Unmanned aerial vehicle safety interval calibration method and system based on collision risk prediction

Abstract

The invention discloses an unmanned aerial vehicle safety interval calibration method and system based on collision risk prediction, and relates to the technical field of unmanned aerial vehicle flight management, and the method comprises the steps: obtaining an initial state parameter and an inherent size of a target unmanned aerial vehicle, and obtaining a historical positioning error set of the target unmanned aerial vehicle and a historical airspace error set of a target airspace; obtaining an initial collision risk domain based on the inherent size, and adjusting the initial collision risk domain based on the initial state parameter and the motion intention of the target unmanned aerial vehicle to obtain a collision risk domain; performing error prediction based on the historical positioning error set and the historical airspace error set to obtain an error risk parameter; and scaling the collision risk domain based on the error risk parameter to obtain the unmanned aerial vehicle safety interval. The technical problem of poor accuracy caused by the fact that unmanned aerial vehicle safety interval calibration cannot be matched with the risk in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) flight management technology, and more specifically to a method and system for calibrating safe intervals for UAVs based on collision risk prediction. Background Technology

[0002] In UAV operation and management, safety interval calibration is one of the core issues in ensuring flight safety and airspace utilization. Existing methods are usually based on fixed geometric models and simplified probabilistic assumptions, calibrating by constructing spherical or ellipsoidal collision domains. However, in actual operation, UAVs have diverse motion intentions, and the risk distribution in different modes such as hovering, low-speed cruise, and high-speed sprint exhibits significant anisotropy. At the same time, the complex low-altitude environment leads to positioning errors, and the level of environmental interference varies in different airspaces. Traditional methods are difficult to dynamically adapt to risks, resulting in either overly conservative safety intervals that waste airspace, or insufficient risk assessments that may lead to safety hazards. Summary of the Invention

[0003] This application provides a method and system for calibrating the safe distance of unmanned aerial vehicles (UAVs) based on collision risk prediction, which is used to address the technical problem of poor accuracy caused by the inability of existing UAV safe distance calibration to match risks.

[0004] In view of the above problems, this application provides a method and system for calibrating the safe distance of unmanned aerial vehicles based on collision risk prediction.

[0005] In a first aspect, this application provides a method for calibrating the safe distance of unmanned aerial vehicles (UAVs) based on collision risk prediction, the method comprising: Obtain the initial state parameters and inherent dimensions of the target UAV, and obtain the historical positioning error set of the target UAV and the historical airspace error set of the target airspace; Based on the inherent dimensions, an initial collision risk domain is obtained, and the initial collision risk domain is adjusted based on the initial state parameters and motion intention of the target UAV to obtain a new collision risk domain. Error prediction is performed based on the historical positioning error set and the historical spatial error set to obtain error risk parameters; The collision risk domain is scaled down based on the error risk parameters to obtain the safe interval for the UAV.

[0006] Secondly, this application provides a drone safety distance calibration system based on collision risk prediction, including: The information acquisition module is used to acquire the initial state parameters and inherent dimensions of the target UAV, as well as the historical positioning error set of the target UAV and the historical airspace error set of the target airspace. The collision risk domain acquisition module is used to acquire an initial collision risk domain based on the inherent size, and to adjust the initial collision risk domain based on the initial state parameters and motion intention of the target UAV to acquire the collision risk domain. The error risk acquisition module is used to perform error prediction based on the historical positioning error set and the historical spatial error set, and to acquire error risk parameters. The safety interval acquisition module is used to scale the collision risk domain based on the error risk parameters to obtain the safety interval of the UAV.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application proposes a method and system for calibrating safe distances for unmanned aerial vehicles (UAVs) based on collision risk prediction. By introducing a deep coupling analysis of motion intention and historical errors, it achieves a dynamic and refined assessment of UAV operational risks. First, the method constructs an initial collision risk domain based on the inherent dimensions of the UAV and predicts future states according to real-time motion intentions. This allows the risk domain to be partitioned and adjusted so that its shape and size adaptively reflect the UAV's motion characteristics in different directions. For example, the longitudinal risk domain automatically expands during high-speed forward flight, while the risks in all directions tend to be balanced during hovering or low-speed maneuvers, thus overcoming the shortcomings of traditional fixed geometric models that cannot reflect anisotropic collision risks. Second, by fusing the UAV's own historical positioning errors with environmental error data from the target airspace, quantified error risk parameters are obtained. The frequency of motion intentions in historical data is used to weight and fuse the two types of errors, making the error estimation more consistent with the statistical patterns of the current operational scenario. Simultaneously, the confidence level of the predicted state is assessed based on historical errors, and this is used to correct the partitioned risks, further improving the reliability of risk prediction. Based on this, the error risk parameter is applied as a scaling factor to the adjusted collision risk domain, ultimately outputting a three-dimensional safety interval that matches the safety target level. Compared with traditional methods, the technical solution provided in this application can adapt to the differences in UAV models and payloads, significantly improving the matching degree between safety interval calibration and actual operational risks. It avoids overly conservative intervals or underestimation of risks due to ignoring differences in motion intentions, thereby effectively improving airspace resource utilization while ensuring the same level of safety, achieving the technical effect of enabling UAV safety intervals to adapt to motion intentions and errors. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating the UAV safety interval calibration method based on collision risk prediction provided in an embodiment of this application.

[0010] Figure 2 This is a schematic diagram of the structure of the UAV safety interval calibration system based on collision risk prediction provided in an embodiment of this application.

[0011] The components represented by each number in the attached diagram are explained below: Information acquisition module 100, collision risk domain acquisition module 200, error risk acquisition module 300, and safety interval acquisition module 400. Detailed Implementation

[0012] This application provides a method and system for calibrating the safe distance of unmanned aerial vehicles (UAVs) based on collision risk prediction, which addresses the technical problem of poor accuracy in existing UAV safe distance calibrations due to mismatch between collision risk and risk.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0015] Example 1, as Figure 1 As shown, this application provides a method for calibrating the safe distance of unmanned aerial vehicles (UAVs) based on collision risk prediction, wherein the method includes: S10: Obtain the initial state parameters and inherent dimensions of the target UAV, and obtain the historical positioning error set of the target UAV and the historical airspace error set of the target airspace.

[0016] In the process of calibrating safe intervals for unmanned aerial vehicles (UAVs), existing methods typically rely solely on the UAV's current instantaneous state parameters, such as position and velocity, without fully incorporating historical positioning errors and environmental interference information. Ignoring historical positioning errors makes it difficult to assess the accuracy fluctuations of the positioning system, while ignoring historical environmental errors in the target airspace fails to reflect the differences in external interference under different airspace conditions.

[0017] Step S10 in the method provided in this application embodiment includes: Obtain the initial state parameters and inherent dimensions of the target UAV, wherein the initial state parameters include position coordinates and velocity vector; The historical positioning error of the target UAV is obtained, wherein the historical positioning error includes multiple historical positioning errors; Obtain the historical airspace error set of the target airspace, wherein the historical airspace error set includes the historical airspace errors of multiple UAVs in the target airspace.

[0018] In this embodiment of the application, the initial state parameters and inherent dimensions of the target UAV are obtained, and the historical positioning error set of the target UAV and the historical airspace error set of the target airspace are obtained.

[0019] Specifically, firstly, the initial state parameters and inherent dimensions of the target UAV are obtained. The initial state parameters include position coordinates and velocity vectors. For example, the current position coordinates are obtained through the UAV's onboard positioning system, such as 118.7 degrees East longitude, 32.0 degrees North latitude, and an altitude of 120 meters; and the current velocity vector is obtained through the inertial measurement unit, including a ground speed of 15 meters per second, a heading angle of 90 degrees (due east), and a vertical velocity of 0 meters per second. Further, the inherent dimensions of the target UAV are read from its model parameters, such as a body length of 1.2 meters, a width of 1.2 meters, a height of 0.4 meters, and a rotor sweep diameter of 0.3 meters. If the UAV contains a payload exceeding its inherent dimensions, the maximum value between the payload and the UAV's dimensions is taken as the inherent dimensions. For example, if the UAV carries a long, strip-shaped payload with a length of 1.3 meters, the inherent dimensions are a body length of 1.3 meters, a width of 1.2 meters, a height of 0.4 meters, and a rotor sweep diameter of 0.3 meters.

[0020] Furthermore, the historical positioning errors of the target UAV are obtained, wherein the historical positioning errors include multiple historical positioning errors. For example, the historical positioning errors are extracted from the log file of the UAV flight control system. If the log records the deviation between the satellite positioning solution and the actual position during each positioning in the last 100 flight missions, a residual sequence containing horizontal and vertical errors is formed. For example, one set of data is a horizontal error of 0.3 meters and a vertical error of 0.1 meters.

[0021] Furthermore, a historical airspace error set for the target airspace is obtained, wherein the historical airspace error set includes the historical airspace errors of multiple UAVs in the target airspace. For example, the historical airspace error set is obtained from a shared database of an airspace management platform. This database aggregates historical positioning error data recorded by multiple other UAVs that have flown in the same airspace. Each flight includes horizontal and vertical errors; for example, one flight recorded a horizontal error of 0.4 meters and a vertical error of 0.2 meters, while another recorded a horizontal error of 0.5 meters and a vertical error of 0.3 meters. These data collectively constitute the historical airspace error set for the target airspace, reflecting the statistical characteristics of positioning errors in that airspace environment.

[0022] By acquiring the initial state parameters and inherent dimensions of the target UAV, and simultaneously collecting its historical positioning error set and the historical airspace error set of the target airspace, a comprehensive data foundation covering the current motion state, physical attributes, its own accuracy history, and environmental interference history was constructed.

[0023] S20: Based on the inherent dimensions, obtain the initial collision risk domain, and adjust the initial collision risk domain based on the initial state parameters and motion intention of the target UAV to obtain the collision risk domain.

[0024] During operation, the collision risk of a drone is not uniformly distributed in all directions of space, but is closely related to its flight intention. For example, when flying forward at high speed, the longitudinal risk is much greater than the lateral risk, while when hovering, the risks in all directions are relatively balanced.

[0025] Step S20 in the method provided in this application embodiment includes: Based on the inherent dimensions and combined with the standard safety constraints of the target UAV model, the initial collision risk domain is obtained; Based on the initial state parameters and motion intention, a predicted state is obtained, and the initial collision risk domain is adjusted based on the predicted state to obtain a collision risk domain, wherein the motion intention includes the target position; The process of obtaining the predicted state based on the initial state parameters and the motion intention includes: Construct a state prediction model framework; Obtain historical state parameters and historical motion intentions as the sample input set; The state after the time interval corresponding to the sample input set is obtained as the sample output set, wherein the time interval is obtained based on the initial state parameters of the target UAV; The state prediction model framework is trained using the sample input set and the sample output set until convergence, thereby obtaining the state prediction model. The initial state parameters and the motion intention are input into the state prediction model to obtain the predicted state, wherein the predicted state includes the predicted velocity vector; Based on the distribution characteristics of the predicted state in the sample output set, the confidence level of the predicted state is evaluated, and the prediction confidence level parameter is obtained. The process of adjusting the initial collision risk domain based on the predicted state to obtain the collision risk domain includes: Based on the predicted state, a collision risk analysis is performed to obtain multi-region collision risks, wherein the multi-region collision risks include longitudinal risk, lateral risk and vertical risk. Based on the multi-region collision risk, the initial collision risk domain is partitioned and adjusted to obtain the collision risk domain; The collision risk analysis based on the predicted state is used to obtain multi-region collision risks, including: Based on the predicted state, the spatial range that the target UAV may occupy in the future period is determined. The ratios of the projected lengths of the spatial range in the longitudinal, lateral, and vertical directions to the inherent dimensions are normalized and used as the initial longitudinal risk, initial lateral risk, and initial vertical risk, respectively. The initial longitudinal risk, initial lateral risk, and initial vertical risk are corrected using the prediction confidence parameters to obtain the longitudinal risk, lateral risk, and vertical risk.

[0026] In this embodiment of the application, an initial collision risk domain is obtained based on the inherent size, and the initial collision risk domain is adjusted based on the initial state parameters and motion intention of the target UAV to obtain a new collision risk domain.

[0027] Specifically, firstly, based on the inherent dimensions and the standard safety constraints of the target UAV model, an initial collision risk domain is obtained. For example, the inherent dimensions of the target UAV include a body length of 1.2 meters, a width of 1.2 meters, a height of 0.4 meters, and a rotor sweep diameter of 0.3 meters. Exemplarily, the standard safety constraints of the target UAV model stipulate a safety margin of 1 times the geometric dimensions. Specifically, the safety margin can be obtained by classifying the target UAV based on its function; for example, a UAV transporting important biological materials might have a safety margin level of 3. Further, a safety margin requirement is obtained based on the safety margin level; for example, when the safety margin level is 3, the safety margin requirement is 3 times. In specific scenarios, the safety margin requirement corresponding to the safety margin level can be adjusted according to the actual scenario. Furthermore, the initial collision risk domain is shaped like an ellipsoid, with its longitudinal half-axis length being half the fuselage length plus the rotor sweep radius multiplied by 2, its lateral half-axis length being half the fuselage width plus the rotor sweep radius multiplied by 2, and its vertical half-axis length being half the fuselage height plus the rotor sweep radius multiplied by 2. That is, the longitudinal half-axis length is 1.5 meters, the lateral half-axis length is 1.5 meters, and the vertical half-axis length is 0.7 meters. These three half-axis lengths of the initial collision risk domain are calculated and used to describe the basic safety profile of the UAV.

[0028] Furthermore, based on the initial state parameters and motion intention, a predicted state is obtained, and the initial collision risk domain is adjusted based on the predicted state to obtain a collision risk domain, wherein the motion intention includes the target position.

[0029] Specifically, first, a state prediction model framework is constructed. For example, a three-layer fully connected neural network is used: the input layer contains 5 nodes to receive initial state parameters and motion intentions; the hidden layer contains 10 nodes, using the ReLU (Rectified Linear Unit) activation function; and the output layer contains 1 node to output the predicted state. This network is implemented using the TensorFlow framework.

[0030] Furthermore, historical state parameters and historical motion intentions are obtained as a sample input set. For example, 500 sets of historical state parameters and corresponding historical motion intentions are extracted from the UAV flight log as a sample input set. Each set of historical state parameters includes the position coordinates and velocity vector at that time, and the historical motion intention is the target position at that time.

[0031] Further, the state after the time interval corresponding to the sample input set is obtained as the sample output set, wherein the time interval is obtained based on the initial state parameters of the target UAV. For example, the actual state after the time interval corresponding to each group of sample inputs is obtained as the sample output set. For each group of historical samples, the straight-line distance between the sample's starting position and the corresponding target position is calculated. This distance is divided by the sample's initial velocity to obtain the estimated arrival time. The smaller value between the estimated arrival time and a preset maximum time window is taken as the time interval. The preset maximum time window is 5 seconds. For example, in a historical flight, the initial velocity was 10 meters per second, the distance to the target point was 80 meters, and the estimated arrival time was 8 seconds, exceeding 5 seconds, so 5 seconds is taken as the time interval; in another flight, the initial velocity was 8 meters per second, the distance to the target point was 20 meters, and the estimated arrival time was 2.5 seconds, less than 5 seconds, so 2.5 seconds is taken as the time interval. This ensures that the time interval for each sample is both relevant to the flight mission and does not exceed a reasonable prediction range.

[0032] Further, the state prediction model framework is trained using the sample input set and the sample output set until convergence, thus obtaining the state prediction model. For example, the sample input set and sample output set are input into the network, mean squared error is used as the loss function, and the Adam optimizer (Adaptive Moment Estimation) is trained for 200 epochs with a learning rate of 0.001 until the loss converges.

[0033] Further, the initial state parameters and the motion intention are input into the state prediction model to obtain the predicted state, wherein the predicted state includes a predicted velocity vector. For example, the current initial state parameters and motion intention are input into the model to obtain the predicted velocity vector, such as predicting a velocity of 12 meters per second in the longitude direction, a velocity of 3 meters per second in the latitude direction (due south), and a velocity of 1 meter per second in the altitude direction.

[0034] Further, based on the distribution characteristics of the predicted state in the sample output set, the confidence level of the predicted state is evaluated, and a prediction confidence parameter is obtained. For example, all historical velocity vectors are extracted from the sample output set. The predicted velocity vector is compared with each historical velocity vector, and the Euclidean distance between them is calculated. The Euclidean distance is calculated by summing the squares of the velocity differences in the longitude, latitude, and altitude directions, and then taking the square root. A distance threshold is set, for example, 1 meter per second. The number of historical velocity vectors whose Euclidean distance to the predicted velocity vector is less than this threshold is counted, and this number is divided by the total number of historical velocity vectors in the sample output set to obtain a frequency value. This frequency value is between 0 and 1; a higher frequency indicates that the predicted state occurred more frequently in historical flights, and the higher the prediction confidence. Assuming there are 500 historical velocity vectors in the sample output set, and calculations show that 50 historical velocity vectors have an Euclidean distance to the predicted velocity vector of less than 1 meter per second, the frequency value is 50 divided by 500, which equals 0.1. This frequency value is used as the prediction confidence parameter, i.e., 0.1. This approach has low reliability, reflecting that the current forecast state is relatively rare in historical data, and the forecast results may have significant uncertainty.

[0035] Furthermore, based on the predicted state, a collision risk analysis is performed to obtain multi-region collision risks, wherein the multi-region collision risks include longitudinal risk, lateral risk and vertical risk.

[0036] Specifically, firstly, based on the predicted state, the spatial range that the target UAV may occupy in the future is determined. The ratios of the projected lengths of the spatial range in the longitudinal, lateral, and vertical directions to the inherent dimensions are normalized and used as the initial longitudinal risk, initial lateral risk, and initial vertical risk, respectively. For example, assuming the UAV maintains its current speed for the next 3 seconds, the displacements of the UAV in the longitudinal, lateral, and vertical directions during this time can be calculated. The longitudinal direction is defined as the direction in which the velocity vector is projected onto the horizontal plane, and the lateral direction is perpendicular to the longitudinal direction. The displacement components are added to the lengths in the corresponding directions of the inherent dimensions to obtain the projected lengths of the potentially occupied spatial range in the three directions. For example, the longitudinal projected length = body length 1.2 meters + longitudinal displacement 12 meters / second × 3 seconds = 37.2 meters; the lateral projected length = body width 1.2 meters + lateral displacement 3 meters / second × 3 seconds = 10.2 meters; and the vertical projected length = body height 0.4 meters + vertical displacement 1 meter / second × 3 seconds = 3.4 meters. Divide these projected lengths by their corresponding inherent dimensions to obtain the initial longitudinal risk value of 31, the initial lateral risk value of 8.5, and the initial vertical risk value of 8.5.

[0037] Furthermore, the initial longitudinal risk, initial lateral risk, and initial vertical risk are corrected using the predicted confidence parameter to obtain the longitudinal risk, lateral risk, and vertical risk. For example, risk = initial risk value × (2 - predicted confidence). For instance, if the initial longitudinal risk value is 31 and the predicted confidence is 0.1, then the longitudinal risk value = 31 × (2 - 0.1) = 58.9. Using the same method, the lateral risk value is calculated to be 16.15, and the vertical risk value is also 16.15.

[0038] Furthermore, based on the multi-region collision risk, the initial collision risk domain is partitioned and adjusted to obtain a new collision risk domain. For example, the three semi-axis lengths of the initial collision risk domain are multiplied by their corresponding corrected risk values ​​to obtain the adjusted semi-axis lengths of the collision risk domain. For instance, the adjusted semi-axis lengths are calculated to be 88.35 meters longitudinally, 24.23 meters laterally, and 11.3 meters vertically. This yields a final collision risk domain that matches the current motion intention and prediction confidence.

[0039] By first constructing an initial collision risk domain based on the inherent dimensions of the UAV, then combining initial state parameters and motion intentions to predict future states, and subsequently adjusting the risk domain by partitioning it, the final collision risk domain exhibits anisotropic characteristics in the longitudinal, lateral, and vertical directions. This dynamic adjustment mechanism allows the shape of the risk domain to change in real time with the motion intention; for example, the longitudinal risk domain naturally extends when accelerating forward, and the lateral risk domain expands accordingly when turning, thus more accurately depicting the potential collision space in actual UAV operation.

[0040] S30: Based on the historical positioning error set and the historical spatial error set, perform error prediction and obtain error risk parameters.

[0041] Positioning error is one of the key factors affecting drone collision risk assessment, and the error distribution often varies in actual operation.

[0042] Step S30 in the method provided in this application embodiment includes: The average historical positioning error and the average spatial positioning error are obtained by calculating the mean values ​​of the historical positioning error set and the historical spatial error set, respectively. The average historical positioning error and the average spatial positioning error are weighted and calculated to obtain the error risk parameter; Specifically, the average historical positioning error and the average spatial positioning error are weighted and calculated to obtain error risk parameters, including: Based on the frequency of the stated motion intention in historical motion intentions, a weighted reorganization is obtained; The average historical positioning error and the average spatial positioning error are weighted and calculated using the weighted reassembly method to obtain error risk parameters.

[0043] In this embodiment of the application, error prediction is performed based on the historical positioning error set and the historical spatial error set to obtain error risk parameters.

[0044] Specifically, firstly, the average values ​​of the historical positioning error set and the historical airspace error set are calculated to obtain the average historical positioning error and the average airspace positioning error. For example, the arithmetic mean of the horizontal and vertical components of all historical positioning errors is calculated to obtain the average historical positioning error, such as a horizontal average error of 0.45 meters and a vertical average error of 0.25 meters. The historical airspace error set comes from the shared database of the airspace management platform and contains positioning error data recorded by multiple UAVs in the same airspace. Each flight also includes horizontal and vertical errors; for example, a record shows a horizontal error of 0.4 meters and a vertical error of 0.2 meters. Similarly, the average value of all airspace errors is calculated to obtain the average airspace positioning error, such as a horizontal average error of 0.35 meters and a vertical average error of 0.15 meters.

[0045] Furthermore, the average historical positioning error and the average spatial positioning error are weighted and calculated to obtain error risk parameters.

[0046] Specifically, firstly, a weighted reassembly is obtained based on the frequency of the stated motion intention within historical motion intentions. For example, historical motion intentions are extracted from the UAV's historical flight logs, with each intention represented by target location coordinates. The target location of the current motion intention is 118.8 degrees East longitude, 32.1 degrees North latitude, and 150 meters above sea level. The number of historical motion intentions falling within a 100-meter radius of the current target location is counted and divided by the total number of historical motion intentions (e.g., 500) to obtain a frequency value, for example, 0.9. The weighted reassembly includes two weights, corresponding to the weights of the average historical positioning error and the average airspace positioning error, respectively. A higher frequency indicates more experience with this type of flight mission, and a greater reference value for the UAV's own historical positioning error. The weight allocation rule is set so that the weight of the historical positioning error equals the frequency value, and the weight of the airspace positioning error equals 1 minus the frequency value. Therefore, the weight of the historical positioning error is 0.9, and the weight of the airspace positioning error is 0.1.

[0047] Further, the average historical positioning error and the average spatial positioning error are weighted and calculated using the weighted reassembly method to obtain the error risk parameter. The weighted calculation is performed separately for the horizontal and vertical components. The horizontal component of the error risk parameter is calculated as 0.35 × 0.1 + 0.45 × 0.9 = 0.44 meters. The vertical component of the error risk parameter is calculated as 0.15 × 0.1 + 0.25 × 0.9 = 0.24 meters. This error risk parameter contains values ​​in both the horizontal and vertical directions, representing the level of positioning uncertainty under the current motion intention, considering both historical accuracy and spatial environmental interference. Further, the dimensional error risk parameter is converted into a dimensionless coefficient ranging from 0 to 1. The horizontal and vertical errors of all samples are extracted from the historical positioning error set and the historical spatial error set, identifying the maximum horizontal error value as 1.0 meter and the maximum vertical error value as 0.5 meters. Dividing the horizontal component of the error risk parameter (0.44 meters) by the maximum horizontal error value of 1.0 meter yields a horizontal risk coefficient of 0.44; dividing the vertical component of the error risk parameter (0.24 meters) by the maximum vertical error value of 0.5 meters yields a vertical risk coefficient of 0.48. The higher of the two coefficients, 0.48, is taken as the final error risk parameter. This coefficient ranges from 0 to 1; a higher value indicates a higher positioning risk under the current motion intention, considering both historical accuracy and airspace interference.

[0048] By statistically processing the historical positioning error set and the historical airspace error set respectively, the average historical positioning error and the average airspace positioning error are obtained, and then a quantified error risk parameter is obtained through weighted calculation. This weighting process can adjust the weights of the two types of errors according to the frequency of occurrence of motion intentions in history, so that the error risk parameter can reflect both the accuracy level of the UAV itself and the environmental interference characteristics of a specific airspace, thereby more accurately characterizing the positioning uncertainty in the current scenario.

[0049] S40: Based on the error risk parameters, scale the collision risk domain to obtain the safe interval for the UAV.

[0050] Existing methods fail to integrate error parameters with the anisotropic characteristics of the risk domain, resulting in a discrepancy between the calibrated interval and the actual operational risk.

[0051] Step S40 in the method provided in this application embodiment includes: Calculate the product of the error risk parameter and the preset scaling factor to obtain the scaling factor; The collision risk domain is scaled using the scaling factor to obtain the UAV safety interval, wherein the UAV safety interval includes longitudinal safety interval, lateral safety interval and vertical safety interval.

[0052] In this embodiment of the application, the collision risk domain is scaled down based on the error risk parameter to obtain the safe interval of the UAV.

[0053] Specifically, firstly, the product of the error risk parameter and the preset scaling factor is calculated to obtain the scaling factor. For example, the preset scaling factor is determined based on the target safety level, such as a value of 10. The preset scaling factor is derived from the target safety level and can convert the target safety level into a scaling factor for the collision risk domain. When the error risk parameter is constant, the larger the risk domain, the lower the collision probability. Therefore, it is necessary to find a scaling factor such that the overall collision probability after considering the error does not exceed the target safety level. This factor is usually pre-calibrated using historical operating data. For example, the collision probabilities corresponding to different scaling factors are statistically analyzed in a large number of current airspace scenarios, and the factor that makes the probability equal to the target safety level is selected as the preset value. The scaling factor is obtained by multiplying the error risk parameter by the preset scaling factor, i.e., scaling factor = 0.48 × 10 = 4.8. This scaling factor is used for subsequent overall scaling of the collision risk domain.

[0054] Furthermore, the collision risk domain is scaled down using the scaling factors to obtain the UAV safety distance, which includes a longitudinal safety distance, a lateral safety distance, and a vertical safety distance. Specifically, the collision risk domain is scaled down using scaling factors, that is, the semi-axis lengths in the three directions are multiplied by the scaling factor. The longitudinal safety distance equals the longitudinal semi-axis length multiplied by the scaling factor; the lateral safety distance equals the lateral semi-axis length multiplied by the scaling factor; and the vertical safety distance equals the vertical semi-axis length multiplied by the scaling factor. For example, the calculated longitudinal safety distance is 424 meters, the lateral safety distance is 116.3 meters, and the vertical safety distance is 54.24 meters. Finally, the scaled collision risk domain is output as the final UAV safety distance. This safety distance comprehensively considers the inherent size of the UAV, its current motion intention, historical positioning errors, and airspace environmental errors, and matches a preset safety target level, used to guide the safe distance maintained by the UAV from other aircraft or obstacles during operation.

[0055] The scaling factor is obtained by multiplying the error risk parameter by a preset scaling factor. This scaling factor is then used to scale the previously obtained anisotropic collision risk domain as a whole, ultimately outputting three-dimensional safety intervals in the longitudinal, lateral, and vertical directions. This scaling process incorporates error uncertainty into the risk domain in a quantifiable way, so that the final safety interval simultaneously reflects the directional risk brought about by the motion intention and the statistical impact of historical and environmental errors.

[0056] Example 2, as Figure 2 As shown, based on the same inventive concept as the UAV safety distance calibration method based on collision risk prediction provided in Embodiment 1, this embodiment of the invention also provides a UAV safety distance calibration system based on collision risk prediction, including: The information acquisition module 100 is used to acquire the initial state parameters and inherent dimensions of the target UAV, and to acquire the historical positioning error set of the target UAV and the historical airspace error set of the target airspace. The collision risk domain acquisition module 200 is used to acquire an initial collision risk domain based on the inherent size, and to adjust the initial collision risk domain based on the initial state parameters and motion intention of the target UAV to acquire the collision risk domain. The error risk acquisition module 300 is used to perform error prediction based on the historical positioning error set and the historical spatial error set, and to acquire error risk parameters. The safety interval acquisition module 400 is used to scale the collision risk domain based on the error risk parameters to obtain the UAV safety interval.

[0057] In one embodiment, the information acquisition module 100 is further configured to: Obtain the initial state parameters and inherent dimensions of the target UAV, wherein the initial state parameters include position coordinates and velocity vector; The historical positioning error of the target UAV is obtained, wherein the historical positioning error includes multiple historical positioning errors; Obtain the historical airspace error set of the target airspace, wherein the historical airspace error set includes the historical airspace errors of multiple UAVs in the target airspace.

[0058] In one embodiment, the collision risk domain acquisition module 200 is further configured to: Based on the inherent dimensions and combined with the standard safety constraints of the target UAV model, the initial collision risk domain is obtained; Based on the initial state parameters and motion intention, a predicted state is obtained, and the initial collision risk domain is adjusted based on the predicted state to obtain a collision risk domain, wherein the motion intention includes the target position; The process of obtaining the predicted state based on the initial state parameters and the motion intention includes: Construct a state prediction model framework; Obtain historical state parameters and historical motion intentions as the sample input set; The state after the time interval corresponding to the sample input set is obtained as the sample output set, wherein the time interval is obtained based on the initial state parameters of the target UAV; The state prediction model framework is trained using the sample input set and the sample output set until convergence, thereby obtaining the state prediction model. The initial state parameters and the motion intention are input into the state prediction model to obtain the predicted state, wherein the predicted state includes the predicted velocity vector; Based on the distribution characteristics of the predicted state in the sample output set, the confidence level of the predicted state is evaluated, and the prediction confidence level parameter is obtained. The process of adjusting the initial collision risk domain based on the predicted state to obtain the collision risk domain includes: Based on the predicted state, a collision risk analysis is performed to obtain multi-region collision risks, wherein the multi-region collision risks include longitudinal risk, lateral risk and vertical risk. Based on the multi-region collision risk, the initial collision risk domain is partitioned and adjusted to obtain the collision risk domain; The collision risk analysis based on the predicted state is used to obtain multi-region collision risks, including: Based on the predicted state, the spatial range that the target UAV may occupy in the future period is determined. The ratios of the projected lengths of the spatial range in the longitudinal, lateral, and vertical directions to the inherent dimensions are normalized and used as the initial longitudinal risk, initial lateral risk, and initial vertical risk, respectively. The initial longitudinal risk, initial lateral risk, and initial vertical risk are corrected using the prediction confidence parameters to obtain the longitudinal risk, lateral risk, and vertical risk.

[0059] In one embodiment, the error risk acquisition module 300 is further configured to: The average historical positioning error and the average spatial positioning error are obtained by calculating the mean values ​​of the historical positioning error set and the historical spatial error set, respectively. The average historical positioning error and the average spatial positioning error are weighted and calculated to obtain the error risk parameter; Specifically, the average historical positioning error and the average spatial positioning error are weighted and calculated to obtain error risk parameters, including: Based on the frequency of the stated motion intention in historical motion intentions, a weighted reorganization is obtained; The average historical positioning error and the average spatial positioning error are weighted and calculated using the weighted reassembly method to obtain error risk parameters.

[0060] In one embodiment, the safety interval acquisition module 400 is further configured to: Calculate the product of the error risk parameter and the preset scaling factor to obtain the scaling factor; The collision risk domain is scaled using the scaling factor to obtain the UAV safety interval, wherein the UAV safety interval includes longitudinal safety interval, lateral safety interval and vertical safety interval.

[0061] In summary, the embodiments of this application have at least the following technical effects: This application proposes a method and system for calibrating safe distances for unmanned aerial vehicles (UAVs) based on collision risk prediction. By introducing a deep coupling analysis of motion intention and historical errors, it achieves a dynamic and refined assessment of UAV operational risks. First, the method constructs an initial collision risk domain based on the inherent dimensions of the UAV and predicts future states according to real-time motion intentions. This allows the risk domain to be partitioned and adjusted so that its shape and size adaptively reflect the UAV's motion characteristics in different directions. For example, the longitudinal risk domain automatically expands during high-speed forward flight, while the risks in all directions tend to be balanced during hovering or low-speed maneuvers, thus overcoming the shortcomings of traditional fixed geometric models that cannot reflect anisotropic collision risks. Second, by fusing the UAV's own historical positioning errors with environmental error data from the target airspace, quantified error risk parameters are obtained. The frequency of motion intentions in historical data is used to weight and fuse the two types of errors, making the error estimation more consistent with the statistical patterns of the current operational scenario. Simultaneously, the confidence level of the predicted state is assessed based on historical errors, and this is used to correct the partitioned risks, further improving the reliability of risk prediction. Based on this, the error risk parameter is applied as a scaling factor to the adjusted collision risk domain, ultimately outputting a three-dimensional safety interval that matches the safety target level. Compared with traditional methods, the technical solution provided in this application can adapt to the differences in UAV models and payloads, significantly improving the matching degree between safety interval calibration and actual operational risks. It avoids overly conservative intervals or underestimation of risks due to ignoring differences in motion intentions, thereby effectively improving airspace resource utilization while ensuring the same level of safety, achieving the technical effect of enabling UAV safety intervals to adapt to motion intentions and errors.

[0062] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0063] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0064] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for calibrating safe intervals for unmanned aerial vehicles (UAVs) based on collision risk prediction, characterized in that, include: Obtain the initial state parameters and inherent dimensions of the target UAV, and obtain the historical positioning error set of the target UAV and the historical airspace error set of the target airspace; Based on the inherent dimensions, an initial collision risk domain is obtained, and the initial collision risk domain is adjusted based on the initial state parameters and motion intention of the target UAV to obtain a new collision risk domain. Error prediction is performed based on the historical positioning error set and the historical spatial error set to obtain error risk parameters; The collision risk domain is scaled down based on the error risk parameters to obtain the safe interval for the UAV.

2. The method for calibrating the safe interval of a UAV based on collision risk prediction according to claim 1, characterized in that, Obtain the initial state parameters and inherent dimensions of the target UAV, and obtain the historical positioning error set of the target UAV and the historical airspace error set of the target airspace, including: Obtain the initial state parameters and inherent dimensions of the target UAV, wherein the initial state parameters include position coordinates and velocity vector; The historical positioning error of the target UAV is obtained, wherein the historical positioning error includes multiple historical positioning errors; Obtain the historical airspace error set of the target airspace, wherein the historical airspace error set includes the historical airspace errors of multiple UAVs in the target airspace.

3. The method for calibrating the safe interval of a UAV based on collision risk prediction according to claim 1, characterized in that, Based on the inherent dimensions, an initial collision risk domain is obtained, and the initial collision risk domain is adjusted based on the initial state parameters and motion intention of the target UAV to obtain the collision risk domain, including: Based on the inherent dimensions and combined with the standard safety constraints of the target UAV model, the initial collision risk domain is obtained; Based on the initial state parameters and motion intention, a predicted state is obtained, and the initial collision risk domain is adjusted based on the predicted state to obtain a collision risk domain, wherein the motion intention includes the target position.

4. The method for calibrating the safe interval of a UAV based on collision risk prediction according to claim 3, characterized in that, Based on the initial state parameters and motion intention, the predicted state is obtained, including: Construct a state prediction model framework; Obtain historical state parameters and historical motion intentions as the sample input set; The state after the time interval corresponding to the sample input set is obtained as the sample output set, wherein the time interval is obtained based on the initial state parameters of the target UAV; The state prediction model framework is trained using the sample input set and the sample output set until convergence, thereby obtaining the state prediction model. The initial state parameters and the motion intention are input into the state prediction model to obtain the predicted state, wherein the predicted state includes the predicted velocity vector; Based on the distribution characteristics of the predicted state in the sample output set, the confidence level of the predicted state is evaluated, and the prediction confidence level parameter is obtained.

5. The method for calibrating the safe interval of a UAV based on collision risk prediction according to claim 4, characterized in that, The initial collision risk domain is adjusted based on the predicted state to obtain the collision risk domain, including: Based on the predicted state, a collision risk analysis is performed to obtain multi-region collision risks, wherein the multi-region collision risks include longitudinal risk, lateral risk and vertical risk. Based on the multi-region collision risk, the initial collision risk domain is partitioned and adjusted to obtain the collision risk domain.

6. The method for calibrating the safe interval of a UAV based on collision risk prediction according to claim 5, characterized in that, Based on the predicted state, a collision risk analysis is performed to obtain multi-region collision risks, including: Based on the predicted state, the spatial range that the target UAV may occupy in the future period is determined. The ratios of the projected lengths of the spatial range in the longitudinal, lateral, and vertical directions to the inherent dimensions are normalized and used as the initial longitudinal risk, initial lateral risk, and initial vertical risk, respectively. The initial longitudinal risk, initial lateral risk, and initial vertical risk are corrected using the prediction confidence parameters to obtain the longitudinal risk, lateral risk, and vertical risk.

7. The method for calibrating the safe interval of a UAV based on collision risk prediction according to claim 1, characterized in that, Error prediction is performed based on the historical positioning error set and the historical spatial error set to obtain error risk parameters, including: The average historical positioning error and the average spatial positioning error are obtained by calculating the mean values ​​of the historical positioning error set and the historical spatial error set, respectively. The average historical positioning error and the average spatial positioning error are weighted and calculated to obtain the error risk parameter.

8. The method for calibrating the safe interval of a UAV based on collision risk prediction according to claim 7, characterized in that, The average historical positioning error and the average spatial positioning error are weighted and calculated to obtain error risk parameters, including: Based on the frequency of the stated motion intention in historical motion intentions, a weighted reorganization is obtained; The average historical positioning error and the average spatial positioning error are weighted and calculated using the weighted reassembly method to obtain error risk parameters.

9. The method for calibrating the safe interval of a UAV based on collision risk prediction according to claim 1, characterized in that, Based on the error risk parameters, the collision risk domain is scaled down to obtain the safe distance for the UAV, including: Calculate the product of the error risk parameter and the preset scaling factor to obtain the scaling factor; The collision risk domain is scaled using the scaling factor to obtain the UAV safety interval, wherein the UAV safety interval includes longitudinal safety interval, lateral safety interval and vertical safety interval.

10. A UAV safety interval calibration system based on collision risk prediction, characterized in that, The system is used to implement the UAV safety interval calibration method based on collision risk prediction as described in any one of claims 1-9, the system comprising: The information acquisition module is used to acquire the initial state parameters and inherent dimensions of the target UAV, as well as the historical positioning error set of the target UAV and the historical airspace error set of the target airspace. The collision risk domain acquisition module is used to acquire an initial collision risk domain based on the inherent size, and to adjust the initial collision risk domain based on the initial state parameters and motion intention of the target UAV to acquire the collision risk domain. The error risk acquisition module is used to perform error prediction based on the historical positioning error set and the historical spatial error set, and to acquire error risk parameters. The safety interval acquisition module is used to scale the collision risk domain based on the error risk parameters to obtain the safety interval of the UAV.

Images