Clustered uav confrontation training collaborative optimization method, system, device and storage medium of distributed digital twin architecture

Abstract

The application relates to the technical field of cluster UAV confrontation training, and provides a cluster UAV confrontation training cooperative optimization method, system and device of a distributed digital twin architecture and a storage medium, and solves the problem of cooperative failure. The method comprises the following steps: correcting the control parameters of each UAV individual according to abnormal fluctuation amplitude and spatial gradient change information; constructing a confrontation training scene through a digital twin model of the distributed digital twin architecture; loading the corresponding corrected control parameters, recording the actual response results of the UAV individuals in the confrontation training scene; comparing the deviation between the actual response results and the expected behavior of the digital twin model; when the deviation exceeds a dynamic fault tolerance threshold, reconstructing the correction logic of the control parameters; and based on the corrected control parameters and the actual response results of the multiple UAV individuals exchanged by the distributed nodes, cooperatively generating an anti-interference group flight strategy. The application improves the anti-interference ability and group cooperative flight performance of the cluster UAV.

Description

Technical Field

[0001] This application relates to the field of swarm drone adversarial training technology, and in particular to a collaborative optimization method and system for swarm drone adversarial training using a distributed digital twin architecture. Background Technology

[0002] In the scenario of swarm UAV adversarial training, the technical requirements are mainly reflected in three dimensions: First, it is necessary to integrate UAV power system state data (such as the rate of change of rotation speed fluctuation amplitude and voltage anomaly offset) with hydrodynamic environment parameters (such as air pressure change rate vector and temperature field contour density) in real time to construct an adversarial scenario that can dynamically reflect changes in spatial gradient; Second, the control parameter correction mechanism is required to be fault-tolerant, and when the actual response result deviates from the expected value of the digital twin model by more than the limit, it can automatically reconstruct the correction logic; Third, it is necessary to realize data interaction and collaborative decision-making of distributed nodes, and generate anti-interference swarm flight strategies while ensuring communication efficiency, so as to meet the real-time adversarial training needs of large-scale swarms.

[0003] A typical solution to the above requirements is a hybrid digital twin training architecture based on edge-cloud collaboration. This approach collects UAV power and environmental data at edge nodes, preprocesses it, and uploads it to a cloud-based digital twin model. The cloud then centrally generates adversarial training scenarios, calculates control parameter adjustments, and distributes the data to each UAV for execution. For example, a company's swarm adversarial simulation platform uses this architecture, with the edge handling data filtering and the cloud utilizing a parallel computing engine to generate virtual adversarial rules, enabling simultaneous simulation of virtual and real scenarios.

[0004] However, this solution has three significant drawbacks: First, centralized cloud processing causes data transmission latency to increase exponentially with the cluster size. When the number of drones is large, the control parameter correction lag time is long, which cannot meet the real-time requirements of highly dynamic adversarial scenarios. Second, environmental modeling relies on global parameters in the cloud, making it difficult to adapt to local spatial gradient differences. For example, the virtual scene in the region of sudden change in air pressure rate vector has a large deviation from the physical entity. Third, the fault tolerance mechanism relies on a single decision node in the cloud. Once communication is interrupted or computing fails, the cluster will lose its ability to coordinate and adjust, resulting in low efficiency in generating anti-interference group strategies. Summary of the Invention

[0005] This application provides a distributed digital twin architecture-based collaborative optimization method and system for clustered UAV adversarial training, which solves the problems of latency caused by centralized processing, difficulty in adapting global parameters to local differences, and collaborative failure caused by single node failure in the prior art.

[0006] Firstly, this application provides a collaborative optimization method for swarm drone adversarial training based on a distributed digital twin architecture, including:

[0007] Acquire the power system status data and hydrodynamic environment parameters of multiple individual drones in the cluster;

[0008] Based on the abnormal fluctuation amplitude of the power system state data and the spatial gradient change information of the hydrodynamic environment parameters, the control parameters of each UAV are corrected.

[0009] Using a digital twin model with a distributed digital twin architecture, an adversarial training scenario is constructed based on the fluctuation characteristics of the dynamic system state data and the spatial gradient distribution of the fluid dynamic environment parameters.

[0010] After loading the corresponding modified control parameters into each of the aforementioned drones, the actual response results of the drones in the adversarial training scenario are recorded.

[0011] By comparing the deviation between the actual response result and the expected behavior of the digital twin model, when the deviation exceeds the dynamic fault tolerance threshold, a fault tolerance adjustment mechanism is triggered to reconstruct the correction logic of the control parameters.

[0012] Based on the corrected control parameters of multiple individual UAVs exchanged through distributed nodes and the actual response results, an anti-interference swarm flight strategy is collaboratively generated.

[0013] Optionally, the digital twin model using a distributed digital twin architecture constructs an adversarial training scenario based on the fluctuation characteristics of the dynamic system state data and the spatial gradient distribution of the hydrodynamic environment parameters, including:

[0014] Through the feature extraction module in the digital twin model, corresponding fluctuation features are extracted from the dynamic state parameters of each UAV individual. The fluctuation features include the rate of change of rotational speed fluctuation amplitude and the voltage anomaly offset. At the same time, the spatial gradient distribution is extracted from the hydrodynamic environment parameters of each UAV individual. The spatial gradient distribution includes the air pressure change rate vector and the temperature field contour density.

[0015] The first rule setting module in the digital twin model maps the rate of change of the rotational speed fluctuation amplitude to the motion performance attenuation coefficient of the virtual entity, and generates a first boundary constraint rule to characterize the collision volume of the virtual entity based on the motion performance attenuation coefficient.

[0016] The pressure change rate vector is converted into a virtual aerodynamic disturbance direction through the second rule setting module in the digital twin model, and a second boundary constraint rule is generated based on the aerodynamic disturbance direction to characterize the range of environmental disturbance.

[0017] The communication delay parameters are reset based on the voltage anomaly offset using the first strategy generation module in the digital twin model to obtain an anti-interference communication strategy.

[0018] The second strategy generation module in the digital twin model redistributes the coordinates of the virtual obstacle based on the temperature field contour density to obtain an environment-adaptive obstacle avoidance strategy.

[0019] Through the fusion module in the digital twin model, an adversarial training scenario is constructed based on the first boundary constraint rule, the second boundary constraint rule, the anti-interference communication strategy, and the environment adaptive obstacle avoidance strategy.

[0020] Optionally, the step of mapping the rate of change of the rotational speed fluctuation amplitude to the maneuverability attenuation coefficient of the virtual entity through the first rule setting module in the digital twin model, and generating a first boundary constraint rule to characterize the collision volume of the virtual entity based on the maneuverability attenuation coefficient, includes:

[0021] The mapping unit in the first rule setting module divides the rate of change of fluctuation amplitude according to the threshold corresponding to the preset interval to obtain the interval corresponding to the rate of change of rotational speed fluctuation amplitude.

[0022] Based on the calculation method corresponding to the interval, a value for the degree of propulsion reduction is generated;

[0023] After classifying and mapping the propulsion reduction values, they are converted into the motion performance attenuation coefficient of the virtual entity;

[0024] The rule generation unit in the first rule setting module determines the baseline collision volume parameters of the virtual entity under normal conditions and the adjustment ratio corresponding to the baseline collision volume parameters based on the mobility performance attenuation coefficient. The baseline collision volume parameters include baseline length, baseline width and baseline height.

[0025] By adjusting the corresponding adjustment ratios, the reference length, the reference width, and the reference height are adjusted respectively to obtain the adjusted length, adjusted width, and adjusted height;

[0026] The three-dimensional spatial range defined by the adjusted length, the adjusted width, and the adjusted height is converted into the maximum offset value along the X-axis, Y-axis, and Z-axis directions. The maximum offset value is the boundary value that the virtual entity cannot cross when it moves.

[0027] Based on the boundary values, a first boundary constraint rule is generated to characterize the collision volume of the virtual entity.

[0028] Optionally, the step of converting the pressure change rate vector into a virtual aerodynamic disturbance direction through the second rule setting module in the digital twin model, and generating a second boundary constraint rule to characterize the range of environmental disturbance based on the aerodynamic disturbance direction, includes:

[0029] The conversion unit in the second rule setting module decomposes the directional information of air pressure change in the air pressure change rate vector into components along the X-axis, Y-axis and Z-axis.

[0030] A corresponding interference weight coefficient is set for the components in each coordinate axis direction. The interference weight coefficient is used to represent the degree of influence of the components in the corresponding coordinate axis direction on the virtual aerodynamic interference direction.

[0031] The intensity information in the pressure change rate vector is converted into an intensity coefficient;

[0032] The intensity coefficient is multiplied by the interference weight coefficient in each coordinate axis direction to obtain the weighting coefficient in each coordinate axis direction.

[0033] Based on the weighting coefficients, the components of each coordinate axis direction are weighted and calculated to obtain the weighted direction components;

[0034] The weighted directional components are synthesized to obtain the virtual aerodynamic interference direction;

[0035] The rule generation unit in the second rule setting module calculates the angular deviation between the virtual aerodynamic interference direction and the preset environmental disturbance range reference direction.

[0036] Based on the angular deviation, the preset environmental disturbance range is rotated and adjusted so that the central axis of the adjusted environmental disturbance range is consistent with the direction of the virtual aerodynamic interference, thereby obtaining the boundary parameters of the adjusted environmental disturbance range.

[0037] The boundary parameters of the adjusted environmental disturbance range are integrated to generate a second boundary constraint rule for characterizing the environmental disturbance range.

[0038] Optionally, the step of correcting the control parameters of each individual UAV based on the abnormal fluctuation amplitude of the power system state data and the spatial gradient change information of the hydrodynamic environment parameters includes:

[0039] Extract the magnitude and frequency of abnormal fluctuations from the dynamic system state data;

[0040] The magnitude of the abnormal fluctuation is normalized to obtain the corresponding basic value. The occurrence frequency is converted into a correction coefficient according to a preset ratio. The fluctuation characteristic value is generated by multiplying the basic value and the correction coefficient.

[0041] The variation of the fluid dynamic environment parameters under different spatial gradients is analyzed to obtain spatial gradient variation information;

[0042] The fluctuation characteristic value is correlated with the spatial gradient change information to form a comprehensive information set;

[0043] Based on the comprehensive information set, the control parameter adjustment direction and initial adjustment value for each individual UAV are determined.

[0044] The initial adjustment value is adjusted according to the preset control parameter adjustment constraints to obtain the target adjustment value that meets the constraints.

[0045] Based on the target adjustment value, the control parameters of each individual UAV are modified according to the control parameter adjustment direction to obtain the modified control parameters of each individual UAV.

[0046] Optionally, the step of associating the fluctuation feature value with the spatial gradient change information to form a comprehensive information set includes:

[0047] For each individual drone, the corresponding fluctuation feature value is matched with the value at the same position in the spatial gradient change information according to the spatial gradient position of the individual drone, and the correspondence of each pair of matched data is recorded.

[0048] The frequency of occurrence of each pair of matching data in the corresponding drone individual is counted, and the corresponding relationships with a frequency greater than or equal to a preset frequency threshold are marked as strong associations;

[0049] The strongly correlated matching data is grouped according to the individual identifier of each drone to obtain data groups;

[0050] The extreme values ​​and median values ​​of the fluctuation characteristic values ​​in the data group are statistically analyzed to form a characteristic interval. The upper and lower limits and the mean of the spatial gradient change information are calculated simultaneously to form a numerical interval.

[0051] A comprehensive information set is formed based on the individual identifiers of each UAV, the feature ranges, and the numerical ranges.

[0052] Optionally, the modified control parameters of multiple individual UAVs based on distributed node exchange and the actual response results are used to collaboratively generate an anti-interference swarm flight strategy, including:

[0053] Based on distributed nodes, each individual drone is controlled to send its actual location information, corrected control parameters, and actual response results to other individuals in the cluster at preset time intervals. At the same time, it receives the actual location information, corrected control parameters, and actual response results sent by other individuals to form a cluster dataset containing the actual location information, corrected control parameters, and actual response results of all individual drones.

[0054] Extract the adjustment magnitude and adjustment frequency from the corrected control parameters of each individual UAV from the cluster dataset;

[0055] The degree of deviation between the adjustment range and adjustment frequency and the actual response results is compared, and the maneuver requirements of each individual UAV are determined based on the degree of deviation. Different maneuver requirements correspond to different levels of urgency.

[0056] Extract the actual position information of each individual drone from the cluster dataset, and calculate the distance and relative position change rate between adjacent drones.

[0057] Based on the distance and relative position change rate between adjacent UAV individuals, the cluster formation constraint is determined, wherein the distance between adjacent UAV individuals is inversely proportional to the strength of the corresponding cluster formation constraint;

[0058] Based on the urgency of the maneuvering needs of each individual UAV and the strength of the swarm formation constraints, the adjustment schemes of all individual UAVs are integrated into an anti-interference swarm flight strategy.

[0059] Secondly, this application provides a distributed digital twin architecture-based swarm drone adversarial training collaborative optimization system, comprising:

[0060] The acquisition module is used to acquire the power system status data and hydrodynamic environment parameters of multiple individual UAVs in the cluster;

[0061] The correction module is used to correct the control parameters of each UAV based on the abnormal fluctuation amplitude of the power system state data and the spatial gradient change information of the hydrodynamic environment parameters.

[0062] The building module is used to construct adversarial training scenarios based on the fluctuation characteristics of dynamic system state data and the spatial gradient distribution of hydrodynamic environment parameters through a digital twin model of a distributed digital twin architecture.

[0063] The recording module is used to load the corresponding modified control parameters into each of the individual drones and record the actual response results of the individual drones in the adversarial training scenario.

[0064] The adjustment module is used to compare the deviation between the actual response result and the expected behavior of the digital twin model. When the deviation exceeds the dynamic fault tolerance threshold, the fault tolerance adjustment mechanism is triggered to reconstruct the correction logic of the control parameters.

[0065] The generation module is used to collaboratively generate an anti-interference swarm flight strategy based on the corrected control parameters of multiple individual UAVs exchanged by distributed nodes and the actual response results.

[0066] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are to be invoked and executed by the processing component to implement a distributed digital twin architecture-based collaborative optimization method for swarm UAV adversarial training as described in any of the first aspects.

[0067] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a distributed digital twin architecture-based collaborative optimization method for swarm UAV adversarial training as described in any of the first aspects.

[0068] This application provides a collaborative optimization method for adversarial training of swarmed UAVs using a distributed digital twin architecture, comprising: acquiring the power system state data and hydrodynamic environment parameters of multiple individual UAVs in the swarm; correcting the control parameters of each individual UAV based on the abnormal fluctuation amplitude of the power system state data and the spatial gradient change information of the hydrodynamic environment parameters; constructing an adversarial training scenario based on the fluctuation characteristics of the power system state data and the spatial gradient distribution of the hydrodynamic environment parameters using a digital twin model of the distributed digital twin architecture; loading the corresponding corrected control parameters into each individual UAV and recording the actual response results of the individual UAVs in the adversarial training scenario;

[0069] By comparing the deviation between the actual response results and the expected behavior of the digital twin model, when the deviation exceeds the dynamic fault tolerance threshold, a fault tolerance adjustment mechanism is triggered to reconstruct the correction logic of the control parameters. Based on the corrected control parameters and actual response results of multiple individual UAVs exchanged by distributed nodes, an anti-interference swarm flight strategy is collaboratively generated.

[0070] This application has the following advantages:

[0071] By acquiring the power system state data and hydrodynamic environment parameters of multiple individual UAVs in the cluster, basic data support can be provided for subsequent control parameter correction and adversarial scenario construction. By correcting the control parameters of each individual UAV based on the abnormal fluctuation amplitude of the power system state data and the spatial gradient change information of the hydrodynamic environment parameters, the control parameters can be adapted to the actual operating state and environmental changes. By constructing adversarial training scenarios through digital twin models with a distributed digital twin architecture, a high-fidelity virtual adversarial environment can be generated based on real data. By loading the corrected control parameters and recording the actual response results of individual UAVs in the adversarial training scenarios, experimental evidence can be provided for fault tolerance adjustment and group strategy generation. By comparing the deviation between the actual response results and the expected behavior of the digital twin model and triggering the fault tolerance adjustment mechanism, the correction logic can be dynamically reconstructed when the deviation exceeds the limit, improving the system's fault tolerance. By collaboratively generating anti-interference group flight strategies based on distributed node data exchange, the overall anti-interference collaborative optimization of the cluster can be achieved.

[0072] Furthermore, the digital twin model first extracts fluctuation features such as the rate of change of rotational speed fluctuation amplitude and voltage anomaly offset from the dynamic state parameters, and extracts spatial gradient distributions such as the air pressure change rate vector and temperature field contour density from the hydrodynamic environment parameters. Then, through the first rule setting module, the rate of change of rotational speed fluctuation amplitude is mapped to the maneuverability attenuation coefficient of the virtual entity and generates the first boundary constraint rule. This module classifies the rate of change of fluctuation amplitude to generate a propulsion reduction value, which is then converted into a maneuverability attenuation coefficient. Based on this, the baseline collision volume parameters of the virtual entity are adjusted and boundary values ​​are generated to determine the first rule. By refining feature extraction and classifying mapping rules, the boundary constraints of the virtual entity's collision volume can better match the actual maneuverability of the physical UAV, improving the realism of the virtual entity and the accuracy of scene construction in adversarial training scenarios.

[0073] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0074] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0075] Figure 1 A flowchart illustrating a collaborative optimization method for swarm UAV adversarial training using a distributed digital twin architecture, provided in this application embodiment;

[0076] Figure 2This application provides a schematic diagram of the structure of a distributed digital twin architecture-based swarm UAV adversarial training collaborative optimization system.

[0077] Figure 3 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0078] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0079] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 11, 12, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0080] 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 some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0081] Figure 1 A flowchart illustrating a collaborative optimization method for swarm UAV adversarial training using a distributed digital twin architecture, as provided in this application embodiment, is shown below. Figure 1 As shown, the method includes:

[0082] S11. Obtain the power system status data and hydrodynamic environment parameters of multiple individual UAVs in the cluster.

[0083] Among them, the power system status data is the operating data of the UAV's power unit, including changes in motor speed and power supply voltage, reflecting the working status of the power system; the fluid dynamic environment parameters are the data of the fluid environment around the UAV, including air pressure and temperature distribution at different locations, reflecting the characteristics of the environmental fluid; the cluster is a whole composed of multiple UAVs; the individual UAV is a single UAV in the cluster; through this step, a collection of all UAV power and environmental data is finally generated.

[0084] In this embodiment, each drone in the cluster is equipped with devices that can measure the operation of its own power system (such as motor speed and voltage) and the surrounding environment (such as air pressure and temperature). These devices record data in real time, and each drone sends the data to the cluster's distributed processing node. The node then aggregates the data from all drones. For example, in a cluster of 20 drones, each drone records changes in motor speed, voltage fluctuations, and ambient air pressure and temperature using instruments, and then sends the data to the processing node. The node then compiles a set containing data from all drones.

[0085] S12. Based on the abnormal fluctuation amplitude of the power system state data and the spatial gradient change information of the hydrodynamic environment parameters, correct the control parameters of each UAV.

[0086] Among them, abnormal fluctuation amplitude refers to the magnitude of changes in the power system state data that exceed the normal range, such as the degree of sudden change in motor speed; spatial gradient change information refers to the changes in hydrodynamic environmental parameters at different locations, such as the spatial change trend of air pressure; control parameters are the parameters that control the flight of the UAV, such as speed and steering-related parameters; through this step, the corrected control parameters of each UAV are finally generated.

[0087] In this embodiment, the magnitude of abnormal fluctuations is identified from the power data, and the spatial variation trend of environmental parameters is analyzed. Combining these two aspects, the adjustment direction and initial adjustment value of the control parameters are determined. Then, the initial value is modified according to the actual limitations of the UAV (such as the maximum speed of the motor) to obtain the corrected control parameters. For example, if the motor speed of a certain UAV fluctuates abnormally and the surrounding air pressure changes significantly, it is determined that deceleration is necessary. The initial adjustment value is 5, which is changed to 3 after considering the motor limitations, and finally the speed parameter is slowed down by 3.

[0088] S13. Using a distributed digital twin architecture, an adversarial training scenario is constructed based on the fluctuation characteristics of the dynamic system state data and the spatial gradient distribution of the fluid dynamic environment parameters.

[0089] Among them, the distributed digital twin architecture is a virtual simulation system jointly constructed by multiple processing nodes; the digital twin model is a simulation model of the real drone and environment in the system; the fluctuation characteristics are the changing characteristics of the dynamic system state data, such as the speed of rotational speed fluctuation and voltage deviation; the spatial gradient distribution is the spatial distribution characteristics of fluid environment parameters, such as the direction of air pressure change and the density of temperature distribution; through this step, the adversarial training scenario is finally generated.

[0090] In this embodiment, the digital twin model extracts features such as the speed fluctuation and voltage offset from the dynamic data, and extracts information such as the direction of air pressure change and temperature distribution density from the environmental data. The first rule setting module maps the speed fluctuation to the virtual drone's motion capability attenuation coefficient, adjusts the collision volume, and generates the first constraint rule (first, the fluctuation is divided into levels to generate thrust reduction values, then converted into attenuation coefficients, and the baseline volume parameters are adjusted to obtain boundary values). The second rule setting module converts the air pressure change direction into the virtual aerodynamic interference direction and generates the second constraint rule for the environmental disturbance range. The first strategy generation module resets the communication delay based on the voltage offset to obtain an anti-interference communication strategy. The second strategy generation module redistributes virtual obstacles based on the temperature distribution to obtain an obstacle avoidance strategy. Finally, these rules and strategies are fused to construct an adversarial training scenario. For example, the model extracts that a certain drone has fast speed fluctuations and large voltage offsets, and that the air pressure in the environment changes significantly in a certain direction and the temperature distribution is uneven. Then, corresponding constraint rules and strategies are generated and fused to construct a virtual adversarial scenario.

[0091] S14. After loading the corresponding modified control parameters into each individual UAV, record the actual response results of the individual UAV in the adversarial training scenario.

[0092] Among them, the corrected control parameters are the flight control parameters adjusted by S12; the adversarial training scenario is the virtual training environment constructed by S13; the actual response result is the actual performance of the UAV flying in the scenario according to the corrected parameters, such as trajectory, speed changes, etc.; the actual response result is finally generated and recorded through this step.

[0093] In this embodiment, the corrected control parameters obtained in S12 are input into the UAV control system, allowing the UAV to fly in the scenario constructed in S13. A recording device records the flight trajectory, speed changes, obstacle avoidance, and other actual performance data in real time, and then sends these records to the cluster processing nodes. For example, the UAV loads the corrected speed and steering parameters, flies in a virtual scenario, and the recording device records the flight path, whether obstacles are avoided, etc., before sending the records to the nodes.

[0094] S15. Compare the deviation between the actual response result and the expected behavior of the digital twin model. When the deviation exceeds the dynamic fault tolerance threshold, trigger the fault tolerance adjustment mechanism to reconstruct the correction logic of the control parameters.

[0095] Among them, the actual response result is the actual performance of the drone recorded by S14; the expected behavior of the digital twin model is the expected performance of the drone predicted by the model; the deviation is the difference between the actual and the expected; the dynamic fault tolerance threshold is the acceptable range of difference set according to the actual situation; the fault tolerance adjustment mechanism is the way to adjust the correction logic when the deviation exceeds the limit; the correction logic of the control parameters is the method to adjust the control parameters; through this step, the reconstructed correction logic will be generated when the deviation exceeds the limit.

[0096] In this embodiment, the actual response recorded in S14 is compared with the expected behavior of the digital twin model. The deviation between the two is calculated, and it is determined whether the deviation exceeds the dynamic fault tolerance threshold. If it does, the fault tolerance adjustment mechanism is activated, and the correction logic of the control parameters is readjusted. For example, if the actual flight trajectory of a certain UAV differs greatly from the model's expectation and exceeds the threshold, the system readjusts the method of adjusting parameters based on rotational speed and air pressure.

[0097] S16. Based on the corrected control parameters and actual response results of multiple individual UAVs through distributed node exchange, an anti-interference swarm flight strategy is collaboratively generated.

[0098] Among them, distributed nodes are multiple units in the cluster that process and exchange data; the anti-interference swarm flight strategy is the coordinated flight method of the cluster to deal with interference; the anti-interference swarm flight strategy is finally generated through this step.

[0099] In this embodiment, distributed nodes control drones to exchange positions, corrected parameters, and actual response results at fixed intervals, summarizing them to form a cluster dataset. The adjustment range and frequency of control parameters are extracted from the data to determine the flight needs and urgency of each drone. Simultaneously, the distance and position change rate of adjacent drones are calculated to determine formation constraints. Finally, combining the urgency of flight needs and formation constraints, adjustment schemes are integrated to generate an anti-interference swarm flight strategy. For example, after analyzing the dataset formed by the exchange of information among the swarm drones, it is found that some require emergency speed adjustments and others require turning. Combined with formation constraints, a swarm cooperative obstacle avoidance flight mode is generated.

[0100] For example, when a cluster of 25 drones conducts adversarial training, the process begins by collecting data on motor speed changes, voltage fluctuations, and ambient air pressure and temperature for each drone in S11, and then aggregating this data to a distributed processing node. In S12, the data is analyzed, and control parameters are adjusted for drones with large speed fluctuations and significant changes in ambient air pressure, such as slowing down a drone by 2 units. In S13, features are extracted using a digital twin model to generate constraint rules and strategies, constructing a virtual adversarial scenario. In S14, the drones are loaded with corrected parameters and fly in the scenario, recording their trajectories, obstacle avoidance, etc., and sending the data to the node. In S15, the actual flight is compared with the model's expectations, and the parameter correction logic is readjusted for drones with deviations exceeding a threshold. In S16, the drones exchange information to form a dataset, analyze flight requirements and formation constraints, and finally generate a cluster anti-interference cooperative flight strategy.

[0101] By executing S11~S16, this embodiment of the application collects basic data to support subsequent steps, corrects control parameters to enable the UAV to adapt to the actual state and environment, enhances the realism of training by constructing adversarial scenarios based on real data, provides a basis for fault tolerance adjustment and strategy generation by actual response records, enhances system stability by fault tolerance mechanism, and improves the overall anti-interference capability and coordination of the cluster by distributed collaborative generation of group strategies, ultimately achieving efficient collaborative optimization of cluster UAV adversarial training.

[0102] In one possible embodiment, S13, using a digital twin model of a distributed digital twin architecture, based on the fluctuation characteristics of the dynamic system state data and the spatial gradient distribution of the hydrodynamic environment parameters, an adversarial training scenario is constructed, including:

[0103] Step 131: Using the feature extraction module in the digital twin model, extract the corresponding fluctuation features from the dynamic state parameters of each UAV individual. The fluctuation features include the rate of change of rotational speed fluctuation amplitude and the voltage anomaly offset. Simultaneously, extract the spatial gradient distribution from the hydrodynamic environment parameters of each UAV individual. The spatial gradient distribution includes the air pressure change rate vector and the temperature field contour density.

[0104] The digital twin model is a virtual model simulating a real drone and its environment. The feature extraction module is the part that extracts data characteristics from the model. The dynamic state parameters are the operating data of the drone's power system, and the fluctuation characteristics are the changes extracted from them, including the rate of change of the rotational speed fluctuation amplitude (how fast the rotational speed changes) and the voltage anomaly offset (the value of the voltage deviating from the normal range). The hydrodynamic environment parameters are the fluid environment data around the drone, and the spatial gradient distribution is the spatial variation extracted from them, including the air pressure change rate vector (how fast and how far the air pressure changes in different directions) and the temperature field contour density (the density of lines at the same temperature). Through this step, the fluctuation characteristics and spatial gradient distribution are finally generated.

[0105] In this embodiment, the feature extraction module first analyzes the dynamic state parameters of each UAV, calculates the rate of change of rotational speed fluctuation amplitude (e.g., if the rotational speed changes from 1000 rpm to 1200 rpm within 10 seconds, the rate of change is 200 ÷ 1000 ÷ 10 = 0.02, or 2% per second) and the voltage anomaly offset (e.g., normal voltage is 12 volts, measured voltage is 12.3 volts, offset is 0.3 volts), and obtains the fluctuation characteristics. Simultaneously, it analyzes the hydrodynamic environment parameters to determine the changes in air pressure in different directions and the density of temperature isopleths, obtaining the spatial gradient distribution. For example, a UAV's rotational speed fluctuation amplitude changes by 3% per second, its voltage offset is 0.4 volts, the surrounding air pressure changes by 0.2 kPa per meter eastward, and there are 3 temperature isopleths per meter; these are extracted as fluctuation characteristics and spatial gradient distribution.

[0106] Step 132: Through the first rule setting module in the digital twin model, the rate of change of rotation speed fluctuation amplitude is mapped to the motion performance attenuation coefficient of the virtual entity, and the first boundary constraint rule for characterizing the collision volume of the virtual entity is generated based on the motion performance attenuation coefficient.

[0107] The first rule setting module is the part of the digital twin model that sets virtual rules; the rotation speed fluctuation amplitude change rate is the speed at which the rotation speed changes; the virtual entity is the object that simulates a real drone in the virtual scene; the maneuverability attenuation coefficient is the degree to which the virtual entity's mobility decreases; the first boundary constraint rule is the rule that defines the collision range of the virtual entity; the collision volume is the spatial range in which the virtual entity may collide; the first boundary constraint rule is finally generated through this step.

[0108] In this embodiment, firstly, the first rule setting module categorizes the rate of change of rotational speed fluctuation amplitude (e.g., 0-5% is level one, 5%-10% is level two), generating corresponding thrust reduction values ​​(10% for level one, 20% for level two), which are then converted into maneuverability attenuation coefficients (0.1 for level one, 0.2 for level two). Next, the baseline collision volume (length, width, and height) of the virtual entity is adjusted according to the attenuation coefficient. The larger the attenuation coefficient, the larger the volume adjustment ratio (e.g., 0.2 corresponds to 1.2 times). Finally, the first boundary constraint rule is determined based on the adjusted volume. For example, a UAV has a rotational speed fluctuation amplitude change rate of 8% (level two), an attenuation coefficient of 0.2, and a baseline volume of 2 meters long, 1 meter wide, and 0.5 meters high. After adjustment, it becomes 2.4 meters long, 1.2 meters wide, and 0.6 meters high, and the first rule is generated accordingly.

[0109] Step 133: Through the second rule setting module in the digital twin model, the pressure change rate vector is converted into a virtual aerodynamic disturbance direction, and a second boundary constraint rule is generated based on the aerodynamic disturbance direction to characterize the range of environmental disturbance.

[0110] The second rule setting module is the part of the digital twin model that sets environmental rules; the air pressure change rate vector is the rate and direction of air pressure change in different directions, and the virtual aerodynamic interference direction is the direction of air thrust on the drone in the virtual scene; the second boundary constraint rule is the rule that defines the range of environmental disturbance, and the range of environmental disturbance effect is the area where the environment affects the drone; the second boundary constraint rule is finally generated through this step.

[0111] In this embodiment, firstly, the second rule setting module decomposes the pressure change rate vector into components in different directions and assigns weights to each direction (e.g., 0.7 for the horizontal direction and 0.3 for the vertical direction). The components are multiplied by their weights to synthesize a virtual aerodynamic disturbance direction. Then, based on this direction, the preset environmental disturbance range is adjusted so that the central axis of the range aligns with the disturbance direction, boundary parameters are determined, and a second boundary constraint rule is generated. For example, if the pressure change rate vector has a horizontal direction of 0.3 kPa / m and a vertical direction of 0.1 kPa / m, with weights of 0.7 and 0.3 respectively, the weighted average is 0.21 for the horizontal direction and 0.03 for the vertical direction, resulting in a more horizontal synthesized direction. The environmental disturbance range is adjusted accordingly, generating the second rule.

[0112] Step 134: Using the first strategy generation module in the digital twin model, reset the communication delay parameters based on the voltage anomaly offset to obtain an anti-interference communication strategy.

[0113] The first strategy generation module is the part of the digital twin model that generates the communication strategy; the voltage anomaly offset is the value of the voltage deviating from the normal range; the communication delay parameter is the allowable delay time for signal transmission; the anti-interference communication strategy is a method to reduce communication interference, and the anti-interference communication strategy is finally generated through this step.

[0114] In this embodiment, the first strategy generation module first determines the normal voltage range (e.g., 12 ± 0.2 volts) and calculates the abnormal voltage offset (e.g., a measured voltage of 12.4 volts with an offset of 0.2 volts). Then, it adjusts the communication delay parameter according to the offset; the larger the offset, the larger the delay parameter (e.g., when the offset is 0.2 volts, it is adjusted from 50 milliseconds to 70 milliseconds), thus forming an anti-interference communication strategy. For example, if a UAV's voltage offset is 0.3 volts, the module adjusts its communication delay parameter from 40 milliseconds to 60 milliseconds to reduce interference.

[0115] Step 135: Using the second strategy generation module in the digital twin model, the coordinates of the virtual obstacle are redistributed based on the isopleth density of the temperature field to obtain an environment-adaptive obstacle avoidance strategy.

[0116] The second strategy generation module is the part of the digital twin model that generates obstacle avoidance strategies; the temperature field contour density is the density of lines with the same temperature (reflecting the rate of temperature change); the virtual obstacle coordinates are the positions of objects that obstruct flight in the virtual scene; the environment-adaptive obstacle avoidance strategy is an obstacle avoidance method that is adjusted according to the environment, and the environment-adaptive obstacle avoidance strategy is finally generated through this step.

[0117] In this embodiment, firstly, the second strategy generation module analyzes the density of temperature field contour lines; a higher density indicates a more complex environment. Then, it increases the number of virtual obstacles in high-density areas (e.g., in areas with 3 lines per meter, the number of obstacles increases from 5 to 8) and adjusts their distribution. Finally, based on the new distribution, it formulates obstacle avoidance methods (e.g., reducing speed, increasing detection frequency) to form an environment-adaptive obstacle avoidance strategy. For example, if there are 4 temperature contour lines per meter in a certain area, the module adds 5 obstacles to that area, and the strategy requires the UAV to decelerate by 20% in this area.

[0118] Step 136: Construct an adversarial training scenario using the fusion module in the digital twin model, based on the first boundary constraint rule, the second boundary constraint rule, the anti-interference communication strategy, and the environment adaptive obstacle avoidance strategy.

[0119] Among them, the fusion module is the part that integrates information in the digital twin model; the first and second boundary constraint rules respectively define the collision range of virtual entities and the range of environmental disturbances; the anti-interference communication strategy and the environmental adaptive obstacle avoidance strategy are the corresponding communication and obstacle avoidance methods; the adversarial training scenario is a virtual environment used for training, and the adversarial training scenario is finally generated through this step.

[0120] In this embodiment, firstly, the fusion module collects the first and second boundary constraint rules, anti-interference communication strategies, and environmental adaptive obstacle avoidance strategies generated in steps 132-135; then, it integrates these rules and strategies into the same virtual environment to ensure consistency, forming an adversarial training scenario that includes collision rules, environmental disturbances, communication, and obstacle avoidance methods. For example, the module integrates virtual entity collision range, northeast-direction environmental disturbances, 60-millisecond communication latency, and obstacle avoidance requirements for complex temperature areas to construct a complete scenario.

[0121] For example, in a cluster of UAV adversarial training, step 131 first extracts from UAV A the rotational speed fluctuation amplitude change rate of 5% per second, the voltage anomaly offset of 0.2 volts, the air pressure change rate vector southward of 0.2 kPa per meter, and the temperature field isopleth density of 3 lines per meter; step 132 sets the maneuverability attenuation coefficient of the virtual entity of UAV A to 0.1 based on this, and generates the first boundary constraint rule after adjusting the collision volume; step 133 determines the direction of virtual aerodynamic interference southward based on the southward air pressure change, and generates the second boundary constraint rule; step 134 adjusts the communication delay parameter of UAV A from 50 milliseconds to 60 milliseconds to obtain the anti-interference communication strategy; step 135 adds 3 obstacles in the dense area of ​​temperature isopleths to generate an environment-adaptive obstacle avoidance strategy; finally, step 136 integrates the above rules and strategies to construct a complete adversarial training scenario.

[0122] By executing steps 131 to 136, this embodiment of the application extracts key features of UAV power and environment, generates rules and strategies for virtual scenarios in a targeted manner, and integrates them into a unified adversarial training scenario. This enables the scenario to accurately reflect the real power state and environmental characteristics, improves the realism and reliability of the scenario, provides a realistic virtual platform for swarm UAV adversarial training, and helps to improve the effectiveness of training.

[0123] In one possible embodiment, step 132 involves mapping the rate of change of rotational speed fluctuation amplitude to the maneuverability attenuation coefficient of the virtual entity through the first rule setting module in the digital twin model, and generating a first boundary constraint rule to characterize the collision volume of the virtual entity based on the maneuverability attenuation coefficient, including:

[0124] b1. Using the mapping unit in the first rule setting module, the fluctuation amplitude change rate is divided into levels according to the threshold corresponding to the preset interval to obtain the interval corresponding to the speed fluctuation amplitude change rate.

[0125] The mapping unit in the first rule setting module is used to classify the data. The fluctuation amplitude change rate refers to the speed of the rotational speed change. The preset interval is a number of pre-defined numerical ranges. The threshold is the boundary value of each interval. The interval is the numerical range to which the fluctuation amplitude change rate belongs. Through this step, the interval corresponding to the fluctuation amplitude change rate will be determined.

[0126] In this embodiment, the mapping unit first refers to several pre-set intervals and the boundary values ​​of each interval. For example, it sets three intervals: 0-5%, 5%-10%, and above 10%, with boundary values ​​of 5% and 10% respectively. Then, it compares the fluctuation amplitude change rate with these boundary values ​​to determine which interval it belongs to. For example, if the fluctuation amplitude change rate of a certain drone is 7%, it is determined to belong to the 5%-10% interval after comparing it with the boundary values.

[0127] b2. Based on the calculation method corresponding to the interval, generate the value of the degree of reduction in propulsion.

[0128] The calculation method corresponding to each interval means that each interval has a corresponding calculation method. The thrust reduction value refers to the degree of reduction in the thrust of the UAV. Through this step, the thrust reduction value will be obtained.

[0129] In this embodiment of the application, firstly, a corresponding calculation method is set for each interval. For example, the fluctuation amplitude change rate is multiplied by 0.1 for the 0-5% interval and by 0.2 for the 5%-10% interval. Then, the calculation method corresponding to the interval determined in step b1 is used to substitute the fluctuation amplitude change rate for calculation to obtain the propulsion reduction value. For example, if step b1 determines the interval to be 5%-10% and the fluctuation amplitude change rate is 7%, then according to the calculation method for this interval, 7% is multiplied by 0.2 to obtain the propulsion reduction value of 1.4%.

[0130] b3. After mapping the reduction in propulsion force in different grades, it is converted into the reduction coefficient of the maneuverability of the virtual entity.

[0131] Among them, the graded mapping refers to mapping the propulsion reduction value to different values ​​according to different ranges. The virtual entity's maneuverability reduction coefficient is a value that represents the degree of decline in the virtual entity's mobility. Through this step, the maneuverability reduction coefficient will be obtained.

[0132] In this embodiment of the application, firstly, several ranges of the thrust reduction value and the corresponding attenuation coefficient for each range are set in advance, such as 0.1 for 0-1% and 0.2 for 1%-2%. Then, the thrust reduction value obtained in step b2 is checked to see which range it belongs to, and the corresponding attenuation coefficient of maneuverability is obtained. For example, if the thrust reduction value obtained in step b2 is 1.4%, this value belongs to the 1%-2% range, and the corresponding attenuation coefficient is 0.2.

[0133] b4. Using the rule generation unit in the first rule setting module, based on the motor performance attenuation coefficient, determine the baseline collision volume parameters of the virtual entity under normal conditions and the corresponding adjustment ratio of the baseline collision volume parameters. The baseline collision volume parameters include the baseline length, baseline width, and baseline height.

[0134] Among them, the rule generation unit in the first rule setting module is the part used to formulate rules. The reference collision volume parameter of the virtual entity in normal state refers to the basic size of the space where the virtual entity may collide when it moves normally, including the reference length, reference width and reference height. The adjustment ratio corresponding to the reference collision volume parameter refers to the magnification ratio determined according to the attenuation coefficient. Through this step, the reference collision volume parameter and the corresponding adjustment ratio will be determined.

[0135] In this embodiment of the application, firstly, the rule generation unit determines the baseline collision volume parameters of the virtual entity under normal conditions, such as a baseline length of 2 meters, a baseline width of 1 meter, and a baseline height of 0.5 meters. Then, based on the mobility performance attenuation coefficient obtained in step b3, the adjustment ratio is determined. For example, an attenuation coefficient of 0.1 corresponds to 1.1 times, and 0.2 corresponds to 1.2 times. For example, if the attenuation coefficient obtained in step b3 is 0.2, the rule generation unit will determine the adjustment ratio to be 1.2 times, and at the same time determine the baseline length of 2 meters, the baseline width of 1 meter, and the baseline height of 0.5 meters.

[0136] b5. Adjust the reference length, reference width and reference height respectively by the corresponding adjustment ratio to obtain the adjusted length, adjusted width and adjusted height.

[0137] Among them, the adjusted length, adjusted width, and adjusted height refer to the dimensions of the virtual entity collision volume after proportional adjustment. This step will ultimately yield the adjusted length, width, and height.

[0138] In this embodiment of the application, firstly, the adjustment ratio obtained in step b4 is multiplied by the reference length, reference width, and reference height respectively to calculate the adjusted dimensions. For example, if step b4 determines that the adjustment ratio is 1.2 times, the reference length is 2 meters, the reference width is 1 meter, and the reference height is 0.5 meters, then the adjusted length is 2 meters multiplied by 1.2 equals 2.4 meters, the adjusted width is 1 meter multiplied by 1.2 equals 1.2 meters, and the adjusted height is 0.5 meters multiplied by 1.2 equals 0.6 meters.

[0139] b6. The three-dimensional spatial range defined by the adjusted length, adjusted width and adjusted height is converted into the maximum offset value along the X-axis, Y-axis and Z-axis. The maximum offset value is the boundary value that the virtual entity cannot cross when it moves.

[0140] Among them, the three-dimensional spatial range defined by the adjusted length, adjusted width and adjusted height refers to the size of the space in which the virtual entity may collide. The maximum offset value along the X-axis, Y-axis and Z-axis refers to the distance that the virtual entity cannot exceed when moving in these three directions. The boundary values ​​that the virtual entity cannot cross when moving are these maximum offset values. Through this step, the maximum offset values ​​in the three directions will be obtained.

[0141] In this embodiment of the application, firstly, the adjusted length, adjusted width, and adjusted height obtained in step b5 are used as the total range in the X-axis, Y-axis, and Z-axis directions, respectively. Then, the total range in each direction is divided by 2 to obtain the maximum offset value in both positive and negative directions in each direction. For example, if the adjusted length is 2.4 meters, the maximum offset value in the X-axis direction is 2.4 meters divided by 2, which equals ±1.2 meters; if the adjusted width is 1.2 meters, the maximum offset value in the Y-axis direction is 1.2 meters divided by 2, which equals ±0.6 meters; and if the adjusted height is 0.6 meters, the maximum offset value in the Z-axis direction is 0.6 meters divided by 2, which equals ±0.3 meters.

[0142] b7. Based on the boundary values, generate the first boundary constraint rules to characterize the collision volume of virtual entities.

[0143] Here, the boundary value refers to the maximum offset value in the three directions obtained in step b6. The first boundary constraint rule used to characterize the collision volume of the virtual entity is a rule that stipulates that the virtual entity cannot exceed these offset values ​​when it moves. This step will eventually generate the first boundary constraint rule.

[0144] In this embodiment of the application, firstly, the maximum offset values ​​in the X-axis, Y-axis and Z-axis directions obtained in step b6 are collected, and then these offset values ​​are integrated into a rule to clarify that the virtual entity cannot exceed these offset values ​​when moving. For example, the offset values ​​of ±1.2 meters in the X-axis, ±0.6 meters in the Y-axis and ±0.3 meters in the Z-axis are integrated to form the first boundary constraint rule that "when the virtual entity moves, the offset in the X-axis direction cannot exceed ±1.2 meters, the offset in the Y-axis direction cannot exceed ±0.6 meters, and the offset in the Z-axis direction cannot exceed ±0.3 meters".

[0145] For example, in a cluster of UAV combat training, when processing the rate of change of rotational speed fluctuation amplitude of UAV B, in step b1, the mapping unit compares the 8% fluctuation amplitude change rate with the preset intervals 0-5%, 5%-10%, and above 10%, and determines that it belongs to the 5%-10% interval; in step b2, the calculation method corresponding to this interval is multiplied by 0.2, so the thrust reduction value is 8%×0.2=1.6%; in step b3, 1.6% belongs to the 1%-2% range, corresponding to a maneuverability attenuation coefficient of 0.2; in step b4, the rule generation unit determines the baseline collision volume parameters as a baseline length of 3 meters and a baseline width of 1 meter. With a base height of 0.8 meters and an adjustment ratio of 1.2 times based on an attenuation coefficient of 0.2, step b5 calculates the adjusted length as 3 × 1.2 = 3.6 meters, width as 1.5 × 1.2 = 1.8 meters, and height as 0.8 × 1.2 = 0.96 meters using the adjustment ratio. In step b6, the adjusted volume is converted into maximum offset values ​​of ±1.8 meters for the X-axis, ±0.9 meters for the Y-axis, and ±0.48 meters for the Z-axis. In step b7, based on these offset values, a first boundary constraint rule is generated: "When the virtual entity moves, the X-axis offset shall not exceed ±1.8 meters, the Y-axis offset shall not exceed ±0.9 meters, and the Z-axis offset shall not exceed ±0.48 meters."

[0146] By executing b1~b7, the embodiments of this application classify the rate of change of rotation speed fluctuation amplitude, calculate the degree of propulsion reduction, determine the coefficient of motion capability reduction, and then adjust the collision volume parameters of virtual entities and generate constraint rules accordingly. This enables the collision range of virtual entities to accurately reflect the changes in their motion capability, making the collision rules in the virtual scene more consistent with the actual situation, improving the rationality of the virtual scene, and providing a more reliable virtual environment basis for UAV combat training.

[0147] In one possible embodiment, step 133, through the second rule setting module in the digital twin model, converts the pressure change rate vector into a virtual aerodynamic disturbance direction, and generates a second boundary constraint rule to characterize the range of environmental disturbance based on the aerodynamic disturbance direction, including:

[0148] c1. Through the conversion unit in the second rule setting module, the directional information of air pressure change in the air pressure change rate vector is decomposed into components along the X-axis, Y-axis and Z-axis.

[0149] The conversion unit in the second rule setting module is the part that processes the information on the direction of air pressure change. The air pressure change rate vector contains information on the direction and speed of air pressure change. The components along the X-axis, Y-axis and Z-axis are the specific values ​​of the air pressure change direction in three directions: horizontal front and back, horizontal left and right and vertical up and down. Through this step, the components of these three directions will be obtained.

[0150] In this embodiment, the conversion unit first receives a pressure change rate vector, which reflects the direction and speed of pressure change. Then, the direction information of this vector is split into specific values ​​in three directions: horizontal front-back (X-axis), horizontal left-right (Y-axis), and vertical up-down (Z-axis). For example, if a pressure change rate vector shows that the pressure mainly changes in the horizontal front-back direction, the conversion unit will split it into an X-axis component of 0.3, a Y-axis component of 0.1, and a Z-axis component of 0.05.

[0151] c2. Set the corresponding interference weight coefficient for the components in each coordinate axis direction. The interference weight coefficient is used to represent the degree of influence of the components in the corresponding coordinate axis direction on the virtual aerodynamic interference direction.

[0152] The components of each coordinate axis are the air pressure changes in the X, Y, and Z axes. The interference weight coefficient is a value that represents the magnitude of the influence of air pressure changes in each direction on the direction of virtual aerodynamic interference. This step will ultimately determine the interference weight coefficient corresponding to each direction.

[0153] In this embodiment of the application, firstly, based on the actual situation of air pressure affecting the UAV in each direction during flight, interference weight coefficients are set for the X-axis, Y-axis, and Z-axis. For example, if the horizontal direction has a greater impact on the flight of the UAV, the weight coefficients of the X-axis and Y-axis are set to 0.4, and the Z-axis is set to 0.2. For example, interference weight coefficients of 0.4 for the X-axis, 0.4 for the Y-axis, and 0.2 for the Z-axis are set.

[0154] c3. Convert the intensity information in the pressure change rate vector into intensity coefficients.

[0155] Among them, the intensity information in the pressure change rate vector refers to the speed at which the pressure changes, and the intensity coefficient is the specific numerical value converted from this speed. The intensity coefficient is obtained through this step.

[0156] In this embodiment of the application, firstly, the correspondence between the range of air pressure change rate and the intensity coefficient is set. For example, the intensity coefficient is 0.5 when the air pressure change rate is 0-0.5 kPa / m and the intensity coefficient is 1 when it is 0.5-1 kPa / m. Then, the corresponding intensity coefficient is found according to the intensity information in the air pressure change rate vector. For example, if the air pressure change rate is 0.6 kPa / m, the corresponding intensity coefficient is 1.

[0157] c4. Multiply the intensity coefficient by the interference weight coefficient in each coordinate axis direction to obtain the weighting coefficient in each coordinate axis direction.

[0158] Among them, the intensity coefficient is a value representing the rate of change of air pressure, the interference weight coefficient of each coordinate axis direction is a value representing the magnitude of the influence of each direction on the virtual aerodynamic force, and the weighting coefficient of each coordinate axis direction is the result of multiplying the intensity coefficient by the corresponding weight coefficient. Through this step, the weighting coefficients of the three directions will be obtained in the end.

[0159] In this embodiment of the application, firstly, the intensity coefficient obtained in step c3 is taken out, and then multiplied by the interference weight coefficients of the X-axis, Y-axis and Z-axis set in step c2 respectively to obtain the weighting coefficient of each direction. For example, if the intensity coefficient is 1, the weight of the X-axis is 0.4, the weight of the Y-axis is 0.4 and the weight of the Z-axis is 0.2, then the weighting coefficient of the X-axis is 1×0.4=0.4, the weight of the Y-axis is 1×0.4=0.4 and the weight of the Z-axis is 1×0.2=0.2.

[0160] c5. Based on the weighting coefficients, the components of each coordinate axis direction are weighted and calculated to obtain the weighted direction components.

[0161] The weighting coefficient represents the degree of influence in each direction, the component of each coordinate axis represents the pressure change in the three directions, and the weighted directional component is the result of multiplying the component by the corresponding weighting coefficient. Through this step, the weighted components in the three directions will be obtained.

[0162] In this embodiment of the application, firstly, the X-axis, Y-axis, and Z-axis components obtained in step c1 are extracted, and then multiplied by the corresponding directional weighting coefficients obtained in step c4 to obtain the weighted directional components. For example, if the X-axis component is 0.3 and the weighting coefficient is 0.4, then the weighted X-axis component is 0.3 × 0.4 = 0.12; if the Y-axis component is 0.1 and the weighting coefficient is 0.4, then it is 0.1 × 0.4 = 0.04; and if the Z-axis component is 0.05 and the weighting coefficient is 0.2, then it is 0.05 × 0.2 = 0.01.

[0163] c6. Combine the weighted directional components to obtain the virtual aerodynamic interference direction.

[0164] Among them, the weighted directional component is the influence value of air pressure change in three directions, and the virtual aerodynamic interference direction is the virtual aerodynamic force direction obtained by merging these three components. Through this step, the virtual aerodynamic interference direction will be finally obtained.

[0165] In this embodiment of the application, firstly, the weighted components of the X-axis, Y-axis and Z-axis obtained in step c5 are collected, and then these three components are merged to determine a general direction. This direction is the direction of interference of virtual aerodynamic force on the UAV. For example, the weighted components of the X-axis are 0.12, the Y-axis is 0.04 and the Z-axis is 0.01. The direction obtained after merging is mainly along the X-axis direction.

[0166] c7. Calculate the angular deviation between the virtual aerodynamic interference direction and the preset environmental disturbance range reference direction through the rule generation unit in the second rule setting module.

[0167] Among them, the rule generation unit in the second rule setting module is the part that formulates environmental interference rules. The virtual aerodynamic interference direction is the direction of action of the virtual aerodynamic force, the preset environmental disturbance range reference direction is the pre-set environmental interference reference direction, and the angle deviation is the angle between these two directions. The angle deviation will be obtained through this step.

[0168] In this embodiment of the application, firstly, the rule generation unit determines a preset reference direction, such as the horizontal front-back direction (positive X-axis direction). Then, it measures the angle between the virtual aerodynamic interference direction obtained in step c6 and this reference direction. This angle is the angle deviation. For example, if the virtual aerodynamic interference direction is biased to the horizontal left-right direction (positive Y-axis direction) by 10 degrees, and the reference direction is the positive X-axis direction, then the angle deviation is 10 degrees.

[0169] c8. Based on the angle deviation, rotate and adjust the preset environmental disturbance range so that the central axis of the adjusted environmental disturbance range is consistent with the direction of the virtual aerodynamic interference, and obtain the boundary parameters of the adjusted environmental disturbance range.

[0170] Among them, the angle deviation is the angle between the virtual aerodynamic interference direction and the reference direction, the preset environmental disturbance range is the pre-set environmental disturbance influence range, the central axis is the central direction line of this range, and the boundary parameters of the adjusted environmental disturbance range are the boundary data of the adjusted range. Through this step, the adjusted boundary parameters will be obtained.

[0171] In this embodiment of the application, firstly, based on the angle deviation obtained in step c7, the preset environmental disturbance range is rotated around the center so that the central axis of the range is consistent with the direction of the virtual aerodynamic disturbance obtained in step c6. For example, if the angle deviation is 10 degrees, the preset range is rotated by 10 degrees. Then, the boundary data of the rotated range is determined. For example, if the preset range is a fan-shaped area with a radius of 5 meters, after rotating by 10 degrees, the new boundary parameters include the angle range and radius of the fan-shaped area.

[0172] c9. Integrate the boundary parameters of the adjusted environmental disturbance range to generate a second boundary constraint rule for characterizing the environmental disturbance range.

[0173] Among them, the boundary parameters of the adjusted environmental disturbance range are the boundary data of the adjusted disturbance range, and the second boundary constraint rules used to characterize the environmental disturbance range are the rules that define the environmental disturbance influence range. This step will eventually generate the second boundary constraint rules.

[0174] In this embodiment of the application, firstly, the adjusted boundary parameters obtained in step c8 are collected, and then these parameters are organized into a rule to clarify the spatial range of the environmental disturbance. For example, if the boundary parameters show that the disturbance range is a fan-shaped area with a radius of 5 meters and a central axis 10 degrees off the Y-axis in the positive direction of the X-axis, it is organized into a second boundary constraint rule of "the environmental disturbance range is a fan-shaped area with a radius of 5 meters and a central axis 10 degrees off the Y-axis in the positive direction of the X-axis".

[0175] For example, in a swarm of UAVs adversarial training, when processing the air pressure data around UAV A, step c1 involves the conversion unit decomposing the air pressure change rate vector into an X-axis component of 0.2, a Y-axis component of 0.15, and a Z-axis component of 0.08; step c2 sets the X-axis interference weight coefficient to 0.4, the Y-axis to 0.4, and the Z-axis to 0.2; step c3 sets the air pressure change rate to 0.7 kPa / m, corresponding to an intensity coefficient of 1; step c4 calculates the weighting coefficients: X-axis 1 × 0.4 = 0.4, Y-axis 1 × 0.4 = 0.4, and Z-axis 1 × 0.2 = 0.2; step c5 obtains the weighted components: X-axis 0.2 × 0.4 = 0.08, Y-axis 0.15 × 0.08, and Z-axis 0.15 × 0.08. 0.4 = 0.06, Z-axis 0.08 × 0.2 = 0.016; Step c6 synthesizes the virtual aerodynamic disturbance direction as 37 degrees off the positive Y-axis direction (by calculating the arctangent of the ratio of the Y-axis to the X-axis components, 0.06 ÷ 0.08 = 0.75, corresponding to an angle of approximately 37 degrees); In step c7, the reference direction is the positive X-axis direction, with an angle deviation of 37 degrees; Step c8 rotates the preset 6-meter radius fan-shaped disturbance range by 37 degrees to obtain the adjusted boundary parameters; Step c9 integrates the parameters to generate the second boundary constraint rule of "the environmental disturbance range is a fan-shaped area with a radius of 6 meters, and the center direction is 37 degrees off the Y-axis along the positive X-axis direction".

[0176] By executing c1 to c9, this embodiment of the application separates the direction of air pressure change, calculates the degree of influence of each direction, determines the direction of virtual aerodynamic interference, adjusts the range of environmental disturbance, and generates rules, so that the direction and range of interference in the virtual environment match the actual air pressure change, making the environmental disturbance closer to the real situation, improving the accuracy of the virtual scene, and providing a more realistic environmental reference for UAV combat training.

[0177] In one possible embodiment, S12, based on the abnormal fluctuation amplitude of the power system state data and the spatial gradient change information of the hydrodynamic environment parameters, the control parameters of each individual UAV are corrected, including:

[0178] Step 121: Extract the magnitude and frequency of abnormal fluctuations from the dynamic system state data.

[0179] Among them, the power system status data is the operating information of the UAV's power unit, including changes in motor speed, power supply voltage, etc.; the magnitude of abnormal fluctuations is the degree to which these data exceed the normal range; the frequency of occurrence is how many times such abnormal changes occur; through this step, the magnitude and frequency of abnormal fluctuations will be obtained.

[0180] In this embodiment, the normal range of the power system status data is first determined. For example, the normal motor speed is 1000-1500 rpm, and the normal voltage is 12-14 volts. Then, the data that exceeds this range is identified from the collected data, the degree of change of these parts is measured, and the number of times they occur within a certain period of time is counted. For example, the motor speed of a certain drone reaches 1600 rpm several times, which is 100 rpm beyond the normal range. This is the magnitude of the abnormal fluctuation. If it occurs 3 times in 1 hour, this is the frequency of occurrence.

[0181] Step 122: Normalize the magnitude of the abnormal fluctuation to obtain the corresponding basic value, convert the occurrence frequency into a correction coefficient according to a preset ratio, and generate the fluctuation characteristic value by multiplying the basic value and the correction coefficient.

[0182] The normalization process converts the magnitude of abnormal fluctuations into values ​​between 0 and 1; the base value is the result obtained after normalization; the preset ratio is a pre-set ratio for converting the occurrence frequency into a value; the correction coefficient is the value after the occurrence frequency is converted according to the preset ratio; the fluctuation characteristic value is the result of multiplying the base value and the correction coefficient; through this step, the fluctuation characteristic value is finally obtained.

[0183] In this embodiment, the magnitude of the abnormal fluctuation obtained in step 121 is first normalized by (magnitude of abnormal fluctuation - minimum abnormal value) ÷ (maximum abnormal value - minimum abnormal value) to obtain a base value. Then, the occurrence frequency is converted into a correction coefficient according to a preset ratio, such as 0.2 for each occurrence. Finally, the two are multiplied to obtain the fluctuation characteristic value. For example, if the magnitude of the abnormal fluctuation is 100 rpm, the minimum abnormal value is 50 rpm, and the maximum is 150 rpm, the base value is (100-50) ÷ (150-50) = 0.5. If the occurrence frequency is 3 times, the correction coefficient is 3 × 0.2 = 0.6, and the fluctuation characteristic value is 0.5 × 0.6 = 0.3.

[0184] Step 123: Analyze the changes in fluid dynamic environment parameters under different spatial gradients to obtain spatial gradient change information.

[0185] Among them, the hydrodynamic environment parameters are the environmental information of fluids such as air around the UAV, including air pressure and temperature at different locations; the spatial gradient change is the trend of these parameters changing at different locations; the spatial gradient change information is a summary of these trends; through this step, the spatial gradient change information will be obtained.

[0186] In this embodiment of the application, hydrodynamic environmental parameters at different locations around the drone are first collected, such as air pressure and temperature in front, behind, and to the left and right. Then, the parameters at these locations are compared and the changing trends are analyzed. For example, if the air pressure around a drone is 101.3 kPa to the north and 101.1 kPa to the south, the temperature is 25 degrees to the east and 23 degrees to the west, the spatial gradient change information obtained after analysis is that "the air pressure in the north is 0.2 kPa higher than that in the south, and the temperature in the east is 2 degrees higher than that in the west".

[0187] Step 124: Associate the fluctuation feature values ​​with the spatial gradient change information to form a comprehensive information set.

[0188] Among them, the fluctuation characteristic value is the value obtained in step 122 that reflects the abnormal situation of the dynamic system; the spatial gradient change information is the information obtained in step 123 that reflects the environmental change; the correlation is to find the correspondence between the two; the comprehensive information set is the information combination that integrates these correspondences; through this step, the comprehensive information set will be obtained in the end.

[0189] In this embodiment of the application, the fluctuation feature value obtained in step 122 is first matched with the spatial gradient change information obtained in step 123 according to the location of the UAV. For example, if the fluctuation feature value is 0.3 at a certain location, the spatial gradient change at the same location is "high air pressure in the north and low air pressure in the south". Then, the number of times this matching relationship occurs is counted, and those that occur more often are considered as strong correlations. Finally, these strong correlation information are organized according to individual UAVs. For example, if a certain UAV has a fluctuation feature value of 0.3 multiple times at the location of "high air pressure in the north and low air pressure in the south", this relationship is organized into the comprehensive information set.

[0190] Step 125: Based on the comprehensive information set, determine the adjustment direction and initial adjustment value of the control parameters for each individual UAV.

[0191] Among them, the comprehensive information set is the set of integrated power and environmental information obtained in step 124; the control parameters are the parameters that control the flight of the UAV, such as speed and turning angle; the adjustment direction is the direction in which the parameters need to be increased or decreased; the initial adjustment value is the initially determined adjustment amount; through this step, the adjustment direction and initial adjustment value of the control parameters will be obtained in the end.

[0192] In this embodiment of the application, the comprehensive information set obtained in step 124 is first analyzed to see the combination of fluctuation characteristic value and spatial gradient change information. For example, when the fluctuation characteristic value is large and the air pressure changes drastically, it is determined that the speed needs to be reduced. Then, the adjustment direction and initial adjustment amount are determined based on this judgment. For example, if the comprehensive information set shows that a certain UAV is unstable when the fluctuation characteristic value is 0.3 and the air pressure is "higher in the north and lower in the south", the adjustment direction is determined to be to reduce the speed, and the initial adjustment value is set to 2 meters / second.

[0193] Step 126: Adjust the initial adjustment value according to the preset control parameter adjustment constraints to obtain the target adjustment value that meets the constraints.

[0194] The preset control parameter adjustment restrictions are the adjustment range set to ensure the safety of the UAV, such as the speed not being lower than 1 m / s; the initial adjustment value is the preliminary adjustment amount obtained in step 125; the target adjustment value is the final adjustment amount that meets the restrictions; the target adjustment value will be obtained through this step.

[0195] In this embodiment of the application, the preset limiting conditions are first defined, such as the speed after adjustment cannot be lower than 1 m / s. Then, it is checked whether the initial adjustment value obtained in step 125 meets these conditions. If it does not meet the conditions, it is adjusted until it does. For example, if the initial adjustment value is 2 m / s, the original speed is 5 m / s, and the adjusted speed is 3 m / s, the target adjustment value is 2 m / s. If the initial adjustment value is 4 m / s, and the adjusted speed is 1 m / s, the target value is 4 m / s.

[0196] Step 127: Adjust the values ​​according to the target, modify the control parameters of each individual UAV according to the control parameter adjustment direction, and obtain the modified control parameters of each individual UAV.

[0197] Among them, the target adjustment value is the adjustment amount that meets the constraints obtained in step 126; the control parameter adjustment direction is the direction of increase or decrease obtained in step 125; the modified control parameters are the parameters that are finally used to control the flight of the UAV after adjustment; through this step, the modified control parameters of each individual UAV will be obtained.

[0198] In this embodiment of the application, the adjustment direction obtained in step 125 and the target adjustment value obtained in step 126 are first determined. Then, the original control parameters are modified according to this direction and value. For example, the original speed parameter of a certain UAV is 5 m / s, the adjustment direction is to slow down, the target adjustment value is 2 m / s, and the modified speed parameter is 5-2=3 m / s.

[0199] For example, in a swarm UAV combat training exercise, when adjusting the control parameters of UAV A, step 121 extracts the abnormal fluctuation amplitude from its power data, finding it to be 100 rpm (out of normal range), occurring 3 times within 1 hour. In step 122, the abnormal fluctuation amplitude is normalized, with a minimum abnormal value of 50 rpm and a maximum of 150 rpm. The base value is (100-50) ÷ (150-50) = 0.5. The frequency of occurrence (3 times) is converted to a correction coefficient of 0.6 according to a preset ratio (0.2 per occurrence). The fluctuation characteristic value is 0.5 × 0.6 = 0.3. Step 123 analyzes the surrounding environment and obtains the spatial gradient change information as "the air pressure in the north is 0.2 kPa higher than in the south, and the temperature in the east is 2 degrees higher than in the west"; Step 124 associates the fluctuation characteristic value of 0.3 with this environmental information to form a comprehensive information set; Step 125 determines the adjustment direction as slowing down based on this set, with an initial adjustment value of 2 m / s; Step 126 checks and finds that the original speed of 5 m / s and the adjusted speed of 3 m / s meet the restriction condition of "not less than 1 m / s", and the target adjustment value is 2 m / s; Step 127 modifies according to the direction and target value to obtain the modified speed parameter of 3 m / s.

[0200] By executing steps 121 to 127, this embodiment of the application extracts information on abnormal conditions of the power system and changes in the surrounding environment, correlates the two, determines the adjustment direction and value of the control parameters, and ensures that the adjustment complies with safety limits, ultimately obtaining suitable control parameters. This enables the UAV's control parameters to better adapt to abnormalities in its own power system and changes in the surrounding environment, allowing the UAV to fly more stably in combat training and better cope with complex situations.

[0201] In one possible embodiment, step 124 involves associating the fluctuation feature values ​​with spatial gradient change information to form a comprehensive information set, including:

[0202] d1. For each individual drone, match the corresponding fluctuation feature value with the value at the same position in the spatial gradient change information according to the spatial gradient position of the individual drone, and record the correspondence of each pair of matching data.

[0203] Among them, an individual drone is a single drone in a cluster; spatial gradient position is the specific spatial location of the drone; fluctuation characteristic value is a value reflecting abnormal conditions of the drone's power system; spatial gradient change information is environmental change data at different locations; the value at the same location is the environmental data corresponding to the location of the drone; matching is to match the fluctuation characteristic value and environmental data at the same location; the correspondence relationship is the correlation between the two; through this step, the correspondence relationship of each pair of matched data will be recorded.

[0204] In this embodiment of the application, firstly, the specific spatial location of each drone is determined, then the environmental data corresponding to that location is found in the spatial gradient change information, and then the fluctuation characteristic value of the drone is correlated with these environmental data, and this correlation is recorded. For example, if the fluctuation characteristic value of a drone at position P is 0.4, and the temperature change in the spatial gradient change information of position P is 2℃ / meter, then the correspondence between "fluctuation characteristic value 0.4 and temperature change 2℃ / meter" is recorded.

[0205] d2. Count the frequency of occurrence of each pair of matching data in the corresponding drone individual, and mark the corresponding relationship with the frequency of occurrence greater than or equal to the preset frequency threshold as strong association.

[0206] In this process, the correspondence between each pair of matching data is the association between the fluctuation feature value recorded in step d1 and the environmental data; the frequency of occurrence is the number of times this correspondence occurs in the same drone; the preset frequency threshold is a pre-set standard for judging whether it is a strong correlation; a strong correlation is a correspondence whose frequency of occurrence reaches or exceeds the threshold; through this step, the strong correlation correspondence will be marked.

[0207] In this embodiment of the application, firstly, the number of times each pair of correspondences recorded in step d1 appears in the same drone is counted, and then this number is compared with a preset frequency threshold (e.g., 4 times). If the number reaches or exceeds the threshold, the pair of correspondences is marked as strongly associated. For example, if a certain correspondence appears 5 times in the same drone and the preset threshold is 4 times, it is marked as strongly associated.

[0208] d3. Group the strongly correlated matching data according to the individual identifiers of each UAV to obtain data groups.

[0209] Among them, the individual drone identifier is a mark (such as a number) that distinguishes different drones; the strongly associated matching data is the association relationship marked in step d2; the data grouping is to group the strongly associated data of the same drone together; through this step, the data grouping by individual drone will be obtained.

[0210] In this embodiment of the application, firstly, a unique identifier (such as number A, B, C) is assigned to each drone. Then, the strongly correlated data marked in step d2 is classified according to the corresponding identifier. Strongly correlated data belonging to the same identifier are grouped together to form data groups. For example, if drone number A has 2 pairs of strongly correlated data, these 2 pairs of data are grouped together and called data group number A.

[0211] d4. Calculate the extreme values ​​and median values ​​of the fluctuation characteristics in the statistical data group to form a characteristic interval, and simultaneously calculate the upper and lower limits and mean values ​​of the spatial gradient change information to form a numerical interval.

[0212] In this process, data grouping refers to the strongly correlated data divided by individual drones in step d3; the extreme values ​​of the fluctuation feature values ​​are the largest and smallest fluctuation feature values ​​in the data; the median value is the middle value after sorting the fluctuation feature values ​​by size; the feature interval is the range of fluctuation feature values ​​determined by the extreme values ​​and the median value; the upper and lower limits of the spatial gradient change information are the largest and smallest values ​​in the environmental data; the mean value is the average value of the environmental data; and the numerical interval is the range of environmental data determined by the upper and lower limits and the mean value. Through this step, the feature interval and the numerical interval will be formed in the end.

[0213] In this embodiment of the application, firstly, the maximum, minimum, and median values ​​of the fluctuation characteristic value are found from each data group (e.g., the maximum is 0.6, the minimum is 0.2, and the median is 0.4 in a certain group), and these three values ​​are used to determine the characteristic interval (0.2-0.6); at the same time, the maximum, minimum, and average values ​​of the spatial gradient change information in the group are calculated (e.g., the maximum air pressure change is 0.4, the minimum is 0.1, and the average is 0.25), and these three values ​​are used to determine the value interval (0.1-0.4).

[0214] d5. Based on the individual identifiers, feature ranges, and numerical ranges of each UAV, a comprehensive information set is formed.

[0215] Among them, the individual drone identifier is a mark that distinguishes different drones; the feature interval is the range of fluctuation feature values ​​determined in step d4; the numerical interval is the range of environmental data determined in step d4; the comprehensive information set is a combination of information integrating the identifier, feature interval, and numerical interval; through this step, a comprehensive information set will be formed in the end.

[0216] In this embodiment of the application, firstly, the identifier, corresponding feature range, and numerical range of each UAV are integrated together to form a set containing these three parts of information. For example, for UAV number B, the feature range is 0.3-0.7 and the numerical range is 0.2-0.5. These three are integrated into the comprehensive information of number B. The comprehensive information set is formed by summarizing this type of information of all UAVs.

[0217] For example, a cluster consists of 3 drones (numbered 1, 2, and 3). The processing procedure is as follows: In step d1, the spatial position of each drone is determined. The fluctuation characteristic value of drone number 1 at position Q (0.5) is matched with the air pressure change at that position (0.3 kPa / m), and the correspondence is recorded. In step d2, it is found that this correspondence occurs 5 times in drone number 1, and the preset frequency threshold is 4 times, so it is marked as a strong correlation. In step d3, all strongly correlated data of drone number 1 are grouped together. Drones number 2 and 3 are processed in the same way to obtain three data groups. In step d4, the fluctuation characteristic value of the data group of drone number 1 is 0.6, 0.4, and 0.5, with a characteristic range of 0.4-0.6. The air pressure change information is 0.4, 0.2, and 0.3, with a value range of 0.2-0.4. In step d5, the identifier, characteristic range of 0.4-0.6, and value range of 0.2-0.4 of drone number 1 are integrated. Other drones are processed in the same way to finally form a comprehensive information set.

[0218] By executing d1~d5, this embodiment of the application identifies frequently occurring correlations by correlating the abnormal power data of the UAV with the environmental data of its location. Then, it classifies the UAVs and determines the data range. The resulting comprehensive information set can clearly reflect the correlation between the power characteristics of each UAV and environmental changes in different environments. This provides a clear and reliable reference for subsequent adjustments to the UAV control parameters, making the adjustments more in line with actual operating conditions.

[0219] In one possible embodiment, S16, based on the corrected control parameters and actual response results of multiple individual UAVs according to distributed node exchange, collaboratively generates an anti-interference swarm flight strategy, including:

[0220] Step 161: Based on distributed nodes, control each individual drone to send its actual location information, corrected control parameters, and actual response results to other individuals in the cluster at preset time intervals, and at the same time receive the actual location information, corrected control parameters, and actual response results sent by other individuals, so as to form a cluster dataset containing the actual location information, corrected control parameters, and actual response results of all individual drones.

[0221] Among them, distributed nodes are multiple units in the cluster responsible for processing and exchanging information, preset time interval is a fixed time for drones to send information, actual location information is the current location of the drone, corrected control parameters are the adjusted flight control data, actual response result is the performance of the drone flying according to the corrected parameters, and cluster dataset is the summary of all this information from all drones. Through this step, a cluster dataset containing information from all drones will eventually be formed.

[0222] In this embodiment, firstly, the distributed nodes control each drone to send its actual position, corrected control parameters, and actual flight performance to other drones in the cluster at fixed intervals (e.g., 8 seconds). At the same time, each drone receives this information sent by other drones. Finally, all the received information and its own information are aggregated to form a set containing all drone data. For example, if a cluster has 15 drones, each drone sends its own position coordinates, adjusted flight speed, and actual flight trajectory every 8 seconds, while receiving information from the other 14 drones. This aggregated information forms a cluster dataset.

[0223] Step 162: Extract the adjustment magnitude and adjustment frequency from the corrected control parameters of each individual UAV from the cluster dataset.

[0224] Among them, the cluster dataset is all the UAV information summarized in step 161, the corrected control parameters are the adjusted flight control data, the adjustment magnitude is the size of the change of the control parameters compared with the original parameters, and the adjustment frequency is the number of times the control parameters change per unit time. Through this step, the adjustment magnitude and adjustment frequency of each UAV will be extracted.

[0225] In this embodiment of the application, firstly, the corrected control parameters of each UAV are found from the cluster dataset, and these parameters are subtracted from their original parameters. The difference obtained is the adjustment range. At the same time, the number of times the control parameters of each UAV change within a unit of time (e.g., 1 minute) is counted to obtain the adjustment frequency. For example, if the original flight speed of a UAV is 6 m / s and the corrected speed is 4 m / s, the adjustment range is 2 m / s. If the speed is adjusted 3 times within 1 minute, the adjustment frequency is 3 times / minute.

[0226] Step 163: Compare the degree of deviation between the adjustment range and the adjustment frequency and the actual response results. Determine the maneuver requirements of each individual UAV based on the degree of deviation. Different maneuver requirements correspond to different levels of urgency.

[0227] Among them, the adjustment range is the magnitude of the change in control parameters, the adjustment frequency is the number of times the parameters change per unit time, the actual response result is the flight performance of the UAV, the deviation degree is the gap between the adjusted parameters and the actual flight performance, the maneuvering requirement is the necessity for the UAV to adjust its flight state, and the urgency level is the priority of the maneuvering requirement. Through this step, the maneuvering requirements and corresponding urgency level of each UAV will be determined.

[0228] In this embodiment, firstly, the adjustment range and frequency of each UAV are compared with the actual response result, and the difference between them is calculated (for example, the adjustment range is 2 m / s, but the actual flight speed is only reduced by 1 m / s, so the deviation is 1 m / s). Then, different thresholds are set according to the degree of deviation (for example, a deviation exceeding 1.2 m / s is considered high urgency). Based on the threshold, the maneuver requirements and corresponding urgency of each UAV are determined. For example, if the deviation of a certain UAV is 1.5 m / s, which exceeds the set high urgency threshold, it is determined to be a high urgency deceleration requirement.

[0229] Step 164: Extract the actual position information of each individual drone from the cluster dataset, and calculate the distance and relative position change rate between adjacent individual drones.

[0230] Among them, the actual location information is the current location of the drone, the neighboring drones are drones that are close to each other, the distance is the spatial interval between the neighboring drones, and the relative position change rate is the speed at which the distance between the neighboring drones changes. Through this step, the distance and relative position change rate between the neighboring drones will be calculated.

[0231] In this embodiment of the application, firstly, the actual position information (usually represented by coordinates) of each UAV is extracted from the cluster dataset, and the spatial interval (distance) between the coordinates of adjacent UAVs is calculated by taking the square root of the sum of the squares of the differences between the two coordinates; then, the distance at different time points is recorded, and the change in distance is divided by the change in time to obtain the relative position change rate.

[0232] Step 165: Determine the cluster formation constraints based on the distance and relative position change rate between adjacent UAV individuals, wherein the distance between adjacent UAV individuals is inversely proportional to the strength of the corresponding cluster formation constraints.

[0233] Among them, the distance between adjacent UAVs is the spatial interval between them, the relative position change rate is the speed at which the distance changes, the swarm formation constraint is the rule to maintain the swarm formation, the strength refers to the strictness of these rules, and the distance between adjacent UAVs is inversely proportional to the strength of the swarm formation constraint, that is, the closer the distance, the stricter the rule, and the farther the distance, the more relaxed the rule can be. Through this step, the swarm formation constraint will be determined.

[0234] In this embodiment of the application, firstly, based on the distance and relative position change rate between adjacent drones obtained in step 164, different constraint strengths are set. For example, when the distance is 2-3 meters and the change rate is small (e.g., within 0.2 m / s), the constraint strength is high (the distance must be strictly maintained); when the distance is 4-5 meters and the change rate is large (e.g., around 0.5 m / s), the constraint strength is low (the distance can be appropriately relaxed). For example, if the distance between adjacent drones is 3.5 meters and the relative position change rate is 0.3 m / s, the constraint strength is determined to be medium, requiring the distance between the two to be maintained between 3-4 meters.

[0235] Step 166: Based on the urgency of the maneuvering needs of each individual UAV and the strength of the cluster formation constraints, integrate the adjustment schemes of all individual UAVs into an anti-interference group flight strategy.

[0236] Among them, the urgency of maneuvering needs refers to the priority of UAVs adjusting their flight status, the strength of swarm formation constraints refers to the strictness of maintaining swarm formation rules, the adjustment plan is the flight adjustment plan of a single UAV, and the anti-interference swarm flight strategy is the overall collaborative flight plan of the swarm. Through this step, the anti-interference swarm flight strategy will be generated.

[0237] In this embodiment, firstly, the urgency of the maneuvering needs of each UAV is comprehensively considered (high-urgency needs are processed first) and the strength of the cluster formation constraints (high-strength constraints must prioritize maintaining the formation). Then, the adjustment schemes of all UAVs are integrated together to ensure that both the urgent adjustment needs of individual UAVs are met and the formation rules of the cluster are followed. For example, if a UAV has a high-urgency deceleration need, while its neighboring UAVs have medium-strength formation constraints (need to maintain a distance of 3 meters), the integrated strategy may be "the UAV decelerates by 2 meters / second, and the neighboring UAVs simultaneously decelerate by 1 meter / second to maintain a distance of 3 meters".

[0238] For example, a cluster consists of 12 drones. In step 161, each drone sends its own position coordinates, corrected flight speed, and actual flight trajectory to the other drones every 10 seconds, while simultaneously receiving information from the other 11 drones, summarizing them to form a cluster dataset. In step 162, the speed adjustment range of drone C is extracted from the dataset as 1.5 m / s (original speed 5 m / s, corrected to 3.5 m / s), with an adjustment frequency of once per minute. In step 163, it is found that the deviation between the adjustment range of drone C and the actual response result is 1.3 m / s, which is identified as a high-urgency deceleration requirement. In step 164, calculations are performed... The distance between UAV C and its neighboring UAV D is 4 meters. After 10 seconds, the distance becomes 4.5 meters, and the relative position change rate is (4.5-4) / 10=0.05 meters / second. In step 165, based on the distance of 4 meters and the change rate of 0.05 meters / second, the formation constraint between the two is determined to be of medium intensity (maintaining a distance of 3.5-4.5 meters). In step 166, taking into account the high emergency deceleration requirement of UAV C and the medium formation constraint, the adjustment schemes of all UAVs are integrated to generate an anti-interference group flight strategy of "UAV C decelerates by 1.5 meters / second, UAV D decelerates by 0.5 meters / second to maintain a distance of 4 meters, and other UAVs maintain their current speed".

[0239] By executing steps 161 to 166, this embodiment of the application realizes information sharing among UAVs through distributed nodes, extracts key adjustment parameters and determines the urgent needs of individuals in combination with actual flight performance, and formulates the formation constraints of the cluster based on positional relationships. Finally, the integrated anti-interference group flight strategy can not only meet the urgent adjustment needs of individual UAVs, but also ensure the stability of the cluster formation, thereby improving the cluster's ability to cooperate and cope with interference in complex environments.

[0240] Figure 2 This is a schematic diagram of the structure of an XX device (or system) provided in an embodiment of this application, such as... Figure 2 As shown, the device includes:

[0241] The acquisition module 21 is used to acquire the power system status data and hydrodynamic environment parameters of multiple individual UAVs in the cluster.

[0242] The correction module 22 is used to correct the control parameters of each UAV based on the abnormal fluctuation amplitude of the power system state data and the spatial gradient change information of the hydrodynamic environment parameters.

[0243] Module 23 is used to construct adversarial training scenarios based on the fluctuation characteristics of dynamic system state data and the spatial gradient distribution of fluid dynamic environment parameters through a digital twin model of a distributed digital twin architecture.

[0244] The recording module 24 is used to load the corresponding modified control parameters into each individual UAV and record the actual response results of the individual UAV in the adversarial training scenario.

[0245] The adjustment module 25 is used to compare the deviation between the actual response result and the expected behavior of the digital twin model. When the deviation exceeds the dynamic fault tolerance threshold, the fault tolerance adjustment mechanism is triggered to reconstruct the correction logic of the control parameters.

[0246] The generation module 26 is used to collaboratively generate an anti-interference swarm flight strategy based on the corrected control parameters and actual response results of multiple individual UAVs exchanged by distributed nodes.

[0247] Figure 2 The distributed digital twin architecture-based swarm drone adversarial training collaborative optimization system can execute... Figure 1 The implementation principle and technical effects of the distributed digital twin architecture-based swarm UAV adversarial training collaborative optimization method described in the illustrated embodiment will not be repeated here. The specific methods by which each module and unit of the distributed digital twin architecture-based swarm UAV adversarial training collaborative optimization system in the above embodiments are described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0248] In one possible design, Figure 2 The distributed digital twin architecture-based swarm drone adversarial training collaborative optimization system of the illustrated embodiment can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32.

[0249] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0250] The processing component 32 is used to perform the following process: acquire the power system status data and hydrodynamic environment parameters of multiple UAVs in the cluster; and correct the control parameters of each UAV based on the abnormal fluctuation amplitude of the power system status data and the spatial gradient change information of the hydrodynamic environment parameters.

[0251] Using a distributed digital twin architecture, an adversarial training scenario is constructed based on the fluctuation characteristics of the dynamic system state data and the spatial gradient distribution of the hydrodynamic environment parameters. Corresponding modified control parameters are loaded into each individual UAV, and the actual response results of the individual UAVs in the adversarial training scenario are recorded. The deviation between the actual response results and the expected behavior of the digital twin model is compared. When the deviation exceeds the dynamic fault tolerance threshold, a fault tolerance adjustment mechanism is triggered to reconstruct the correction logic of the control parameters. Based on the modified control parameters and actual response results of multiple individual UAVs exchanged through distributed nodes, an anti-interference swarm flight strategy is collaboratively generated.

[0252] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0253] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as Random Access Memory (RAM), Static Random-Access Memory (SRAM), Erasable Programmable Read Only Memory (EPROM), Electrically Erasable Programmable Read Only Memory (EEPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0254] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0255] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0256] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0257] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0258] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The illustrated embodiment presents a collaborative optimization method for clustered UAV adversarial training using a distributed digital twin architecture.

[0259] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0260] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0261] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0262] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A collaborative optimization method for swarm UAV adversarial training based on a distributed digital twin architecture, characterized in that, include: Acquire the power system status data and hydrodynamic environment parameters of multiple individual drones in the cluster; Based on the abnormal fluctuation amplitude of the power system state data and the spatial gradient change information of the hydrodynamic environment parameters, the control parameters of each UAV are corrected. Using a digital twin model with a distributed digital twin architecture, an adversarial training scenario is constructed based on the fluctuation characteristics of the dynamic system state data and the spatial gradient distribution of the fluid dynamic environment parameters. After loading the corresponding modified control parameters into each of the aforementioned drones, the actual response results of the drones in the adversarial training scenario are recorded. By comparing the deviation between the actual response result and the expected behavior of the digital twin model, when the deviation exceeds the dynamic fault tolerance threshold, a fault tolerance adjustment mechanism is triggered to reconstruct the correction logic of the control parameters. Based on the corrected control parameters of multiple individual UAVs and the actual response results obtained from distributed node exchange, an anti-interference swarm flight strategy is collaboratively generated. The digital twin model, built using a distributed digital twin architecture, constructs an adversarial training scenario based on the fluctuation characteristics of the dynamic system state data and the spatial gradient distribution of the hydrodynamic environment parameters, including: Through the feature extraction module in the digital twin model, corresponding fluctuation features are extracted from the dynamic state parameters of each UAV individual. The fluctuation features include the rate of change of rotational speed fluctuation amplitude and the voltage anomaly offset. At the same time, the spatial gradient distribution is extracted from the hydrodynamic environment parameters of each UAV individual. The spatial gradient distribution includes the air pressure change rate vector and the temperature field contour density. The first rule setting module in the digital twin model maps the rate of change of the rotational speed fluctuation amplitude to the motion performance attenuation coefficient of the virtual entity, and generates a first boundary constraint rule to characterize the collision volume of the virtual entity based on the motion performance attenuation coefficient. The pressure change rate vector is converted into a virtual aerodynamic disturbance direction through the second rule setting module in the digital twin model, and a second boundary constraint rule is generated based on the aerodynamic disturbance direction to characterize the range of environmental disturbance. The communication delay parameters are reset based on the voltage anomaly offset using the first strategy generation module in the digital twin model to obtain an anti-interference communication strategy. The second strategy generation module in the digital twin model redistributes the coordinates of the virtual obstacle based on the temperature field contour density to obtain an environment-adaptive obstacle avoidance strategy. Through the fusion module in the digital twin model, an adversarial training scenario is constructed based on the first boundary constraint rule, the second boundary constraint rule, the anti-interference communication strategy, and the environment adaptive obstacle avoidance strategy.

2. The method for collaborative optimization of swarm UAV adversarial training based on a distributed digital twin architecture according to claim 1, characterized in that, The first rule setting module in the digital twin model maps the rate of change of the rotational speed fluctuation amplitude to the maneuverability attenuation coefficient of the virtual entity, and generates a first boundary constraint rule to characterize the collision volume of the virtual entity based on the maneuverability attenuation coefficient, including: The mapping unit in the first rule setting module divides the rate of change of fluctuation amplitude according to the threshold corresponding to the preset interval to obtain the interval corresponding to the rate of change of rotational speed fluctuation amplitude. Based on the calculation method corresponding to the interval, a value for the degree of propulsion reduction is generated; After classifying and mapping the propulsion reduction values, they are converted into the motion performance attenuation coefficient of the virtual entity; The rule generation unit in the first rule setting module determines the baseline collision volume parameters of the virtual entity under normal conditions and the adjustment ratio corresponding to the baseline collision volume parameters based on the mobility performance attenuation coefficient. The baseline collision volume parameters include baseline length, baseline width and baseline height. By adjusting the corresponding adjustment ratios, the reference length, the reference width, and the reference height are adjusted respectively to obtain the adjusted length, adjusted width, and adjusted height; The three-dimensional spatial range defined by the adjusted length, the adjusted width, and the adjusted height is converted into the maximum offset value along the X-axis, Y-axis, and Z-axis directions. The maximum offset value is the boundary value that the virtual entity cannot cross when it moves. Based on the boundary values, a first boundary constraint rule is generated to characterize the collision volume of the virtual entity.

3. The method for collaborative optimization of swarm UAV adversarial training based on a distributed digital twin architecture according to claim 1, characterized in that, The step involves using the second rule setting module in the digital twin model to convert the pressure change rate vector into a virtual aerodynamic disturbance direction, and generating a second boundary constraint rule based on the aerodynamic disturbance direction to characterize the range of environmental disturbance effects, including: The conversion unit in the second rule setting module decomposes the directional information of air pressure change in the air pressure change rate vector into components along the X-axis, Y-axis and Z-axis. A corresponding interference weight coefficient is set for the components in each coordinate axis direction. The interference weight coefficient is used to represent the degree of influence of the components in the corresponding coordinate axis direction on the virtual aerodynamic interference direction. The intensity information in the pressure change rate vector is converted into an intensity coefficient; The intensity coefficient is multiplied by the interference weight coefficient in each coordinate axis direction to obtain the weighting coefficient in each coordinate axis direction. Based on the weighting coefficients, the components of each coordinate axis direction are weighted and calculated to obtain the weighted direction components; The weighted directional components are combined to obtain the virtual aerodynamic interference direction; The rule generation unit in the second rule setting module calculates the angular deviation between the virtual aerodynamic interference direction and the preset environmental disturbance range reference direction. Based on the angular deviation, the preset environmental disturbance range is rotated and adjusted so that the central axis of the adjusted environmental disturbance range is consistent with the direction of the virtual aerodynamic interference, thereby obtaining the boundary parameters of the adjusted environmental disturbance range. The boundary parameters of the adjusted environmental disturbance range are integrated to generate a second boundary constraint rule for characterizing the environmental disturbance range.

4. The method for collaborative optimization of swarm UAV adversarial training based on a distributed digital twin architecture according to claim 1, characterized in that, The step of correcting the control parameters of each individual UAV based on the abnormal fluctuation amplitude of the power system state data and the spatial gradient change information of the hydrodynamic environment parameters includes: Extract the magnitude and frequency of abnormal fluctuations from the dynamic system state data; The magnitude of the abnormal fluctuation is normalized to obtain the corresponding basic value. The occurrence frequency is converted into a correction coefficient according to a preset ratio. The fluctuation characteristic value is generated by multiplying the basic value and the correction coefficient. The variation of the fluid dynamic environment parameters under different spatial gradients is analyzed to obtain spatial gradient variation information; The fluctuation characteristic value is correlated with the spatial gradient change information to form a comprehensive information set; Based on the comprehensive information set, the control parameter adjustment direction and initial adjustment value for each individual UAV are determined. The initial adjustment value is adjusted according to the preset control parameter adjustment constraints to obtain the target adjustment value that meets the constraints. Based on the target adjustment value, the control parameters of each individual UAV are modified according to the control parameter adjustment direction to obtain the modified control parameters of each individual UAV.

5. The method for collaborative optimization of swarm UAV adversarial training based on a distributed digital twin architecture according to claim 4, characterized in that, The step of associating the fluctuation feature value with the spatial gradient change information to form a comprehensive information set includes: For each individual drone, the corresponding fluctuation feature value is matched with the value at the same position in the spatial gradient change information according to the spatial gradient position of the individual drone, and the correspondence of each pair of matched data is recorded. The frequency of occurrence of each pair of matching data in the corresponding drone individual is counted, and the corresponding relationships with a frequency greater than or equal to a preset frequency threshold are marked as strong associations; The strongly correlated matching data is grouped according to the individual identifier of each drone to obtain data groups; The extreme values ​​and median values ​​of the fluctuation characteristic values ​​in the data group are statistically analyzed to form a characteristic interval. The upper and lower limits and the mean of the spatial gradient change information are calculated simultaneously to form a numerical interval. A comprehensive information set is formed based on the individual identifiers of each UAV, the feature ranges, and the numerical ranges.

6. The method for collaborative optimization of swarm UAV adversarial training based on a distributed digital twin architecture according to claim 1, characterized in that, The corrected control parameters of multiple individual UAVs based on distributed node exchange and the actual response results are used to collaboratively generate an anti-interference swarm flight strategy, including: Based on distributed nodes, each individual drone is controlled to send its actual location information, corrected control parameters, and actual response results to other individuals in the cluster at preset time intervals. At the same time, it receives the actual location information, corrected control parameters, and actual response results sent by other individuals to form a cluster dataset containing the actual location information, corrected control parameters, and actual response results of all individual drones. Extract the adjustment magnitude and adjustment frequency from the corrected control parameters of each individual UAV from the cluster dataset; The degree of deviation between the adjustment range and adjustment frequency and the actual response results is compared, and the maneuver requirements of each individual UAV are determined based on the degree of deviation. Different maneuver requirements correspond to different levels of urgency. Extract the actual position information of each individual drone from the cluster dataset, and calculate the distance and relative position change rate between adjacent drones. Based on the distance and relative position change rate between adjacent UAV individuals, the cluster formation constraint is determined, wherein the distance between adjacent UAV individuals is inversely proportional to the strength of the corresponding cluster formation constraint; Based on the urgency of the maneuvering needs of each individual UAV and the strength of the swarm formation constraints, the adjustment schemes of all individual UAVs are integrated into an anti-interference swarm flight strategy.

7. A distributed digital twin architecture-based collaborative optimization system for swarm UAV adversarial training, characterized in that, include: The acquisition module is used to acquire the power system status data and hydrodynamic environment parameters of multiple individual UAVs in the cluster; The correction module is used to correct the control parameters of each UAV based on the abnormal fluctuation amplitude of the power system state data and the spatial gradient change information of the hydrodynamic environment parameters. The building module is used to construct adversarial training scenarios based on the fluctuation characteristics of dynamic system state data and the spatial gradient distribution of hydrodynamic environment parameters through a digital twin model of a distributed digital twin architecture. The recording module is used to load the corresponding modified control parameters into each of the individual drones and record the actual response results of the individual drones in the adversarial training scenario. The adjustment module is used to compare the deviation between the actual response result and the expected behavior of the digital twin model. When the deviation exceeds the dynamic fault tolerance threshold, the fault tolerance adjustment mechanism is triggered to reconstruct the correction logic of the control parameters. The generation module is used to collaboratively generate an anti-interference swarm flight strategy based on the corrected control parameters of multiple individual UAVs exchanged by distributed nodes and the actual response results. The digital twin model, built using a distributed digital twin architecture, constructs an adversarial training scenario based on the fluctuation characteristics of the dynamic system state data and the spatial gradient distribution of the hydrodynamic environment parameters, including: Through the feature extraction module in the digital twin model, corresponding fluctuation features are extracted from the dynamic state parameters of each UAV individual. The fluctuation features include the rate of change of rotational speed fluctuation amplitude and the voltage anomaly offset. At the same time, the spatial gradient distribution is extracted from the hydrodynamic environment parameters of each UAV individual. The spatial gradient distribution includes the air pressure change rate vector and the temperature field contour density. The first rule setting module in the digital twin model maps the rate of change of the rotational speed fluctuation amplitude to the motion performance attenuation coefficient of the virtual entity, and generates a first boundary constraint rule to characterize the collision volume of the virtual entity based on the motion performance attenuation coefficient. The pressure change rate vector is converted into a virtual aerodynamic disturbance direction through the second rule setting module in the digital twin model, and a second boundary constraint rule is generated based on the aerodynamic disturbance direction to characterize the range of environmental disturbance. The communication delay parameters are reset based on the voltage anomaly offset using the first strategy generation module in the digital twin model to obtain an anti-interference communication strategy. The second strategy generation module in the digital twin model redistributes the coordinates of the virtual obstacle based on the temperature field contour density to obtain an environment-adaptive obstacle avoidance strategy. Through the fusion module in the digital twin model, an adversarial training scenario is constructed based on the first boundary constraint rule, the second boundary constraint rule, the anti-interference communication strategy, and the environment adaptive obstacle avoidance strategy.

8. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement a distributed digital twin architecture for collaborative optimization of clustered UAV adversarial training as described in any one of claims 1-6.

9. A computer storage medium, characterized in that, The system stores a computer program, which, when executed by a computer, implements a collaborative optimization method for swarm drone adversarial training based on a distributed digital twin architecture as described in any one of claims 1-6.

Images