Multi-agent collaborative unmanned aerial vehicle cluster simulation training method, system and device and storage medium

Abstract

The invention relates to the technical field of multi-agent cooperation, provides a multi-agent cooperative unmanned aerial vehicle cluster simulation training method, system and device, and a storage medium, and solves the problems of low penetration success rate and poor formation cooperation efficiency of an unmanned aerial vehicle cluster. The method comprises the following steps: collecting pigeon flock biological motion data, cluster trajectory data when an unmanned aerial vehicle cluster executes a historical penetration task, and magnetic field gradient data of a simulation environment; constructing a three-dimensional path library, and calculating geomagnetic course angle deviation; on the basis of the geomagnetic course angle deviation and the three-dimensional path library, a multi-target collaborative path is generated in combination with a swarm intelligent optimization model; generating a flight control instruction based on the multi-target cooperation path in combination with the real-time state data of the unmanned aerial vehicle cluster; and loading the flight control instruction to a simulation model of the unmanned aerial vehicle cluster, and driving the unmanned aerial vehicle cluster to carry out cooperative defense penetration training in a simulation environment. According to the invention, the penetration success rate and formation cooperation efficiency of the unmanned aerial vehicle cluster in a complex electromagnetic environment are improved.

Description

Technical Field

[0001] The present application relates to the field of multi-agent collaborative technology, and in particular to a multi-agent collaborative drone cluster simulation training method and system. Background Art

[0002] UAV swarm penetration training in complex electromagnetic interference environments requires addressing heading deviations caused by strong magnetic field disturbances, dynamic obstacle avoidance, and multi-aircraft coordinated path planning. Traditional simulation training methods struggle to adapt to the real-time path optimization requirements in complex terrains such as mountains and urban canyons. There is an urgent need for an efficient training solution that integrates environmental perception, bionic path optimization, and intelligent decision-making.

[0003] An existing solution uses a reinforcement learning-based multi-UAV collaborative path planning method. This method generates swarm flight trajectories using a pre-trained neural network model and dynamically adjusts the path based on real-time sensor data. This solution uses historical mission data to optimize model parameters and validates the effectiveness of swarm collaboration in a simulation environment.

[0004] This solution relies on a large amount of pre-training data, and the model's generalization ability is limited by the coverage of the training samples; the path planning process lacks an active compensation mechanism for geomagnetic interference, resulting in insufficient heading stability during actual flight; in addition, bionic behavioral characteristics are not deeply integrated into the decision-making model, making it difficult to adapt to the rapid response needs of sudden terrain changes. Summary of the Invention

[0005] The present application provides a multi-agent collaborative drone swarm simulation training method and system to solve the problems of low penetration success rate and poor formation coordination efficiency of drone swarms in complex electromagnetic environments in the existing technology.

[0006] In a first aspect, the present application provides a multi-agent collaborative drone swarm simulation training method, comprising: Collect biological movement data of pigeon flocks, cluster trajectory data of drone swarms when performing historical penetration missions, and magnetic field gradient data of simulated environments; Constructing a three-dimensional path library based on the cluster trajectory data and the pigeon flock biological motion data; Calculating the geomagnetic heading angle deviation according to the magnetic field gradient data; Based on the geomagnetic heading angle deviation and the three-dimensional path library, a multi-objective collaborative path is generated in combination with a swarm intelligence optimization model; Based on the multi-target collaborative path and combined with the real-time status data of the UAV cluster, a flight control instruction is generated; The flight control instructions are loaded into a simulation model of the UAV cluster to drive the UAV cluster to perform coordinated penetration training in a simulated environment.

[0007] Optionally, the generating of a multi-objective collaborative path based on the geomagnetic heading angle deviation and the three-dimensional path library in combination with a swarm intelligence optimization model includes: Retrieving a set of path points matching the current terrain from the three-dimensional path library; Correcting the heading coordinates of each path point in the path point set according to the geomagnetic heading angle deviation; identifying terrain feature change points in the corrected path point set, and determining an avoidance type identifier based on the terrain feature change points; Inputting the avoidance type identifier and the corrected path point set into a swarm intelligence optimization model, and outputting a terrain avoidance constraint path segment corresponding to the avoidance type identifier; All the terrain avoidance constraint path segments are spliced ​​together to form a multi-target collaborative path.

[0008] Optionally, inputting the avoidance type identifier and the corrected path point set into a swarm intelligence optimization model and outputting a terrain avoidance constraint path segment corresponding to the avoidance type identifier includes: Inputting the avoidance type identifier and the corrected path point set into a swarm intelligence optimization model, and calling a corresponding terrain spatial constraint relationship from a preset constraint rule library based on the avoidance type identifier, wherein the terrain spatial constraint relationship includes an obstacle spacing threshold and a height safety margin; Performing a path feasibility check on the corrected path point set according to the terrain spatial constraint relationship, and eliminating conflicting path points that are smaller than the obstacle spacing threshold or smaller than the height safety margin; assigning a corresponding first priority value to the avoidance type identifier based on a degree of association between the avoidance type identifier and the task; Based on the first priority value, the non-conflicting path points in the verified path point set are reorganized in sequence to generate a terrain avoidance constraint path segment.

[0009] Optionally, the step of sequentially reorganizing the non-conflicting path points in the verified path point set based on the first priority value to generate a terrain avoidance constraint path segment includes: Determining an urgency level identifier corresponding to the first priority level value based on a preset mapping table between a first priority level value range and an urgency level; According to the first priority value and in combination with the terrain adaptability score, a corresponding second priority value is assigned to each non-conflicting path point; Sort the second priority values ​​of all non-conflicting waypoints according to the sorting rules corresponding to the urgency identifiers to generate an ordered waypoint sequence; Performing smooth interpolation processing on adjacent non-conflicting path points in the ordered path point sequence to form a continuous flight trajectory; A terrain constraint condition corresponding to the avoidance type identifier is applied to the continuous flight trajectory to generate a terrain avoidance constraint path segment.

[0010] Optionally, constructing a three-dimensional path library based on the cluster trajectory data and the pigeon flock biological motion data includes: Classifying the cluster trajectory data into canyon crossing trajectory, climbing trajectory and diving trajectory according to terrain type; Extracting flight turning features and altitude change features from the pigeon flock biological motion data; Mapping the flight turning feature to the canyon crossing segment trajectory, and mapping the altitude change feature to the climbing segment trajectory and the diving segment trajectory; Based on the mapping results, corresponding terrain-adaptive path units are generated, and all the terrain-adaptive path units are aggregated to form a three-dimensional path library.

[0011] Optionally, calculating the geomagnetic heading angle deviation according to the magnetic field gradient data includes: extracting the magnetic field intensity component of the current measurement point from the magnetic field gradient data; Determining the theoretical magnetic field direction of the current measurement point according to a preset geomagnetic reference field model; Calculating a direction deviation between the magnetic field intensity component and the theoretical magnetic field direction; The direction deviation is converted into an angular offset in a heading coordinate system, and the angular offset is used as the geomagnetic heading angle deviation.

[0012] Optionally, generating a flight control instruction based on the multi-target collaborative path and in combination with real-time status data of the UAV cluster includes: Extracting the flight attitude of each UAV and the relative position of the enemy aircraft from the real-time status data; Analyzing the coordinates and arrival time limit of each UAV at the next navigation point from the multi-target collaborative path; For each UAV, the coordinates, the arrival time limit, the flight attitude, and the relative position of the enemy aircraft are integrated to calculate the roll angle control parameter and the angle of attack control parameter; The roll angle control parameters and the attack angle control parameters of all UAVs are encapsulated as flight control instructions.

[0013] In a second aspect, the present application provides a multi-agent collaborative drone swarm simulation training system, comprising: The acquisition module is used to collect biological movement data of pigeon flocks, cluster trajectory data of drone clusters when performing historical penetration missions, and magnetic field gradient data of simulated environments; A construction module, for constructing a three-dimensional path library based on the cluster trajectory data and the pigeon flock biological motion data; A calculation module, configured to calculate a geomagnetic heading angle deviation based on the magnetic field gradient data; A first generation module is configured to generate a multi-objective collaborative path based on the geomagnetic heading angle deviation and the three-dimensional path library in combination with a swarm intelligence optimization model; A second generation module is used to generate flight control instructions based on the multi-target collaborative path and in combination with the real-time status data of the UAV cluster; The loading module is used to load the flight control instructions into the simulation model of the UAV cluster, driving the UAV cluster to perform coordinated penetration training in a simulated environment.

[0014] In a third aspect, the present application provides a computing device comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a multi-agent collaborative drone cluster simulation training method as described in any one of the first aspects.

[0015] In a fourth aspect, the present application provides a computer storage medium having computer program instructions stored thereon, which, when executed by a processor, implement a multi-agent collaborative drone cluster simulation training method as described in any one of the first aspects.

[0016] In the present application, a multi-agent collaborative drone cluster simulation training method is provided, which includes: collecting pigeon flock biological motion data, cluster trajectory data of the drone cluster when performing historical penetration missions, and magnetic field gradient data of the simulated environment; constructing a three-dimensional path library based on the cluster trajectory data and the pigeon flock biological motion data; calculating the geomagnetic heading angle deviation according to the magnetic field gradient data; generating a multi-target collaborative path based on the geomagnetic heading angle deviation and the three-dimensional path library in combination with a swarm intelligence optimization model; generating flight control instructions based on the multi-target collaborative path in combination with the real-time status data of the drone cluster; loading the flight control instructions into a simulation model of the drone cluster to drive the drone cluster to perform collaborative penetration training in a simulated environment.

[0017] The technical solution provided by this application has the following beneficial effects: This application provides a biomimetic basis, historical experience data, and environmental perception foundation for the training system. It forms a standardized path reference system that integrates biomimetic features and historical experience. It accurately quantifies heading errors in complex electromagnetic environments. It outputs an optimized path plan that takes into account terrain avoidance, formation coordination, and electromagnetic compensation. It converts the planned path into executable, high-precision flight control parameters. This enables full-process closed-loop verification and enhances swarm coordination capabilities.

[0018] Furthermore, this application also ensures that the path planning matches the real-time environment by calling the terrain matching path point set and correcting the heading coordinates; determines the avoidance type identification by identifying the terrain feature change points to achieve targeted path optimization; and finally splices the optimized path segments to form a complete collaborative path.

[0019] Moreover, this process achieves dynamic adaptability of drone cluster path planning and accuracy of tactical avoidance in complex terrain, improving the authenticity and effectiveness of penetration training.

[0020] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0022] Figure 1 A flowchart of a multi-agent collaborative drone swarm simulation training method provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of a multi-agent collaborative drone swarm simulation training system provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to enable people skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0024] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "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 being different types.

[0025] Existing reinforcement learning-based approaches to drone swarm collaborative training suffer from three key issues: First, the training process requires massive amounts of data, but the availability of high-quality training samples is limited, leading to unstable model performance in unfamiliar environments. Second, the system lacks a dedicated mechanism for handling geomagnetic interference, making it prone to heading deviations in complex electromagnetic environments. Finally, path planning lacks the ability to leverage the intelligence of biological swarms, making it inflexible in the face of sudden terrain changes. These issues make existing training systems incapable of meeting the demands of practical training.

[0026] In response to these problems, this application proposes a multi-agent collaborative drone swarm simulation training method. This method innovatively introduces pigeon flock biological motion data to optimize path planning, and automatically corrects heading deviations by measuring geomagnetic field data in real time. Specifically, the system analyzes the flight characteristics of the pigeon flock to establish a more intelligent path library, and dynamically adjusts the flight route in combination with real-time geomagnetic data. This design not only reduces the dependence on the amount of training data, but also effectively offsets electromagnetic interference, and also enables drone swarms to have environmental adaptability similar to biological groups. Through simulation verification, this method improves the collaborative combat effectiveness of drone swarms in complex environments and solves the problem of insufficient adaptability of existing technologies.

[0027] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0028] Figure 1 A flowchart of a multi-agent collaborative drone swarm simulation training method provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes: Step 101: Collect biological movement data of pigeon flocks, cluster trajectory data of drone clusters when performing historical penetration missions, and magnetic field gradient data of simulated environments.

[0029] In step 101, the pigeon bio-motion data represents the motion characteristics of the pigeons during flight, including flight direction changes and altitude adjustments. The cluster trajectory data represents the position and attitude changes of the drone swarm during mission execution. The magnetic field gradient data represents information acquired by sensors reflecting changes in spatial magnetic field strength.

[0030] In this application's implementation, observation equipment is used to collect flight data from a flock of pigeons in their natural state, focusing on recording their turning angles and altitude changes. Simultaneously, flight trajectory information from previous drone missions, including location coordinates and flight attitude, is retrieved. A magnetic field sensor network is deployed in a simulated training environment to monitor changes in magnetic field intensity at various points in space in real time. These three types of data serve as the basis for subsequent path planning and course correction.

[0031] For example, in preparation for a simulation training exercise, researchers first observed and recorded the flight paths of 100 homing pigeons in a mountainous area, focusing on their average turning angles and altitude adjustment rates when crossing canyons. They also retrieved data from 20 penetration missions conducted by drone swarms in similar terrain last year. Fifty magnetic field sensors were deployed at the training site to continuously monitor the changing geomagnetic field characteristics in the training area. This data was then compiled and stored in the training system database.

[0032] Step 102: Construct a three-dimensional path library based on the cluster trajectory data and the pigeon flock biological motion data.

[0033] In step 102, the three-dimensional path library represents a path reference database including spatial location information and terrain features.

[0034] In an embodiment of the present application, the collected historical trajectories of the drone are classified and organized according to the terrain features, including level flight segments, climbing segments, and diving segments; at the same time, the motion characteristics in the pigeon flock data are analyzed, and its characteristic parameters are matched and optimized with the corresponding types of historical trajectories; finally, the optimized path segments of various types are integrated into a standardized three-dimensional path library.

[0035] For example, the system categorizes historical trajectory data into three categories: sections between 2000 and 2200 meters above sea level are classified as canyon crossings, sections with high ascent rates are classified as climbing sections, and sections with high descent rates are classified as diving sections. It then optimizes turning angles in canyon sections based on the turning characteristics of the flock data, and optimizes altitude adjustment rates in climbing and diving sections based on the flock's altitude fluctuations. This ultimately creates a three-dimensional path library containing 30 optimized paths.

[0036] Step 103: Calculate the geomagnetic heading angle deviation according to the magnetic field gradient data.

[0037] In step 103 , the geomagnetic heading angle deviation represents the angular difference between the actual heading and the theoretical heading.

[0038] In an embodiment of the present application, the magnetic field strength component of the current measurement point is obtained from the magnetic field sensor; the theoretical magnetic field direction of the point is calculated based on a preset geomagnetic model; the directional difference between the measured value and the theoretical value is obtained through vector calculation; and finally, the difference is converted into an angular offset in a heading coordinate system.

[0039] For example, during a training session, the system measured the magnetic field strength components at the current point as 45 units in the X direction, -12 units in the Y direction, and 38 units in the Z direction. According to the model's theoretical values, these should be X48, Y-8, and Z40. The angular deviation, calculated using the formula θ=arccos[(A·B) / (|A||B|)], is 8.2 degrees, where A is the measured vector and B is the theoretical vector. This deviation is ultimately converted into a heading correction parameter.

[0040] Step 104: Based on the geomagnetic heading angle deviation and the three-dimensional path library, a multi-objective collaborative path is generated in combination with a swarm intelligence optimization model.

[0041] In step 104, the swarm intelligence optimization model refers to a computational model that simulates the collaborative behavior of biological swarms. Its physical meaning is reflected in the coordination of the flight paths of multiple drones through a distributed decision-making mechanism, enabling the swarm to autonomously adapt to environmental changes like a flock of birds, dynamically adjust its flight path, avoid collisions, maintain formation, and optimize overall penetration efficiency. This model does not rely on centralized control, but instead achieves global optimization through local interaction. The multi-objective collaborative path refers to the optimized flight path generated by the drone swarm in complex terrain environments to simultaneously meet multiple requirements such as heading deviation compensation, terrain avoidance, and formation coordination. This path is composed of multiple path segments with different tactical objectives. Each path segment corresponds to a specific flight strategy, such as altitude hold segments and sharp turns, ultimately forming a complete flight trajectory that balances safety and mission efficiency.

[0042] In an embodiment of the present application, a set of path points that matches the current terrain is selected from a path library; the heading coordinates of each point are corrected according to the geomagnetic deviation; key terrain change points in the path are identified and the avoidance strategy type is determined; the strategy type and the corrected path points are input into the optimization model; the path segments that meet the requirements of each strategy are output; and finally, the path segments are spliced ​​together to form a complete collaborative path.

[0043] For example, the system selects a set of path points near an altitude of 2,150 meters from the path library and corrects the heading according to a deviation of 8.2 degrees; it identifies three key points: the altitude mutation point, the starting point and the ending point of the turn; these correspond to the three strategies of altitude hold, turn preparation and turn execution respectively; after optimization, it generates three path segments: a horizontal flight segment, a turning segment and a recovery segment; and finally splices them into a complete path.

[0044] Step 105: Generate flight control instructions based on the multi-target collaborative path and the real-time status data of the UAV cluster.

[0045] In step 105, real-time state data is collected from real-time sensors and historical mission records in the simulated training environment. Specifically, it includes wind speed data (obtained in real time by meteorological sensors) and enemy aircraft position information (generated using the red and blue coordinate data of the simulated adversarial system). This data is used to describe the spatiotemporal state changes of the drone swarm in the dynamic adversarial environment. Flight control commands convert the planned path into specific control parameters for the drone's executable actions, including adjustments to the roll and angle of attack. Their physical meaning is to control flight direction and altitude by changing the drone's attitude angles (left and right bank angles and up and down pitch angles), ensuring the drone accurately tracks the planned path. Command content is dynamically generated based on the real-time calculated waypoint locations, time limits, and current flight status, and ultimately transmitted as digital signals to the drone's actuators.

[0046] In an embodiment of the present application, the next waypoint information in the collaborative path is parsed; the current flight status and threat information of the drone are obtained; the required roll angle and angle of attack adjustment amounts are comprehensively calculated; and the calculation results are converted into a control instruction format.

[0047] For example, the system resolves the coordinates of the next waypoint to be X degrees east longitude, Y degrees north latitude, and 2,153 meters above sea level; detects that the current drone attitude is a horizontal tilt angle of 3 degrees and a pitch angle of 2 degrees; calculates that it needs to roll left 15 degrees and pitch up 3 degrees; and generates corresponding binary control instructions.

[0048] Step 106: Load the flight control instructions into the simulation model of the UAV cluster, and drive the UAV cluster to perform coordinated penetration training in a simulated environment.

[0049] In step 106, the simulation model represents a mathematical model that simulates the physical characteristics and motion patterns of the UAV. The coordinated penetration training represents a mission exercise performed by multiple UAVs in coordination.

[0050] In an embodiment of the present application, control instructions are input into a simulation system; a swarm of virtual drones is driven to fly along a planned path; training effects are monitored in real time; and training data is recorded for subsequent optimization.

[0051] For example, the system sends control instructions to the simulation platform, and the virtual drone swarm begins to perform the mission, maintaining formation flight in the simulated canyon environment, successfully avoiding simulated threat targets, maintaining a stable heading throughout the entire process, and completing the scheduled penetration mission.

[0052] This method integrates biological motion characteristics with geomagnetic environmental data to construct an intelligent drone swarm training system. The system automatically generates flight paths adapted to complex terrain, accurately compensates for geomagnetic interference, and enables coordinated control of multiple drones. During simulated training, the system demonstrated excellent environmental adaptability and mission completion capabilities, enhancing training effectiveness.

[0053] To solve the collaborative path planning problem of drone swarms in complex terrain, in some embodiments, step 104: generating a multi-objective collaborative path based on the geomagnetic heading angle deviation and the three-dimensional path library in combination with a swarm intelligence optimization model includes: Step 201: Calling a path point set that matches the current terrain from the three-dimensional path library.

[0054] In step 201, the current terrain refers to the actual terrain environment characteristics of the drone cluster in real time. This is determined by presetting a digital elevation model of mountain terrain in the simulation training system and matching the corresponding terrain type data, including altitude, slope, and obstacle distribution characteristics, to the real-time positioning coordinates of the drone cluster. Areas with an altitude of 2000-2500 meters and a slope greater than 30 degrees are defined as the complex mountain terrain currently being processed. These parameters are derived from the three-dimensional terrain data of the training area provided by the geographic information system. A path point set is a collection of path points selected from a three-dimensional path library that match the terrain characteristics of the training area. These point sets contain location coordinates, altitude values, and terrain type tags, serving as a basic reference for subsequent path planning.

[0055] In an embodiment of the present application, the system first determines the terrain type based on the current position of the drone, and then retrieves historical path points stored under the same terrain category from the path library. These point sets have been classified and stored according to scenes such as canyons, climbs, and dives to ensure that the retrieved path points match the current environmental characteristics.

[0056] Step 202: Correct the heading coordinates of each path point in the path point set according to the geomagnetic heading angle deviation.

[0057] In step 202, the correction of the heading coordinates is the process of adjusting the direction angle of the waypoint according to the geomagnetic deviation. The corrected coordinates can offset the navigation error caused by geomagnetic interference.

[0058] In an embodiment of the present application, the system reads the current geomagnetic deviation angle and superimposes the angle value on the original direction angle of each path point, so that the heading of all path points is uniformly compensated for the influence of geomagnetic interference, ensuring that the subsequent planned path maintains the correct direction during actual flight.

[0059] Step 203: Identify terrain feature change points in the corrected path point set, and determine an avoidance type identifier based on the terrain feature change points.

[0060] In step 203, the terrain feature change point refers to a special point in the path point set that identifies a sudden change in terrain. The avoidance type identifier is a classification label for the response strategy defined based on the sudden change feature.

[0061] In an embodiment of the present application, the system scans the corrected path point set, automatically identifies key location points such as sudden altitude changes and the start of turns, and assigns a type identifier to each feature point, such as "altitude hold" or "sharp turn". These identifiers will guide the selection of subsequent path optimization strategies.

[0062] Step 204: Input the avoidance type identifier and the corrected path point set into a swarm intelligence optimization model, and output a terrain avoidance constraint path segment corresponding to the avoidance type identifier.

[0063] In step 204, a terrain avoidance-constrained path segment is an optimized path segment that meets specific avoidance requirements, and its geometric characteristics strictly correspond to the avoidance type identifier. There are multiple terrain avoidance-constrained path segments, and each avoidance type identifier generates an independent path segment. Ultimately, all generated path segments are concatenated to form a complete multi-target coordinated path. Specifically, each avoidance type identifier corresponds to a set of consecutive path points (rather than a single path point), and each group of path points constitutes an independent path segment. Each path segment has different geometric characteristics and constraints depending on the corresponding sub-target type (such as high ground capture or obstacle avoidance).

[0064] In an embodiment of the present application, the system inputs the type identifier and the path point set into the optimization model. The model first matches the corresponding constraint conditions for each identifier, and then adjusts the position of the path point to satisfy the constraints. For example, the sharp turn path segment must ensure that the turning radius meets the standard. Finally, it outputs several segments of optimized paths that meet the requirements.

[0065] Step 205: Splice all the terrain avoidance constraint path segments to form a multi-target collaborative path.

[0066] In an embodiment of the present application, the system connects the horizontal flight segment, the turning segment, and the recovery segment end to end in the order from low altitude to high altitude to form a coherent flight path that simultaneously meets the requirements of heading stability, terrain avoidance, and mission timing.

[0067] Here's a specific example: During a drone swarm mountain penetration training mission, the system first retrieved a set of path points that matched the current canyon terrain from the established three-dimensional path library. This set of points contained 30 path points near an altitude of 2,150 meters, which were derived from historical trajectory data optimized in the early stages. The system then applied the previously calculated 8.2-degree geomagnetic heading deviation to these path points, specifically increasing the heading angle of each point by 8.2 degrees to compensate for the influence of geomagnetic interference. After the correction was completed, the system automatically scanned the path point set and identified three key terrain feature change points: the first was the starting point where the altitude suddenly increased from 2,150 meters to 2,155 meters, marked as the altitude hold mark; the second was the starting point of the sharp turn, marked as the turn preparation mark; and the third was the end point of the turn, marked as the turn execution mark. After inputting these markers and the corrected pathpoints into the swarm intelligence optimization model, the model first matched the altitude hold marker with a vertical fluctuation constraint of no more than 5 meters and the two turn markers with a turning radius constraint of greater than 50 meters. It then iteratively selected and reordered pathpoints that met these constraints. The turning radius of the turn segments was calculated using the formula R=v² / (g×tanφ), where R is the turning radius, v is the preset flight speed of 25 meters per second, g is the acceleration due to gravity, and φ is the roll angle. The model ultimately outputs three optimized paths: the first segment maintains a horizontal flight of 60 meters at an altitude of 2150-2155 meters, the second segment completes a left turn with a radius of 52 meters, and the third segment resumes flight at an altitude of 2150 meters. The system then stitches these three paths together, in mission order, into a complete multi-target coordinated path that compensates for geomagnetic deviation and meets terrain avoidance requirements, providing a precise route for subsequent flight control command generation.

[0068] In the embodiment of the present application, this method intelligently integrates geomagnetic correction and terrain feature recognition to ensure that the generated collaborative path not only ensures heading accuracy but also has good terrain adaptability, effectively improving the penetration capability and collaborative efficiency of the drone cluster in complex environments, while reducing the need for manual intervention.

[0069] To solve the path optimization problem of a drone swarm in complex terrain, in some embodiments, step 204: inputting the avoidance type identifier and the modified path point set into a swarm intelligence optimization model, and outputting a terrain avoidance constraint path segment corresponding to the avoidance type identifier, includes: Step 301: Input the avoidance type identifier and the corrected path point set into the swarm intelligence optimization model. Based on the avoidance type identifier, call the corresponding terrain space constraint relationship from the preset constraint rule library. The terrain space constraint relationship includes an obstacle spacing threshold and a height safety margin.

[0070] In step 301, the terrain spatial constraint relationship refers to the spatial restriction conditions set for different avoidance types, where the obstacle spacing threshold specifies the minimum safe distance between the UAV and the obstacle, and the altitude safety margin limits the adjustable range of the flight altitude.

[0071] In an embodiment of the present application, the system first reads the avoidance type identification, and retrieves the corresponding constraint conditions from the rule library according to the identification type. For example, the sharp turn type corresponds to a larger spacing threshold, and the altitude maintenance type corresponds to a strict altitude margin, to ensure that subsequent path planning meets the special requirements of various terrains.

[0072] Step 302: Perform path feasibility verification on the corrected path point set according to the terrain spatial constraint relationship, and eliminate conflicting path points that are smaller than the obstacle spacing threshold or smaller than the height safety margin.

[0073] In step 302, conflicting path points refer to path points that violate terrain spatial constraints. Path feasibility verification is a process of determining whether each path point satisfies the constraints through geometric calculations.

[0074] In an embodiment of the present application, the system checks the corrected path points one by one, calculates the distance and height deviation between each point and the nearest obstacle, marks the points that do not meet the spacing or height requirements as conflict points and eliminates them, and retains the qualified point set for subsequent optimization.

[0075] Step 303: Based on the degree of association between the avoidance type identifier and the task, a corresponding first priority value is assigned to the avoidance type identifier.

[0076] In step 303, the degree of correlation is determined by analyzing the importance of the training mission objectives and the various terrain avoidance requirements. This refers to the weight of influence that different avoidance types have on mission completion, and is pre-determined by the commander based on the mission's characteristics. For example, in a penetration mission, if maintaining altitude is crucial for concealment, the correlation is high, while a conventional turn has a relatively small impact on the mission, the correlation is low. The system quantifies these correlations into priority values ​​for path optimization. The first priority value is a quantitative indicator reflecting the importance of the avoidance type, and its value depends on its criticality to mission completion.

[0077] In an embodiment of the present application, the system assigns a priority value to each avoidance type based on mission requirements. For example, altitude hold types related to flight safety are given a higher priority, while conventional turn types are given a relatively lower priority. These values ​​will guide the sorting and reorganization of subsequent path points.

[0078] Step 304: Based on the first priority value, the non-conflicting path points in the verified path point set are reorganized in sequence to generate a terrain avoidance constraint path segment.

[0079] In step 304, sequence reorganization refers to the process of reordering the waypoints according to priority.

[0080] In an embodiment of the present application, the system sorts the retained path points according to priority values, high-priority points are arranged at key positions, and then the ordered points are connected into smooth path segments through a curve fitting method to ensure that each path segment meets both terrain constraints and task priority requirements.

[0081] Here's a specific example: During the execution of a certain mountain penetration training mission, the system conducted in-depth optimization processing on the three identified avoidance type identifiers. First, according to the mission safety requirements, the altitude hold identifier was assigned a priority value of 0.9, because it is directly related to the concealed penetration effect. Priority values ​​of 0.7 and 0.6 were assigned to the turn preparation and turn execution identifiers respectively. These values ​​were determined by the commander based on the fact that high concealment is more important than maneuverability in this penetration mission; then the corresponding terrain space constraint relationship was retrieved from the constraint rule library, among which the altitude hold class matched the vertical fluctuation limit of no more than 5 meters, which was derived from the minimum safe flight altitude standard for drones, and the turn class matched the obstacle spacing threshold of 50 meters and the lower limit of the turning radius of 52 meters; then the system verified the 30 corrected path points using the distance formula Calculate the distance between each point and the simulated mountain, where x, y, and z are coordinate values, eliminate three dangerous points with a spacing of less than 50 meters, and check the altitude value and eliminate two points outside the range of 2150-2155 meters; the remaining 25 non-conflicting path points are re-sorted according to priority, with altitude key points first and turning points second, and generate three optimized paths through curve fitting method: the first segment flies 40 meters in a straight line at an altitude of 2150-2152 meters, and the second segment completes the turn with a radius of 53 meters, which is passed through Calculate to ensure safety, where v is 25 m / s and g is , φ is the 25-degree roll angle recommended by the system. The third segment smoothly descends to an altitude of 2,150 meters and continues to fly. The final terrain avoidance constraint path segment not only meets various safety constraints but also reflects the mission priority, providing a reliable route for subsequent coordinated breakthroughs.

[0082] In the embodiment of the present application, the method uses intelligent constraint verification and priority sorting to ensure that the generated path segments not only ensure flight safety but also highlight the key points of the mission, effectively improving the adaptability and mission completion quality of the drone cluster in complex terrain.

[0083] To further improve the accuracy and adaptability of drone swarm path planning, in some embodiments, step 304: sequentially reorganizing the non-conflicting path points in the verified path point set based on the first priority value to generate a terrain avoidance constraint path segment includes: Step 401: Based on a preset mapping table between a first priority value range and urgency, determine an urgency identifier corresponding to the first priority value.

[0084] In step 401, the urgency identifier refers to a level label divided according to the urgency of the task, and the priority value is converted into three urgency levels: high, medium, and low through a pre-set value range mapping table.

[0085] In an embodiment of the present application, the system reads the first priority value of each avoidance type, and determines the corresponding urgency identifier by looking up the table. For example, priority 0.9 is mapped to high urgency, 0.7 is mapped to medium urgency, and 0.6 is mapped to low urgency. These identifiers will guide the sorting strategy of subsequent path points.

[0086] Step 402: According to the first priority value and in combination with the terrain adaptability score, a corresponding second priority value is assigned to each non-conflicting path point.

[0087] In step 402, the terrain adaptability score is calculated by quantitatively analyzing the degree of match between the waypoint and terrain features. Specifically, it considers factors such as the distance from the nearest obstacle, altitude stability, and deviation from the ideal flight path. These metrics are combined into a single score using a weighted calculation method. Waypoints with greater distance from obstacles, greater altitude stability, and smaller deviations receive higher scores. The second priority value is the final priority calculated by combining the first priority and the terrain adaptability.

[0088] In an embodiment of the present application, the system calculates a terrain adaptability score for each non-conflicting path point, taking into account factors such as the distance between the point and the obstacle, altitude stability, etc., and then weights the score with the first priority to obtain a second priority value to ensure that important path points meet the terrain adaptability requirements at the same time.

[0089] Step 403: Sort the second priority values ​​of all non-conflicting waypoints according to the sorting rules corresponding to the urgency identifiers to generate an ordered waypoint sequence.

[0090] In step 403, the sorting rule refers to the arrangement of pathpoints corresponding to different urgency levels: high urgency is sorted in descending order, and medium and low urgency are sorted in ascending order. Within the sorting rule corresponding to the urgency identifier, the sorting of different urgency identifiers is performed independently. The pathpoint sequence corresponding to the high urgency identifier [c, b, a] and the sequence corresponding to the low urgency identifier [d, e] are not mixed. Instead, they are first sorted by urgency level (within the high urgency group, sorted in descending order by [c, b, a], and within the low urgency group, sorted in ascending order by [d, e]). The complete sorting results of the high urgency group are then output as a whole, prioritizing the low urgency group, ultimately forming an overall sequence of [c, b, a, d, e]. An ordered pathpoint sequence refers to a set of pathpoints rearranged according to the urgency identifier and priority rules, with high-urgency, high-priority points placed first and low-urgency, low-priority points placed last, forming a pathpoint queue arranged from important to less important points that meets the task priority requirements.

[0091] In an embodiment of the present application, the system groups the path points according to the urgency identifier, the high urgency group is arranged from high to low according to the second priority, the medium and low urgency groups are arranged from low to high, and finally the high urgency group is placed at the front of the sequence as a whole, forming an ordered sequence that takes into account both urgency and terrain adaptability.

[0092] Step 404: performing smooth interpolation processing on adjacent non-conflicting path points in the ordered path point sequence to form a continuous flight trajectory.

[0093] In step 404, adjacent non-conflicting pathpoints are defined as two non-conflicting pathpoints that are consecutively ranked within the ordered pathpoint sequence. Their adjacency is determined by the order of arrangement within the sequence and is unrelated to the original spatial distribution. Smooth interpolation is the process of adding intermediate points between adjacent pathpoints to create a continuous trajectory. A continuous flight trajectory is a coherent flight path generated by smooth interpolation of an ordered pathpoint sequence. By adding calculated points between the original measurement points, sudden changes in the path are eliminated, ensuring that the drone can fly smoothly and steadily along the trajectory, avoiding dangerous maneuvers such as sharp turns or sudden altitude changes.

[0094] In an embodiment of the present application, the system analyzes the sorted path point sequence, inserts several calculation points between every two actual measurement points, and uses a curve fitting algorithm to ensure that the generated trajectory turns smoothly and the flight is continuous, avoiding sharp turns or sudden changes in altitude.

[0095] Step 405: Applying terrain constraints corresponding to the avoidance type identifier to the continuous flight trajectory to generate a terrain avoidance constraint path segment.

[0096] In step 405 , the terrain spatial constraint relationship is a general constraint rule (such as a spacing threshold) called from the constraint rule library, while the terrain constraint condition is an instance of applying the relationship specifically to the currently generated continuous flight trajectory. The two are the relationship between rules and implementation.

[0097] In an embodiment of the present application, the system checks whether the interpolated continuous trajectory meets the constraints of various avoidance signs throughout the entire process, and makes fine adjustments to local non-compliant areas, such as adjusting the turning radius or altitude change rate, until a fully compliant terrain avoidance constraint path segment is generated.

[0098] Here's a specific example: In a certain UAV swarm mountain penetration training mission, the system performed in-depth optimization on 25 verified non-conflict path points. First, according to the preset mapping table, the 0.9 priority of the altitude hold indicator was mapped to high urgency, the 0.7 priority of the turn preparation was mapped to medium urgency, and the 0.6 priority of the turn execution was mapped to low urgency. Then, the terrain adaptability score was calculated for each path point. The scoring formula is: , where d represents the actual distance between the point and the nearest obstacle, D is the safety threshold of 50 meters, h represents the height fluctuation value, and H is the allowable fluctuation range of 5 meters. and The weight coefficients are 0.6 and 0.4 respectively. The score of each point is calculated to be between 0.5 and 0.9 by this formula. Then, the first priority value and the terrain fitness score are combined to use the weighted formula Calculate the second priority value for each point, where is the first priority, S is the terrain adaptability score; according to the sorting rules corresponding to the urgency identifier, the path points of the high urgency group are arranged in descending order of the second priority, and the medium and low urgency groups are arranged in ascending order, finally forming an ordered sequence; when smoothing the sequence, two intermediate points are inserted between every two adjacent points, and the cubic spline interpolation algorithm is used to ensure the continuity of the trajectory, where the path points of the key turning section are calculated by the formula Verify the turning radius, where v is maintained at 25 m / s and g is , φ is set to 25 degrees to ensure that the radius is no less than 52 meters; finally, the entire trajectory is constrained and reviewed, and the positions of the three points are adjusted to keep the entire altitude within the range of 2150-2155 meters. Three complete paths are output: the first segment is a straight flight of 45 meters at an altitude of 2151-2153 meters, the second segment is a smooth turn with a radius of 53 meters, and the third segment is a steady descent to 2150 meters, forming a terrain avoidance constraint path segment that meets all the constraints.

[0099] In the embodiment of the present application, the method uses multi-level priority division and terrain adaptability optimization to enable the generated path segments to highlight the mission focus while fully considering the terrain characteristics, achieving the optimal balance between safety and mission completion, and improving the combat effectiveness of drone clusters in complex environments.

[0100] In order to construct a UAV path planning database that adapts to complex terrain, in some embodiments, step 102: constructing a three-dimensional path library based on the cluster trajectory data and the pigeon flock biological motion data includes: Step 501: Classify the cluster trajectory data into canyon crossing trajectory, climbing trajectory and diving trajectory according to terrain type.

[0101] In step 501, terrain classification involves dividing the flight trajectory into different segments based on altitude change rate and spatial characteristics. A canyon crossing segment refers to a flight segment that maintains a relatively stable altitude within a confined space. A climbing segment refers to a flight segment that continuously increases in altitude. A diving segment refers to a flight segment that continuously decreases in altitude.

[0102] In an embodiment of the present application, the system analyzes the altitude change curve and spatial position distribution in the historical trajectory data, and automatically identifies three types of flight segments with obvious characteristics. Among them, the area with gentle altitude change and obstacles on both sides is classified as a canyon crossing segment, the segment with continuous altitude increase is classified as a climbing segment, and the segment with continuous altitude decrease is classified as a diving segment, thereby establishing a basic classification framework for subsequent feature mapping.

[0103] Step 502: Extract flight turning features and altitude change features from the pigeon flock biological motion data.

[0104] In step 502, flight turning characteristics refer to the movement characteristics of the pigeon flock when changing direction in complex terrain, including turning angle and turning rate. Altitude change characteristics refer to the movement characteristics of the pigeon flock when adjusting flight altitude, including climbing rate and diving rate.

[0105] In an embodiment of the present application, the system processes observation data of a flock of pigeons, extracts the typical turning angle range and change speed of the flock when crossing obstacles as turning features, and simultaneously calculates the average change rate of the flock when adjusting its altitude as a height feature. These biological motion features provide a natural reference for drone path optimization.

[0106] Step 503: Map the flight turning feature to the canyon crossing segment trajectory, and map the altitude change feature to the climbing segment trajectory and the diving segment trajectory.

[0107] In step 503, feature mapping is the process of matching and optimizing biological motion characteristic parameters with drone trajectory segments. By adjusting drone trajectory parameters, the trajectory segments are aligned with the motion characteristics of biological groups. The specific implementation process for mapping the altitude change characteristics to the climbing and swooping trajectories is as follows: first, extract two characteristic values ​​of typical altitude change rates of 2 meters per second for ascent and 1.8 meters per second for descent from the pigeon flock data. Then, forward mapping is performed on the climbing trajectories to uniformly adjust all ascent rates in the original trajectory to 2 meters per second using the formula "adjusted rate = original rate × 2 / average pigeon flock ascent rate 1.5." Simultaneously, reverse mapping is performed on the swooping trajectories to uniformly adjust the original descent rates to 1.8 meters per second using the formula "adjusted rate = original rate × 1.8 / average pigeon flock descent rate 2.2." For example, the rate of a certain original climbing trajectory is adjusted from 1.2 meters per second to 1.6 meters per second, and the rate of the original swooping trajectory is adjusted from 2.5 meters per second to 2.05 meters per second. Ultimately, the two trajectories are matched to the different altitude change characteristics of the pigeon flocks.

[0108] In an embodiment of the present application, the system applies the typical turning angle of a flock of pigeons to the turning portion of the canyon crossing trajectory, making the UAV's turning action closer to the smooth characteristics of the pigeon flock; at the same time, the climbing and diving rate parameters of the pigeon flock are applied to the corresponding UAV trajectory segment, optimizing the height adjustment process, so that the UAV's height change is more in line with natural flight laws.

[0109] Step 504: Based on the mapping result, corresponding terrain-adaptive path units are generated, and all the terrain-adaptive path units are aggregated to form a three-dimensional path library.

[0110] In step 504, the mapping result refers to the output of matching the pigeon flock's flight characteristics with the drone's historical trajectory classification segments. It includes matching parameters (such as turning angle and altitude change rate) for each terrain classification segment (canyon crossing, climbing, and swooping) and the corresponding pigeon flock characteristics. Specifically, it takes the form of parameterized key-value pairs, for example, {"canyon crossing": turning angle ±30°}, {"climbing": altitude change rate 2m / s}. Three mapping results are generated for each terrain classification segment, each corresponding to a path unit (e.g., a canyon path unit or a climbing path unit). Ultimately, all these units are aggregated to form a three-dimensional path library. Terrain-adapted path units are standardized flight path segments optimized using biometric features.

[0111] In an embodiment of the present application, the system encapsulates the optimized trajectory segments into independent units, each unit containing information such as position coordinates, motion parameters, and applicable terrain. All units are then organized into a structured database according to terrain type and spatial position, providing a modular reference for subsequent path planning.

[0112] Here's a specific example: During the construction of a certain drone swarm training system, the system first analyzed and processed the collected pigeon flight data, and extracted from the flight trajectories of 100 homing pigeons that the average turning angle when crossing the canyon was 35 degrees, the average ascent rate in the climbing phase was 2 meters per second, and the average descent rate in the diving phase was 1.8 meters per second. These values ​​were obtained by taking the arithmetic mean after counting the pigeon flight data; at the same time, the system retrieved 20 historical mission data performed by the drone swarm in similar terrain, and classified 120 trajectories located in the altitude range of 2000-2200 meters and with altitude fluctuations of less than 3 meters as canyon crossing segments, 80 segments with an ascent rate greater than 1.5 meters per second as climbing segments, and 60 segments with a descent rate greater than 2 meters per second as diving segments; then the system mapped the 35-degree turning feature of the pigeon flock to the canyon crossing segment trajectory, using the angle adjustment formula θ_new=θ_o ld+(35-θ_avg) is used to optimize each turning point, where θ_old is the original turning angle and θ_avg is the average turning angle of the historical trajectory of 30 degrees, so that the turning angle of the optimized canyon section is close to 35 degrees; at the same time, the rising and falling rate characteristics of the pigeon flock are mapped to the corresponding type of trajectory segment, and the height change rate of each segment is optimized through the rate adjustment formula v_new=v_old×(v_pigeon / v_avg), where v_pigeon is the pigeon flock rate and v_avg is the average rate of the historical trajectory, so that the climbing segment rate approaches 2 meters per second and the diving segment rate approaches 1.8 meters per second; finally, the system generates 150 optimized path units, including 60 canyon crossing units, 50 climbing units and 40 diving units. Each unit is marked with the applicable altitude range and motion parameters, which are integrated to form a three-dimensional path library containing 30 typical paths.

[0113] In the embodiment of the present application, the method integrates biological swarm intelligence and historical experience data of drones to construct a three-dimensional path library that not only retains the effective experience of actual drone missions, but also incorporates the optimization characteristics of natural flight, thereby improving the natural adaptability and environmental matching of path planning, and providing reliable path support for drone cluster collaborative tasks in complex terrain.

[0114] In order to further improve the heading accuracy of the UAV in a complex geomagnetic environment, in some embodiments, step 103: calculating the geomagnetic heading angle deviation based on the magnetic field gradient data includes: Step 601: extracting the magnetic field intensity component of the current measurement point from the magnetic field gradient data.

[0115] In step 601, the current measurement point is the location of the drone swarm during real-time flight in a simulated training environment. The geomagnetic sensing module installed on the drones measures the magnetic field data at this location in real time. The magnetic field intensity components are the measured values ​​of the magnetic field intensity at a point in space in three mutually perpendicular directions, reflecting the spatial distribution characteristics of the magnetic field at that point.

[0116] In an embodiment of the present application, the system reads the raw data collected by the geomagnetic sensor at the current measurement location and extracts the magnetic field strength values ​​in three directions. These data reflect the actual magnetic field conditions at that location affected by the terrain and environment.

[0117] Step 602: Determine the theoretical magnetic field direction of the current measurement point according to a preset geomagnetic reference field model.

[0118] In step 602, the theoretical magnetic field direction refers to the magnetic field direction vector that should exist at the geographical location calculated according to the standard geomagnetic model, and is used as a heading reference.

[0119] In an embodiment of the present application, the system queries a preset geomagnetic reference field model database based on the latitude and longitude coordinates of the current measurement point to obtain the theoretical magnetic field direction data of the location for comparison and analysis with the actual measurement value.

[0120] Step 603: Calculate the direction deviation between the magnetic field intensity component and the theoretical magnetic field direction.

[0121] In step 603 , the direction deviation refers to the spatial angle difference between the measured magnetic field direction and the theoretical magnetic field direction.

[0122] In an embodiment of the present application, the system forms a space vector with the three measured directional magnetic field components, performs spatial angle calculation with the theoretical magnetic field direction vector, and obtains the degree of directional deviation between the two.

[0123] Step 604: Convert the direction deviation into an angular offset in a heading coordinate system, and use the angular offset as the geomagnetic heading angle deviation.

[0124] In step 604, the heading coordinate system refers to a local coordinate system established based on the drone's current heading. This system originates from a three-dimensional rectangular coordinate system preset in the drone's navigation system, with the drone's longitudinal axis as the X-axis, the horizontal axis perpendicular to the longitudinal axis as the Y-axis, and the vertical axis pointing downward as the Z-axis. This coordinate system dynamically adjusts as the drone's heading changes. It is used to convert geomagnetic deviations into angular offsets, which are directly related to drone flight control. The angular offset is the conversion of spatial directional deviations into a heading correction value in this coordinate system.

[0125] In an embodiment of the present application, the system converts the calculated directional deviation into the body coordinate system based on the current posture of the drone, and outputs an angle correction value that can be directly used for heading control.

[0126] Here's a specific example: During a drone swarm mountain penetration training mission, when the swarm reached a canyon at an altitude of 2,150 meters, the system initiated the geomagnetic heading deviation calculation process. First, the onboard geomagnetic sensor acquired the magnetic field strength components at the current location, measuring 45 units in the X direction, -12 units in the Y direction, and 38 units in the Z direction. These values ​​were directly measured by the sensor. The system then queried the theoretical magnetic field direction from a preset geomagnetic reference field model based on the location's geographic coordinates, determining the standard values ​​to be 48 units in the X direction, -8 units in the Y direction, and 40 units in the Z direction. The system then calculated the directional deviation between the measured and theoretical values ​​using the vector angle formula θ = arccos[(A·B) / (|A||B|)], where A represents the measured vector (45, -12, 38) and B represents the theoretical vector (48, -8, 40). This resulted in a spatial directional deviation of 8.2 degrees. The system then combines the drone's current horizontal tilt angle of 3 degrees and pitch angle of 2 degrees attitude data, converts the spatial deviation into the body coordinate system, and finally outputs a clockwise heading correction value of 8.2 degrees.

[0127] In the embodiment of the present application, the method realizes real-time and accurate correction of the UAV heading in a complex electromagnetic environment by accurately measuring and calculating the geomagnetic deviation, effectively overcomes the navigation error caused by terrain interference, and provides a reliable heading reference for the coordinated flight of UAV clusters in strong interference areas.

[0128] To further improve the collaborative control accuracy of the UAV cluster, in some embodiments, step 105: generating flight control instructions based on the multi-target collaborative path and in combination with the real-time status data of the UAV cluster, includes: Step 701: Extract the flight posture of each UAV and the relative position of the enemy aircraft from the real-time status data.

[0129] In step 701, the flight attitude refers to the current pitch, roll, and yaw angles of the UAV. The relative position of the enemy aircraft refers to the direction and distance of the enemy target relative to the UAV.

[0130] In an embodiment of the present application, the system obtains real-time flight attitude data through the attitude sensors of each drone, and at the same time detects the direction and distance information of surrounding enemy aircraft through a simulated confrontation system, providing an environmental perception basis for subsequent control command generation.

[0131] Step 702: Analyze the coordinates and arrival time limit of each UAV at the next navigation point from the multi-target cooperative path.

[0132] In step 702, the next navigation point refers to the next key path point that the drone cluster will fly to based on the multi-target collaborative path, including its coordinate location and arrival time requirement. The "current measurement point" refers to the drone's location when the geomagnetic sensing module collects magnetic field data in real time. The two are related in that the current measurement point is the reference position for calculating the geomagnetic heading angle deviation, while the next navigation point is the target destination location planned based on the corrected path based on this deviation. The next navigation point is generated only after the current measurement point is determined, and the two constitute a continuous sequence of navigation positions. The arrival time limit refers to the time required to complete the flight segment.

[0133] In an embodiment of the present application, the system parses the collaborative path data and extracts the target point location and specified arrival time that each drone needs to reach in the next stage. These parameters are derived from the previous path planning results to ensure the synchronization and coordination of cluster flight.

[0134] Step 703: For each UAV, the coordinates, the arrival time limit, the flight attitude, and the relative position of the enemy aircraft are integrated to calculate the roll angle control parameters and the angle of attack control parameters.

[0135] In step 703, the roll angle control parameter refers to the angle value for controlling the left and right tilt of the drone. The attack angle control parameter refers to the angle value for controlling the up and down pitch of the drone.

[0136] In the embodiment of the present application, the system comprehensively considers factors such as the target point location, remaining time, current posture and enemy aircraft threat for each drone, and calculates the optimal roll and pitch adjustment amount through the kinematic model to ensure that the drone can reach the target point on time and safely.

[0137] Step 704: Encapsulate the roll angle control parameters and the attack angle control parameters of all UAVs into flight control instructions.

[0138] In an embodiment of the present application, the system encodes and packages the calculated control parameters of each drone according to a predetermined protocol format, generates a digital instruction set that can be received and executed by the drone, and realizes the coordinated control of the cluster.

[0139] Here's a specific example: In a certain UAV cluster canyon penetration training, when the cluster flew to the key area along the planned path, the system began to generate flight control instructions. First, the real-time status of the three UAVs was obtained: the current attitude of the No. 1 drone was a horizontal tilt angle of 3 degrees and a pitch angle of 2 degrees, and the enemy aircraft was detected to be 300 meters northeast of it; the attitude of the No. 2 drone was a horizontal tilt angle of 1 degree and a pitch angle of 1 degree, and the enemy aircraft was 400 meters away; the attitude of the No. 3 drone was a horizontal tilt angle of 0 degrees and a pitch angle of 3 degrees, and the enemy aircraft was 350 meters away. The system parsed from the multi-target collaborative path generated in the early stage that the next waypoint of each aircraft was X degrees east longitude, Y degrees north latitude, and 2153 meters above sea level, and the arrival time limit was unified as 30 seconds. For the No. 1 drone, the system comprehensively considered its current attitude, enemy aircraft distance and arrival time limit, and used the formula The calculation shows that it needs to roll left 18 degrees, where v is the current speed of 25 meters per second and g is the acceleration due to gravity. , R is the path turning radius of 53 meters, K is the adjustment coefficient of 0.6, and θ is the enemy aircraft threat weight angle of 10 degrees. The angle of attack is also calculated to be 4 degrees. Since the enemy aircraft is farther away, aircraft #2 only requires basic adjustments, calculated to roll 12 degrees and angle of attack 3 degrees; aircraft #3 rolls 15 degrees and angle of attack 5 degrees. The system converts these parameters into binary commands. Aircraft #1 is instructed to focus on evasive maneuvers, aircraft #2 maintains standard flight, and aircraft #3 makes appropriate adjustments.

[0140] In the embodiment of the present application, the method generates accurate cluster control instructions by integrating path planning and environmental perception data in real time, thereby realizing the coordinated maneuvering and precise control of the drone swarm in complex environments, and effectively improving the success rate and safety of the penetration mission.

[0141] Figure 2 A schematic diagram of the structure of a multi-agent collaborative drone swarm simulation training system provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the system includes: The acquisition module 21 is used to collect the biological movement data of the pigeon flock, the cluster trajectory data of the drone cluster when performing historical penetration missions, and the magnetic field gradient data of the simulated environment.

[0142] The construction module 22 is used to construct a three-dimensional path library based on the cluster trajectory data and the pigeon flock biological movement data.

[0143] The calculation module 23 is used to calculate the geomagnetic heading angle deviation according to the magnetic field gradient data.

[0144] The first generating module 24 is configured to generate a multi-target collaborative path based on the geomagnetic heading angle deviation and the three-dimensional path library in combination with a swarm intelligence optimization model.

[0145] The second generation module 25 is used to generate flight control instructions based on the multi-target collaborative path and the real-time status data of the UAV cluster.

[0146] The loading module 26 is used to load the flight control instructions into the simulation model of the UAV cluster, driving the UAV cluster to perform coordinated penetration training in a simulated environment.

[0147] Figure 2 The multi-agent collaborative UAV swarm simulation training system can be executed Figure 1 The implementation principle and technical effects of the multi-agent collaborative drone swarm simulation training method described in the illustrated embodiment will not be elaborated on here. The specific manner in which each module and unit performs operations in the multi-agent collaborative drone swarm simulation training system in the above embodiment has been described in detail in the embodiment of the method and will not be elaborated on here.

[0148] In one possible design, Figure 2 The multi-agent collaborative drone swarm simulation training system of the embodiment shown 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; The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0149] The processing component 32 is used to perform the above Figure 1 The embodiment provides a multi-agent collaborative drone swarm simulation training method.

[0150] 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 method. Of course, the processing component may also 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 method.

[0151] The storage component 31 is configured to store various types of data to support operations on the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0152] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0153] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0154] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0155] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0156] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 A multi-agent collaborative drone swarm simulation training method according to the illustrated embodiment.

[0157] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0158] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0159] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion 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, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0160] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-agent collaborative drone swarm simulation training method, characterized by: include: Collect biological movement data of pigeon flocks, cluster trajectory data of drone swarms when performing historical penetration missions, and magnetic field gradient data of simulated environments; Constructing a three-dimensional path library based on the cluster trajectory data and the pigeon flock biological motion data; Calculating the geomagnetic heading angle deviation according to the magnetic field gradient data; Based on the geomagnetic heading angle deviation and the three-dimensional path library, a multi-objective collaborative path is generated in combination with a swarm intelligence optimization model; Based on the multi-target collaborative path and combined with the real-time status data of the UAV cluster, a flight control instruction is generated; The flight control instructions are loaded into a simulation model of the UAV cluster to drive the UAV cluster to perform coordinated penetration training in a simulated environment.

2. The multi-agent collaborative drone swarm simulation training method according to claim 1 is characterized in that: The generating of a multi-objective collaborative path based on the geomagnetic heading angle deviation and the three-dimensional path library in combination with a swarm intelligence optimization model includes: Retrieving a set of path points matching the current terrain from the three-dimensional path library; Correcting the heading coordinates of each path point in the path point set according to the geomagnetic heading angle deviation; identifying terrain feature change points in the corrected path point set, and determining an avoidance type identifier based on the terrain feature change points; Inputting the avoidance type identifier and the corrected path point set into a swarm intelligence optimization model, and outputting a terrain avoidance constraint path segment corresponding to the avoidance type identifier; All the terrain avoidance constraint path segments are spliced ​​together to form a multi-target collaborative path.

3. The multi-agent collaborative drone swarm simulation training method according to claim 2 is characterized in that: The step of inputting the avoidance type identifier and the corrected path point set into a swarm intelligence optimization model and outputting a terrain avoidance constraint path segment corresponding to the avoidance type identifier comprises: Inputting the avoidance type identifier and the corrected path point set into a swarm intelligence optimization model, and calling a corresponding terrain spatial constraint relationship from a preset constraint rule library based on the avoidance type identifier, wherein the terrain spatial constraint relationship includes an obstacle spacing threshold and a height safety margin; Performing a path feasibility check on the corrected path point set according to the terrain spatial constraint relationship, and eliminating conflicting path points that are smaller than the obstacle spacing threshold or smaller than the height safety margin; assigning a corresponding first priority value to the avoidance type identifier based on a degree of association between the avoidance type identifier and the task; Based on the first priority value, the non-conflicting path points in the verified path point set are reorganized in sequence to generate a terrain avoidance constraint path segment.

4. The multi-agent collaborative drone swarm simulation training method according to claim 3 is characterized in that: The step of sequentially reorganizing the non-conflicting path points in the verified path point set based on the first priority value to generate a terrain avoidance constraint path segment includes: Determining an urgency level identifier corresponding to the first priority level value based on a preset mapping table between a first priority level value range and an urgency level; According to the first priority value and in combination with the terrain adaptability score, a corresponding second priority value is assigned to each non-conflicting path point; Sort the second priority values ​​of all non-conflicting waypoints according to the sorting rules corresponding to the urgency identifiers to generate an ordered waypoint sequence; Performing smooth interpolation processing on adjacent non-conflicting path points in the ordered path point sequence to form a continuous flight trajectory; A terrain constraint condition corresponding to the avoidance type identifier is applied to the continuous flight trajectory to generate a terrain avoidance constraint path segment.

5. The multi-agent collaborative drone swarm simulation training method according to claim 1 is characterized in that: The method of constructing a three-dimensional path library based on the cluster trajectory data and the pigeon flock biological motion data includes: Classifying the cluster trajectory data into canyon crossing trajectory, climbing trajectory and diving trajectory according to terrain type; Extracting flight turning features and altitude change features from the pigeon flock biological motion data; Mapping the flight turning feature to the canyon crossing segment trajectory, and mapping the altitude change feature to the climbing segment trajectory and the diving segment trajectory; Based on the mapping results, corresponding terrain-adaptive path units are generated, and all the terrain-adaptive path units are aggregated to form a three-dimensional path library.

6. The multi-agent collaborative drone swarm simulation training method according to claim 1 is characterized in that: Calculating the geomagnetic heading angle deviation according to the magnetic field gradient data includes: extracting the magnetic field intensity component of the current measurement point from the magnetic field gradient data; Determining the theoretical magnetic field direction of the current measurement point according to a preset geomagnetic reference field model; Calculating a direction deviation between the magnetic field intensity component and the theoretical magnetic field direction; The direction deviation is converted into an angular offset in a heading coordinate system, and the angular offset is used as the geomagnetic heading angle deviation.

7. The multi-agent collaborative drone swarm simulation training method according to claim 1 is characterized in that: The generating of flight control instructions based on the multi-target collaborative path and in combination with the real-time status data of the UAV cluster includes: Extracting the flight attitude of each UAV and the relative position of the enemy aircraft from the real-time status data; Analyzing the coordinates and arrival time limit of each UAV at the next navigation point from the multi-target collaborative path; For each UAV, the coordinates, the arrival time limit, the flight attitude, and the relative position of the enemy aircraft are integrated to calculate the roll angle control parameter and the angle of attack control parameter; The roll angle control parameters and the attack angle control parameters of all UAVs are encapsulated as flight control instructions.

8. A multi-agent collaborative drone swarm simulation training system, characterized by: include: The acquisition module is used to collect biological movement data of pigeon flocks, cluster trajectory data of drone clusters when performing historical penetration missions, and magnetic field gradient data of simulated environments; A construction module, for constructing a three-dimensional path library based on the cluster trajectory data and the pigeon flock biological motion data; A calculation module, configured to calculate a geomagnetic heading angle deviation based on the magnetic field gradient data; A first generation module is configured to generate a multi-objective collaborative path based on the geomagnetic heading angle deviation and the three-dimensional path library in combination with a swarm intelligence optimization model; A second generation module is configured to generate flight control instructions based on the multi-target collaborative path and in combination with the real-time status data of the UAV cluster; The loading module is used to load the flight control instructions into the simulation model of the UAV cluster, driving the UAV cluster to perform coordinated penetration training in a simulated environment.

9. 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 used to be called and executed by the processing component to implement a multi-agent collaborative drone cluster simulation training method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, a multi-agent collaborative drone cluster simulation training method as described in any one of claims 1 to 7 is implemented.

Images