynamics, λ=λ0[1-βI m (t)], Among them, λ is the current attenuation coefficient, λ0 is the static smoothing coefficient, and β is the weight compression coefficient.

5. The method for safely monitoring the flight status of a drone according to claim 1, wherein: The differential privacy mechanism in step S3 specifically includes: A parameter perturbation engine is deployed on the regulatory platform, which adds Gaussian noise to the received encrypted feature parameters and performs federated averaging calculation to generate a global behavioral benchmark that is resistant to reverse engineering.

6. The method for safely monitoring the flight status of a drone according to claim 1, wherein: The dynamic threshold is set based on: No-fly zone safety level; The density of certified drones in the current airspace; Frequency of historical malicious behavior incidents The safety level of the no-fly zone is divided into three categories according to airspace management regulations: core area, buffer area and monitoring area.

7. A method for safely monitoring the flight status of a drone as claimed in claim 6, characterized in that: The dynamic threshold value in step S4 is set as follows: Dynamically adjust the threshold amplitude based on the no-fly zone geofence level; The magnitude of the dynamic threshold is negatively correlated with the safety level of the no-fly zone.

8. A method for safely monitoring the flight status of a drone as claimed in claim 7, characterized in that: During the dynamic threshold setting process of step S4, the threshold amplitude is dynamically adjusted according to the no-fly zone geofence level, including: Divide the geofence into monitoring areas according to security level L g =1, buffer zone L g =2, core area L g =3; To maintain a negative correlation between the threshold amplitude and the security level, first set the exponential decay coefficient: ρ(L g )=exp[-μ(L g -2)], Among them, ρ(L g ) indicates level L g The scaling factor for the reference amplitude, μ is the decay rate constant, L g is the geo-fence security level, exp is the natural exponential function; Apply the attenuation coefficient to the base amplitude Θ ref , get the real-time threshold amplitude Θ(t) = Θ ref ρ(L g ), Among them, Θ(t) is the threshold amplitude at time t, Θ ref is the buffer reference amplitude, t is the time index; Introduce first-order recursive smoothing: Θ f (t)=γΘ(t)+(1-γ)Θ f (t-1), Among them, Θ f (t) is the final threshold amplitude after smoothing, γ is the smoothing weight coefficient (0–1), Θ f (t-1) is the smoothing amplitude at the previous moment.

9. A safety monitoring system for the flight status of an unmanned aerial vehicle, based on a safety monitoring method for the flight status of an unmanned aerial vehicle according to any one of claims 1 to 8, characterized in that: include: A local behavior analysis module deployed on the drone is used to calculate behavior feature vectors in real time. The local behavior analysis module is integrated into the isolated security zone of the flight control chip and includes a hardware acceleration unit and a protected storage area. A federated learning engine in the regulatory platform to aggregate distributed parameters and generate a global behavioral benchmark; Dynamic verification gateway, connecting to signal source positioning base station and radar tracking system; Command execution interface, sending encrypted control commands to the target drone.

10. A safety monitoring system for the flight status of an unmanned aerial vehicle according to claim 9, characterized in that: The dynamic verification gateway pre-stores an airspace-base station mapping table and selects a combination of three base stations with the highest signal strength based on the latitude and longitude coordinates of the target drone and the base station load status.

Images