Unmanned aerial vehicle swarm cooperative hunting method based on predicted hierarchical gene regulation network

Abstract

This invention relates to a collaborative UAV swarm capture method based on a predictive hierarchical gene regulatory network, belonging to the field of intelligent UAV swarm defense. First, a long-short term network is used to predict the target trajectory, and based on the predicted target and obstacle positions, a gene regulatory network is employed to obtain a pre-determined capture pattern. Second, a uniform sequence of capture points is generated on the capture pattern, and tasks are assigned to the UAVs, resulting in a specific target capture point for each UAV. Finally, after a UAV determines its assigned capture point, an improved artificial potential field method incorporating vortex obstacle avoidance force and damping mechanisms drives the UAV to move towards the assigned capture point. This invention is applicable to collaborative UAV swarm capture tasks, providing movement schemes for UAVs when they are capturing a target, ensuring efficient completion of the capture task while reducing energy consumption and avoiding collisions.

Description

Technical Field

[0001] This invention belongs to the field of intelligent encirclement and capture technology of drone swarms, and relates to a collaborative encirclement and capture method for drone swarms based on predictive hierarchical gene regulatory networks. Background Technology

[0002] In recent years, with the deep integration of drone technology and artificial intelligence, drone swarms have demonstrated enormous application potential in defensive missions such as modern electronic warfare, border patrol, and the encirclement of high-value targets. Compared to single-unit systems, swarm systems possess high redundancy, flexibility, and self-organization capabilities, enabling them to perform more challenging tasks in complex adversarial environments. Multi-target encirclement, as one of the core issues in swarm collaborative control, requires drone swarms to autonomously cooperate and form specific encirclement formations above targets in dynamic environments with potential obstacles, in order to monitor and restrict target movement. Its value lies not only in limiting the target's movement space but also in maintaining the continuity and integrity of defense during the encirclement process. Therefore, achieving efficient, accurate, and predictive encirclement has become a key support for improving active airspace defense capabilities in the field of automated control.

[0003] However, existing multi-UAV collaborative encirclement technologies still face severe challenges. On the one hand, in practical applications, the dynamic response to high-speed targets is insufficient. For example, although the Chinese invention patent "Dynamic Encirclement Method for UAV Swarms in Complex 3D Scenes" (CN121742486A) uses model predictive control to solve the safety and trajectory optimization problems in dynamic environments, when the target makes a high-maneuver escape, this method often leads to the swarm being caught in a "chase" situation, and it is prone to oscillations when approaching the target position. If there are obstacles in front, such oscillations can easily cause the UAVs to lag, thereby reducing the timeliness of the encirclement mission. On the other hand, the stability of swarm planning in complex obstacle environments is insufficient. For example, although the Chinese invention patent "Swarm Collaborative Navigation Method, Control System and Storage Medium" (CN121832628A) uses gene regulation networks to solve the problem of maintaining the swarm form under limited communication and complex terrain, in multi-target encirclement scenarios, due to the lack of prediction of the global encirclement formation, it is easy to induce frequent switching of task allocation and formation splitting during obstacle avoidance, which leads to low quality of UAV encirclement formation generation and high energy consumption, making it easy for the captured target to escape.

[0004] In summary, constructing a collaborative control architecture that combines target motion trend prediction with inertial decision constraints, introducing predictive algorithms to respond in advance to the encirclement task, and using inertial decision constraints to suppress position oscillations has become a key direction for breaking through the bottlenecks of existing encirclement technologies. Summary of the Invention

[0005] To address the aforementioned technical challenges, this invention proposes a collaborative UAV swarm capture method based on a predictive hierarchical gene regulatory network. This invention uses a Long Short-Term Memory (LSM) network to predict the escape trajectory of the target over time, transforming the current real-time capture formation into a predicted one. This enables early response to target deployment and improves the system's capture defense performance. Subsequently, a uniform set of capture points is generated, and a combinatorial optimization algorithm with a viscosity factor is defined to establish an optimal mapping between the desired capture points generated by the GRN and individual UAVs, ensuring the efficiency and orderliness of global task allocation. Furthermore, the artificial potential field method is improved by introducing vortex obstacle avoidance forces from nearby obstacles and a damping mechanism for approaching the target, solving the problem of traditional algorithms easily stalling under dense obstacles and ensuring that UAVs smoothly and stably reach the expected position during execution.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A drone swarm cooperative capture method based on a predictive hierarchical gene regulatory network includes the following steps: Step 1: Use long-term and short-term networks to predict the target trajectory, and based on the predicted target and obstacle positions, use a gene regulatory network to obtain the predicted encirclement pattern, specifically: Step 1-1: Predict the target trajectory using a Long Short-Term Memory (LSTM) network to obtain the predicted target position at the next moment. Specifically: Assuming in At any given time, LSTM will store the target's position coordinates over H historical time steps. Instantaneous velocity and the relative displacement vector with respect to nearby obstacles As input, output directly Time to Continuous time velocity vector of the target Define the target's velocity vector. ,…, .in, for The position coordinates of the target at any given time. for The position coordinates of the target at any given time. for The velocity vector of the target at any given moment. for The velocity vector of the target at any given moment. for The relative displacement vector of the target to the nearest obstacle at any given time. for The relative displacement vector of the target to the nearest obstacle at any given time. for The velocity vector of the target at any given moment. for The velocity vector of the target at any given moment. For unit time intervals, for The position coordinates of the target at any given time. for The position coordinates of the target at any given time. for The target's position coordinates at any given time; for The position coordinates of the target at any given time.

[0007] right The predicted velocity vectors of each step are linearly weighted to obtain the original prediction. velocity vector at time As shown in the following formula: (1) in, for The weighting factor is generally taken as This aims to balance the accuracy of near-term forecasts with the guidance of long-term trends; for The velocity vector of the target at any given moment; ,express That moment.

[0008] Considering the potential non-physical fluctuations in neural network prediction results, the original predictions are used directly. velocity vector at time High-frequency oscillations are prone to occur. Therefore, this invention introduces a time-coherent smoothing mechanism, which, through... Smooth predictions are obtained by performing weighted recursive processing. velocity vector at time To ensure the continuity of drone movement in the time domain, as shown in the following formula: (2) in, This is a smoothing factor used to balance the real-time response sensitivity and trajectory stability of commands. The closer it is to 0, the smoother the trajectory; the closer it is to 1, the more sensitive the response. It is generally set to... It balances smoothing and prediction.

[0009] According to the kinematic formula ,in, Represents the coordinates of the object's final position. Indicates the initial position coordinates of the object. Represents the velocity vector of an object. Indicates the time it takes for an object to move. Describes the target's position in time. The time predicted Time coordinates position This yields the predicted target position for the next moment, as shown in the following formula: (3) in, express Time coordinates; This indicates the unit of time the drone spends moving. Step 1-2: Using the target position and obstacle position predicted in Step 1-1 as input to the gene regulatory network, the original morphogenetic concentration is generated. The original morphogenetic concentration is then processed to calculate the comprehensive morphogenetic gradient and obtain the encirclement pattern. Specifically: Step 1-2-1: Generate the original morphogenetic concentration within the region based on the obstacle location and the predicted target location, where the region refers to the limited space of the encirclement action; In the process of generating the encirclement mode, firstly, the UAV uses exponential diffusion to simulate the concentration diffusion process in cells based on the location of obstacles and the predicted target location, and calculates the original morphogenetic concentration of obstacles and predicted targets in the area. The original morphogenetic concentration of obstacles and predicted targets obtained at each moment. As shown below: (4) (5) in, Indicates by the first An obstacle or predicted target in The morphogen concentration induced by time, of which , This refers to the total number of targets or obstacles. (Index) This is the concentration diffusion factor, used to adjust the mapping relationship between distance and morphogen concentration; and They are drones to obstacles or predicted target Horizontal and vertical distances; This indicates a drone or an obstacle.

[0010] Generate the original morpheme concentration of the target or obstacle based on all detected targets or obstacles. The original morphogenetic concentration of the obstacle is denoted as The concentration of the target original form element is denoted as .

[0011] Step 1-2-2: Process the original morphogenetic concentration to obtain the comprehensive morphogenesis gradient field and trapping pattern; The original morphogenetic concentration of the obstacle was calculated. and target original morphogen concentration Then, it needs to be further processed to generate the final coupled integrated morphological gradient field and trapping mode. The calculation process is as follows: (6) (7) (8) (9) In the formula, , and As Intermediate regulatory genes at specific times are used to hierarchically regulate the concentration of primitive morphogens. Here is the sigmoid function, where, , and These are the parameters in the sigmoid function. It refers to the concentration of morphogens; The activation threshold determines the spatial span of the gradient field; As a regulatory factor, it determines the steepness of the gradient. From the target original morphogen concentration Activation threshold With regulatory factors Calculation generated; From the original morphogenetic concentration of the obstacle Activation threshold With regulatory factors Calculated generation. As coupling and Activation threshold With regulatory factors The gradient field generated by the comprehensive morphology is the core output variable.

[0012] Gradient field generated by taking the comprehensive shape Medium concentration is the preset threshold The contour lines form an adaptive closed manifold, which is called... Predictive-Pattern (PP) of time-based capture patterns, where the same preset threshold is used. There are multiple capture modes.

[0013] Step 2: Generate a uniform sequence of capture points based on the capture pattern calculated in Step 1, and assign tasks to the drones to obtain the target capture point determined for each drone, specifically: Step 2-1: Generate capture points evenly distributed on each capture pattern according to the perimeter ratio and length interval mechanism of the capture pattern; Step 2-1-1: Allocate the number of capture points in the encirclement mode according to the proportional fairness principle; First, calculate the number of capture points for each encirclement mode. For multiple encirclement modes with different perimeters in a multi-object scenario, use a proportional fairness criterion to calculate the number of capture points for each encirclement mode, i.e., allocate them to the [number of] capture points. Number of drones in each encirclement mode Based on its manifold perimeter Dynamic calculation, as shown in the following formula: (10) in, The total number of drones; For the first The manifold perimeter of a single encirclement pattern; For the number of encirclement and capture modes, Indicates the first Each encirclement mode is used sequentially. The aforementioned proportionality fairness principle ensures that large-scale, high-volume targets can be surrounded and captured by a denser swarm of drones.

[0014] Given the number of capture points in each encirclement mode, capture point positions are randomly selected with equal arc lengths according to the number of capture points in each encirclement mode. All capture point positions constitute the initial capture point sequence. The equal arc lengths ensure the uniformity of capture points across the encirclement modes. However, since the order in which capture points are selected and the first capture point are random, the initial capture point sequence obtained in this step is unstable. Subsequent steps require fixing the capture point sequence to ensure stability.

[0015] Step 2-1-2: Determine the capture point sequence by using clockwise sorting and defining a global reference vector; The initial capture point sequence is determined by defining the capture point counting order and optimizing the absolute starting point. In this invention, capture points in each encirclement mode are counted clockwise. First, the signed area formed by the capture points in each encirclement mode in the initial capture point sequence is calculated using the Gaussian area formula. Here, we take the first... Signed area formed by each encirclement pattern The following formula provides an explanation: (11) in, To be allocated to the first The number of drones in each encirclement and capture mode; Indicates the first The first encirclement mode One capture point; For capture point The x-coordinate; For capture point The ordinate; For capture point The x-coordinate; For capture point The ordinate; when hour, The value is 1, which is the capture point. Location and capture point coincide; pass The sign of the target is used to determine the order of capture points in each capture mode, unifying the order to a clockwise direction, thereby optimizing the orderliness of the initial capture point sequence. Furthermore, to suppress sequence changes in the capture point sequence due to rapid target turning, this invention defines a global reference vector. (i.e., due east in the inertial frame), the easternmost point of the capture mode is used as the absolute starting point. Finally, the capture point sequence is determined by clockwise sorting and defining a global reference vector.

[0016] Step 2-1-3: Set up a capture point smoothing mechanism to obtain a spatially continuous sequence of capture points that varies over time; Based on the number, order, and absolute starting point position of each capture point assigned to each capture mode, the coordinate positions of all capture points are calculated. Since the capture mode is constantly being reshaped, even with an absolute starting point set, slight oscillations still occur in the movement of the same capture point over time. Therefore, a capture point smoothing mechanism is implemented to ensure that the spatial movement of each capture point in the capture point sequence is continuous and smooth over time; the capture point smoothing mechanism is shown in the following formula: (12) in, for The actual location of the point is captured at any given moment; for The actual location of the point is captured at any given moment; According to The ideal capture point is calculated based on the global reference vector and equal arc length sampling at any given moment; As a smoothing factor to balance the actual capture point at the previous time step with the calculated capture point at this time step, if The capture point moves more slowly and smoothly over time, but has poor anti-interference ability; If the capture point follows the theoretically calculated capture point, although the real-time performance is good, it is prone to jitter and oscillation.

[0017] Step 2-2: Given the sequence of drone capture points, solve the allocation relationship between drones and capture points by combining optimization problem models, and add a stickiness factor to achieve more stable drone task allocation; Step 2-2-1: Establish a model for the UAV task allocation and combination optimization problem; After obtaining the set of capture points indicating the actual trajectory of the drone, it is necessary to implement the mapping between the drone and the capture points. The task assignment problem is modeled as a combinatorial optimization problem of bipartite graph matching. Given... A drone, A number of capture points (i.e., the number of drones equals the number of capture points) are defined. The time allocation matrix is ,when At that time, the current planning of drones Assigned to the capture point ,in, For drone indexing, For the capture point index. The objective of the task allocation problem is to minimize the cost function of the average obstacle avoidance distance of the drone swarm. As shown in the following formula: (13) in, For obstacle avoidance distance matrix The elements include not only the currently planned drones With capture point The Euclidean distance between them also integrates a penalty operator based on obstacle location, thereby pre-excluding physically unreachable paths; The binary decision variable represents the currently planned drone. Whether assigned to the capture point 0 indicates no, and 1 indicates yes; For the current planning of drone index, , This refers to the total number of drones and the total number of capture points; For the capture point index, ; This refers to the decision variables for all possible drone mission assignments. In, find one , making the cost function Obtain the minimum value.

[0018] Meanwhile, the system must meet the following allocation constraints: each drone is assigned only one capture point, and each capture point is occupied by only one drone to ensure the uniqueness and completeness of resource allocation, as shown in the following formula: (14) Solving the constrained combinatorial optimization problem shown in formula (14), we obtain... Time allocation matrix .

[0019] Step 2-2-2: Add a stickiness factor to the task allocation combinatorial optimization problem model to avoid frequent task switching and obtain a more stable UAV task allocation; In situations involving continuous multi-target maneuvering and complex obstacles, the allocation strategy between the UAV and the acquisition point is prone to frequent switching. Therefore, this invention designs a combinatorial optimization allocation strategy based on inertial decision-making to achieve more stable task allocation. Specifically, the optimization strategy involves setting a viscosity factor and resetting the obstacle avoidance distance matrix in the cost function. Obstacle avoidance distance at any time Received Time allocation status The impact, ultimately Obstacle avoidance distance matrix with viscosity factor at all times This yields the corrected obstacle avoidance distance matrix. The inertial decision-making mechanism is shown in the following formula: (15) in, It is a viscosity factor. The closer it is to 0, the harder it is to trigger task redistribution.

[0020] Inertial decision-making mechanism refers to the assumption that the current planning of drones... Currently heading to the capture point If a new capture point that is closer appears in space Only when the efficiency gain brought by the new capture point is sufficient to overcome the resistance of the old capture point, i.e. The algorithm will only trigger task redistribution at this time. Indicates the current planning of drones To capture point Obstacle avoidance distance, Indicates the current planning of drones To the new capture point Obstacle avoidance distance.

[0021] Finally, based on the corrected obstacle avoidance distance matrix, the Hungarian algorithm is used to solve the global minimum weighted complete matching problem, resulting in a more stable UAV task allocation. Through a combinatorial optimization allocation strategy based on inertial decision-making, this invention not only achieves a reasonable allocation of UAVs and capture points but also reduces frequent switching of allocation strategies due to complex environments, thereby making UAV operation more stable and reducing UAV energy waste.

[0022] Step 3: After the UAV determines its assigned capture point in Step 2, it is driven to move towards the assigned capture point using an improved artificial potential field method incorporating vortex obstacle avoidance force and damping mechanism. Specifically: Step 3-1: Calculate the four components affecting the drone's movement; the four components are the target traction force, the drone's repulsive force, the vortex obstacle avoidance force, and the capture point attraction force. Specifically: Step 3-1-1: Calculate the target traction force ; The target traction field is affected by the difference between the concentration value at the location of the drone and a preset threshold, driving the drone to move towards the encirclement mode. The specific formula is expressed as follows: (16) in, Is it a currently planned drone? The concentration value at the current location; This is a threshold setting, typically ranging from 0.65 to 0.95. The coordinates of the current planned drone's location. These are the position coordinates of the corresponding target.

[0023] Step 3-1-2: Calculate the mutual repulsion force of the drones ; When the distance between the two drones is less than the preset radius At the same time, the distance between the drones generates a repulsive force to ensure the safety of the swarm and promote a uniform distribution in the encirclement mode. The mutual repulsion force between the drones... The specific formula is expressed as follows: (17) in, Indicates the mutual exclusion strength coefficient; From drones Pointing to the currently planned drone vector, drone In addition to the currently planned drones Other drones besides those; It is the length of the vector. This indicates that from the drone Pointing to the currently planned drone The unit vector. The closer the drone is, the higher the repulsive force, designed to avoid drone collision risks during tighter encirclement maneuvers or high-speed maneuvers. The preset radius. The range is generally taken arrive between.

[0024] Step 3-1-3: Calculate the vortex obstacle avoidance force ; Traditional artificial potential field methods can cause drones to stop very close to obstacles or stall in narrow gaps within obstacles. To address these issues, this invention incorporates a radial repulsive force. and tangential force composite model The formula for the composite model is as follows: (18) in, Radial repulsion weight; This is the obstacle avoidance sensing radius; beyond this distance, the obstacle will not be affected by the repulsive force. The shortest Euclidean distance from the edge of the drone to the edge of the obstacle; It is a very small constant; For vectors Orthogonal unit normal vectors, where, This represents the displacement vector from the obstacle to the drone; The weight of the tangential force; To ensure smooth operation, the tangential force is applied smoothly at the induced boundary, avoiding abrupt changes; As a direction decision factor, it determines whether the direction of the tangential force is clockwise or counterclockwise; For vectors Orthogonal tangential vectors.

[0025] Specifically, Its use is to prevent The combined force of vortex obstacle avoidance approaches infinity, and the direction decision factor... ,in This represents the displacement vector from the UAV to the assigned capture point.

[0026] Step 3-1-4: Calculate the attraction force at the capture point ; The capture point tendency force is the main component force for achieving precise capture, used to drive the UAV rapidly toward the corresponding capture point calculated by the decision-making layer. To address the oscillations caused by velocity inertia overshoot during dynamic capture, an adaptive damping mechanism is implemented to enable the UAV to rapidly approach the capture point when it is far away and to land stably when it is close to eliminate oscillations. This is achieved by generating a pulling force at the capture point position and using damping to prevent velocity overshoot after reaching the position. The specific expression is: (19) in, The value is a scalar, representing the currently planned drone. Current location With assigned capture points Spatial distance; For current planning of drones The coordinates of the assigned capture point; For current planning of drones Current position coordinates; For current planning of drones Current velocity vector. , The current planned drones about The tensile force coefficient and damping coefficient; Specifically, regarding Tensile coefficient and damping coefficient , The formula is as follows: (20) (twenty one) in, This is the deceleration boundary distance; This represents the magnitude of the traction force before deceleration. This refers to the magnitude of the traction force after deceleration. This represents the magnitude of the damping force before deceleration. The magnitude of the damping force after deceleration, where The repulsion coefficient at close range. For current planning of drones Current location With assigned capture points Spatial distance.

[0027] In particular, the damping coefficient As the drone approaches the target acquisition point, its speed increases exponentially. The damping mechanism slows down the drone as it approaches the target acquisition point coordinates, thereby achieving precise locking and stable landing and preventing oscillations.

[0028] Step 3-2: Sum the four component forces obtained in Step 3-1 with weights to calculate the resultant force and velocity vector of the UAV; After obtaining the force field components, the drone synthesizes them. The final resultant force at any moment : (twenty two) in, , , , Target traction force mutual repulsion of drones Vortex obstacle avoidance force Capture point tendency force The weights are applied to different forces to achieve smooth motion in complex environments.

[0029] From the basic formula of kinematics , It can be known that ,in, The net force acting on the object. For the mass of the object, Let be the acceleration of the object's motion. It is the change in the velocity of the object. Let be the time of the object's motion. For each drone, the final net force acting on it has been calculated as follows: The mass of the drone is known to be The unit movement time of the drone is Calculate the time of a drone per unit of movement velocity change vector within, velocity change vector The calculation formula is as follows: (twenty three) To better reflect practical applications, the cutoff speed is set. As an actual speed constraint. The speed of the drone calculated based on the resultant force. Exceeding the deadline speed If the desired direction remains unchanged, its size is limited to... If its size Not exceeding If the original speed remains unchanged, then the final speed can be obtained. The details are as follows: (twenty four) That is, each drone in Always follow the calculated velocity vector Perform real-time displacement.

[0030] The beneficial effects of this invention are as follows: (1) First, the present invention uses LSTM network to perform time-series modeling of the target’s historical displacement and obstacle constraints to obtain the predicted velocity vector. By converting the real-time formation into the predicted encirclement mode (PP), the cluster can evolve in advance to the area that the target will inevitably pass through in the future, thus realizing the advance deployment and active response to highly maneuverable targets.

[0031] (2) Secondly, the present invention introduces a global reference vector to establish an absolute starting point, and combines equal arc length sampling to eliminate the logical jump caused by the target rotation of the capture point. At the same time, an inertial decision-making mechanism with a sticky factor is introduced in the allocation stage to ensure that the task switching is triggered only when the efficiency gain of the new target is sufficient to overcome the sticky resistance of the old target. This suppresses the frequent switching of allocation strategies and reduces energy consumption from the decision-making level.

[0032] (3) Finally, for motion control in complex environments, this invention designs a vortex obstacle avoidance force that includes radial repulsion and tangential force. The tangential force guides the UAV around obstacles, solving the problem that traditional algorithms are prone to stalling in narrow gaps. Combined with the adaptive damping mechanism in the capture point tending force, the damping coefficient increases exponentially with decreasing distance, effectively absorbing kinetic energy and eliminating the oscillation caused by velocity inertia overshoot. Finally, it realizes the smooth generation of UAV velocity vector and the rapid convergence of the capture task.

[0033] In summary, this invention effectively overcomes the bottleneck of insufficient stability in high-speed maneuvering target capture and complex obstacle environments by constructing a method that combines target motion trend prediction, inertial decision constraints, and an improved artificial potential field method (steps 1, 2, and 3). Attached Figure Description

[0034] Figure 1 This is a flowchart of the present invention.

[0035] Figure 2 This is a schematic diagram illustrating the solution process of this invention.

[0036] Figure 3 A schematic diagram illustrating the process of generating a capture pattern for target locations based on LSTM prediction.

[0037] Figure 4 A top-down animation depicting 10 drones navigating two obstacles to surround and capture a target. Figure 5 Concentration contour map of the encirclement mode of 10 drones navigating two obstacles to capture a target; Figure 6 A 3D model of the encirclement pattern of 10 drones navigating two obstacles to capture a target.

[0038] Figure 7 This is a schematic diagram of the target traction force.

[0039] Figure 8 Directional decision factors At that time, the tangential force of the vortex obstacle avoidance force is in a clockwise direction. Figure 9 Directional decision factors At that time, a schematic diagram of the tangential force of the vortex obstacle avoidance force in the counterclockwise direction is shown. Figure 10 Directional decision factors Schematic diagram showing that the tangential force of the vortex obstacle avoidance is 0 at that time; Figure 11 This is a schematic diagram of vortex obstacle avoidance force.

[0040] Figure 12 This is a physical schematic diagram of the damping mechanism; Figure 12 (a) in the diagram is a schematic diagram of the physical mechanism of the tension coefficient and damping coefficient when the UAV is far from the capture point; Figure 12 (b) in the figure is a schematic diagram of the physical mechanism of the tension coefficient and damping coefficient when the UAV is close to the capture point.

[0041] Figure 13 This is a graph showing the variation of the damping coefficient in the tendency force at the capture point.

[0042] Figure 14 This is a diagram illustrating the resultant force.

[0043] Figure 15 A simulation animation of 12 drones capturing three randomly moving targets in a complex, random obstacle environment; Figure 15 (a) in the middle is Real-time animation simulation; Figure 15 (b) in the middle is Real-time animation simulation; Figure 15 (c) in the middle is Real-time animation simulation; Figure 15 (d) in the middle is Animation simulation of the time.

[0044] Figure 16 This invention demonstrates the uniformity of encirclement during a simulation of 12 drones encircling three randomly moving targets in a complex, random obstacle environment.

[0045] Figure 17 This invention demonstrates the average movement distance of the drones during a simulation of 12 drones surrounding and capturing three randomly moving targets in a complex obstacle environment.

[0046] Figure 18 This invention demonstrates the numerical value of the target escape distance during a simulation of 12 drones surrounding and capturing three randomly moving targets in a complex, random obstacle environment. Detailed Implementation

[0047] The present invention will be further described below with reference to specific embodiments. Figure 1 This is a flowchart of the present invention, considering the problem of encirclement and capture of multiple drones in a complex obstacle environment. An embodiment demonstrates the performance of 12 drones in encircling and capturing three randomly moving targets in a random and complex obstacle environment.

[0048] This embodiment provides a method for collaborative trapping of drone swarms based on a predictive hierarchical gene regulatory network, as shown in the flowchart below. Figure 1 As shown in the diagram, the solution is illustrated below. Figure 2 As shown. Includes the following steps: Step 1: Use long-term and short-term networks to predict the target trajectory, and based on the predicted target and obstacle positions, use a gene regulatory network to obtain the predicted encirclement pattern, specifically: Step 1-1: Predict the target trajectory using a Long Short-Term Memory (LSTM) network to obtain the predicted target position at the next moment. Specifically: Assuming in At any given time, LSTM will store the target's position coordinates over H historical time steps. Instantaneous velocity and the relative displacement vector with respect to nearby obstacles As input, output directly Time to Continuous time velocity vector of the target Define the target's velocity vector. ,…, .in, for The position coordinates of the target at any given time. for The position coordinates of the target at any given time. for The velocity vector of the target at any given moment. for The velocity vector of the target at any given moment. for The relative displacement vector of the target to the nearest obstacle at any given time. for The relative displacement vector of the target to the nearest obstacle at any given time. for The velocity vector of the target at any given moment. for The velocity vector of the target at any given moment. For unit time intervals, for The position coordinates of the target at any given time. for The position coordinates of the target at any given time. for The target's position coordinates at any given time; for The target's position coordinates at any given time. This example uses... right The predicted velocity vectors of each step are linearly weighted to obtain the original prediction. velocity vector at time As shown in the following formula: (1) in, for The weighting factor is generally taken as This aims to balance the accuracy of near-term forecasts with the guidance of long-term trends; for The velocity vector of the target at any given moment; ,express At any given moment. This example takes... , .

[0049] Considering the potential non-physical fluctuations in neural network prediction results, the original predictions are used directly. velocity vector at time High-frequency oscillations are prone to occur. Therefore, this invention introduces a time-coherent smoothing mechanism, which, through... Smooth predictions are obtained by performing weighted recursive processing. velocity vector at time To ensure the continuity of drone movement in the time domain, as shown in the following formula: (2) in, This is a smoothing factor used to balance the real-time response sensitivity and trajectory stability of commands. The closer it is to 0, the smoother the trajectory; the closer it is to 1, the more sensitive the response. It is generally set to... To balance smoothing and prediction, this example takes... .

[0050] According to the kinematic formula ,in, Represents the coordinates of the object's final position. Indicates the initial position coordinates of the object. Represents the velocity vector of an object. Indicates the time it takes for an object to move. Describes the target's position in time. The time predicted Time coordinates position This yields the predicted target position for the next moment, as shown in the following formula: (3) in, express Time coordinates; This indicates the unit of time the drone spends moving. Figure 3This explains the process of predicting the target trajectory using LSTM. Target position at any time The target's capture pattern is The target passed through speed get Target position at any time ,exist Obtain the original prediction at any time through LSTM prediction. velocity vector at time Then, a weighted recursive process is performed to obtain a smooth prediction. velocity vector at time Thus, the prediction is obtained. Target position at any time The follow-up will be based on Get The target's encirclement and capture mode at any time .

[0051] Step 1-2: Using the target position and obstacle position predicted in Step 1-1 as input to the gene regulatory network, the original morphogenetic concentration is generated. The original morphogenetic concentration is then processed to calculate the comprehensive morphogenetic gradient and obtain the encirclement pattern. Specifically: Step 1-2-1: Generate the original morphogenetic concentration within the region based on the obstacle location and the predicted target location, where the region refers to the limited space of the encirclement action; In the process of generating the encirclement mode, firstly, the UAV uses exponential diffusion to simulate the concentration diffusion process in cells based on the location of obstacles and the predicted target location, and calculates the original morphogenetic concentration of obstacles and predicted targets in the area. The original morphogenetic concentration of obstacles and predicted targets obtained at each moment. As shown below: (4) (5) in, Indicates by the first An obstacle or predicted target in The morphogen concentration induced by time, of which , This represents the total number of targets or the total number of obstacles. In this example, we assume a total of 3 targets and a total of 10 obstacles. The value is 3 when representing the total number of targets. The index is 10 when representing the total number of obstacles. For this example, we take the concentration diffusion factor as . This is used to adjust the mapping relationship between distance and morphogen concentration; and They are drones to obstacles or predicted target Horizontal and vertical distances; This indicates a drone or an obstacle.

[0052] Generate the original morpheme concentration of the target or obstacle based on all detected targets or obstacles. The original morphogenetic concentration of the obstacle is denoted as The concentration of the target original form element is denoted as .

[0053] Step 1-2-2: Process the original morphogenetic concentration to obtain the comprehensive morphogenesis gradient field and trapping pattern; The original morphogenetic concentration of the obstacle was calculated. and target original morphogen concentration Then, it needs to be further processed to generate the final coupled integrated morphological gradient field and trapping mode. The calculation process is as follows: (6) (7) (8) (9) In the formula, , and As Intermediate regulatory genes at specific times are used to hierarchically regulate the concentration of primitive morphogens. Here is the sigmoid function, where, , and These are the parameters in the sigmoid function. It refers to the concentration of morphogens; The activation threshold determines the spatial span of the gradient field; As a regulatory factor, it determines the steepness of the gradient. From the target original morphogen concentration Activation threshold With regulatory factors This calculation was generated; this example uses... ; From the original morphogenetic concentration of the obstacle Activation threshold With regulatory factors This calculation was generated; this example uses... . As coupling and Activation threshold With regulatory factors The gradient field generated by the comprehensive morphology is the core output variable.

[0054] Gradient field generated by taking the comprehensive shape Medium concentration is the preset threshold The contour lines form an adaptive closed manifold, which is called... Predictive-Pattern (PP) of time-based capture patterns, where the same preset threshold is used. There are multiple capture modes; this example takes... .

[0055] The drones adaptively generate a comprehensive morphological gradient field within the environment. To more intuitively demonstrate the predicted encirclement pattern, we created an animated scene showing 10 drones navigating two obstacles to encircle a target. The top-down view of the animated scene is shown below. Figure 4 As shown, the concentration contour lines of the containment mode are as follows: Figure 5 As shown, the three-dimensional image of the concentration in the containment mode is as follows: Figure 6 As shown, the generated capture pattern clearly exhibits significant adaptive properties: when the target approaches an obstacle, the generated predicted capture pattern deforms, guiding the group to autonomously avoid the obstacle, demonstrating excellent adaptability.

[0056] Step 2: Generate a uniform sequence of capture points based on the capture pattern calculated in Step 1, and assign tasks to the drones to obtain the target capture point determined for each drone, specifically: Step 2-1: Generate capture points evenly distributed on each capture pattern according to the perimeter ratio and length interval mechanism of the capture pattern; Step 2-1-1: Allocate the number of capture points in the encirclement mode according to the proportional fairness principle; First, calculate the number of capture points for each encirclement mode. For multiple encirclement modes with different perimeters in a multi-object scenario, use a proportional fairness criterion to calculate the number of capture points for each encirclement mode, i.e., allocate them to the [number of] capture points. Number of drones in each encirclement mode Based on its manifold perimeter Dynamic calculation, as shown in the following formula: (10) in, The total number of drones; For the first The manifold perimeter of a single encirclement pattern; For the number of encirclement and capture modes, Indicates the first Each encirclement mode is used sequentially. . This calculation example takes , Obtained from step 1. The above proportionality fairness criterion ensures that large-scale, high-volume targets can be surrounded and captured by a denser swarm of drones.

[0057] Given the number of capture points in each encirclement mode, capture point positions are randomly selected with equal arc lengths according to the number of capture points in each encirclement mode. All capture point positions constitute the initial capture point sequence. The equal arc lengths ensure the uniformity of capture points across the encirclement modes. However, since the order in which capture points are selected and the first capture point are random, the initial capture point sequence obtained in this step is unstable. Subsequent steps require fixing the capture point sequence to ensure stability.

[0058] Step 2-1-2: Determine the capture point sequence by using clockwise sorting and defining a global reference vector; The initial capture point sequence is determined by defining the capture point counting order and optimizing the absolute starting point. In this invention, capture points in each encirclement mode are counted clockwise. First, the signed area formed by the capture points in each encirclement mode in the initial capture point sequence is calculated using the Gaussian area formula. Here, we take the first... Signed area formed by each encirclement pattern The following formula provides an explanation: (11) in, To be allocated to the first The number of drones in each encirclement and capture mode; Indicates the first The first encirclement mode One capture point; For capture point The x-coordinate; For capture point The ordinate; For capture point The x-coordinate; For capture point The ordinate; when hour, The value is 1, which is the capture point. Location and capture point coincide; pass The sign of the target is used to determine the order of capture points in each capture mode, unifying the order to a clockwise direction, thereby optimizing the orderliness of the initial capture point sequence. Furthermore, to suppress sequence changes in the capture point sequence due to rapid target turning, this invention defines a global reference vector. (i.e., due east in the inertial frame), the easternmost point of the capture mode is used as the absolute starting point. Finally, the capture point sequence is determined by clockwise sorting and defining a global reference vector.

[0059] Step 2-1-3: Set up a capture point smoothing mechanism to obtain a spatially continuous sequence of capture points that varies over time; Based on the number, order, and absolute starting point position of each capture point assigned to each capture mode, the coordinate positions of all capture points are calculated. Since the capture mode is constantly being reshaped, even with an absolute starting point set, slight oscillations still occur in the movement of the same capture point over time. Therefore, a capture point smoothing mechanism is implemented to ensure that the spatial movement of each capture point in the capture point sequence is continuous and smooth over time; the capture point smoothing mechanism is shown in the following formula: (12) in, for The actual location of the point is captured at any given moment; for The actual location of the point is captured at any given moment; According to The ideal capture point is calculated based on the global reference vector and equal arc length sampling at any given moment; As a smoothing factor to balance the actual capture point at the previous time step with the calculated capture point at this time step, if The capture point moves more slowly and smoothly over time, but has poor anti-interference ability; If the capture point follows the theoretically calculated capture point, although the real-time performance is good, it is prone to jitter and oscillation. This example uses... .

[0060] Step 2-2: Given the sequence of drone capture points, solve the allocation relationship between drones and capture points by combining optimization problem models, and add a stickiness factor to achieve more stable drone task allocation; Step 2-2-1: Establish a model for the UAV task allocation and combination optimization problem; After obtaining the set of capture points indicating the actual trajectory of the drone, it is necessary to implement the mapping between the drone and the capture points. The task assignment problem is modeled as a combinatorial optimization problem of bipartite graph matching. Given... A drone, This example takes [number] capture points. ,definition The time allocation matrix is ,when At that time, the current planning of drones Assigned to the capture point ,in, For drone indexing, For the capture point index. The objective of the task allocation problem is to minimize the cost function of the average obstacle avoidance distance of the drone swarm. As shown in the following formula: (13) in, For obstacle avoidance distance matrix The elements include not only the currently planned drones With capture point The Euclidean distance between them also integrates a penalty operator based on obstacle location, thereby pre-excluding physically unreachable paths; The binary decision variable represents the currently planned drone. Whether assigned to the capture point 0 indicates no, and 1 indicates yes; For drone indexing, , This refers to the total number of drones and the total number of capture points; For the capture point index, ; This refers to the decision variables for all possible drone mission assignments. In, find one , making the cost function Obtain the minimum value.

[0061] Meanwhile, the system must meet the following allocation constraints: each drone is assigned only one capture point, and each capture point is occupied by only one drone to ensure the uniqueness and completeness of resource allocation, as shown in the following formula: (14) Solving the constrained combinatorial optimization problem shown in formula (14), we obtain... Time allocation matrix .

[0062] Step 2-2-2: Add a stickiness factor to the task allocation combinatorial optimization problem model to avoid frequent task switching and obtain a more stable UAV task allocation; In situations involving continuous multi-target maneuvering and complex obstacles, the allocation strategy between the UAV and the acquisition point is prone to frequent switching. Therefore, this invention designs a combinatorial optimization allocation strategy based on inertial decision-making to achieve more stable task allocation. Specifically, the optimization strategy involves setting a viscosity factor and resetting the obstacle avoidance distance matrix in the cost function. Obstacle avoidance distance at any time Received Time allocation status The impact, ultimately Obstacle avoidance distance matrix with viscosity factor at all times This yields the corrected obstacle avoidance distance matrix. The inertial decision-making mechanism is shown in the following formula: (15) in, It is a viscosity factor. The closer to 0, the less likely it is to trigger task redistribution. This example takes... .

[0063] Inertial decision-making mechanism refers to the assumption that the current planning of drones... Currently heading to the capture point If a new capture point that is closer appears in space Only when the efficiency gain brought by the new capture point is sufficient to overcome the resistance of the old capture point, i.e. The algorithm will only trigger task redistribution at this time. Indicates the current planning of drones To capture point Obstacle avoidance distance, Indicates the current planning of drones To the new capture point Obstacle avoidance distance.

[0064] Finally, based on the corrected obstacle avoidance distance matrix, the Hungarian algorithm is used to solve the global minimum weighted complete matching problem, resulting in a more stable UAV task allocation. Through a combinatorial optimization allocation strategy based on inertial decision-making, this invention not only achieves a reasonable allocation of UAVs and capture points but also reduces frequent switching of allocation strategies due to complex environments, thereby making UAV operation more stable and reducing UAV energy waste.

[0065] Step 3: After the UAV determines its assigned capture point in Step 2, it is driven to move towards the assigned capture point using an improved artificial potential field method incorporating vortex obstacle avoidance force and damping mechanism. Specifically: Step 3-1: Calculate the four components affecting the drone's movement; the four components are the target traction force, the drone's repulsive force, the vortex obstacle avoidance force, and the capture point attraction force. Specifically: Step 3-1-1: Calculate the target traction force ; The target traction field is affected by the difference between the concentration value at the location of the drone and a preset threshold, driving the drone to move towards the encirclement mode. The specific formula is expressed as follows: (16) in, Is it a currently planned drone? The concentration value at the current location; It sets a threshold; in this embodiment, it is taken as... . The location coordinates of the drone. These are the position coordinates of the corresponding target. Target traction force. Schematic diagram as follows Figure 7 As shown, the concentration value is highest at the target location, with a value of 1, and radiates outward from the target location. The red line represents the threshold contour line we set. The image shows the target traction. Three situations: (1) If the concentration at the location of the drone exceeds the threshold, the drone will experience an outward repulsive force; (2) If the concentration at the location of the drone is below the threshold, then move towards the target; (3) If the drone is located exactly on the threshold contour line, then the component is 0.

[0066] Step 3-1-2: Calculate the mutual repulsion force of the drones ; When the distance between the two drones is less than the preset radius At the same time, the distance between the drones generates a repulsive force to ensure the safety of the swarm and promote a uniform distribution in the encirclement mode. The mutual repulsion force between the drones... The specific formula is expressed as follows: (17) in, Indicates the mutual exclusion strength coefficient; From drones Pointing to the currently planned drone The vector, It is the length of the vector. This indicates that from the drone Pointing to the currently planned drone The unit vector. The closer the drone is, the higher the repulsive force, which aims to avoid the risk of drone collisions during tight encirclement or high-speed maneuvers. This embodiment takes... .

[0067] Step 3-1-3: Calculate the vortex obstacle avoidance force ; The design incorporates radial repulsion. and tangential force composite model The formula for the composite model is as follows: (18) in, Radial repulsion weight; This is the obstacle avoidance sensing radius; beyond this distance, the obstacle will not be affected by the repulsive force. The shortest Euclidean distance from the edge of the drone to the edge of the obstacle; For vectors Orthogonal unit normal vectors, where, This represents the displacement vector from the obstacle to the drone; The weight of the tangential force; For smoothing operators; As a directional decision factor; For vectors Orthogonal tangential vectors. In this embodiment, we take... .

[0068] Specifically, Its use is to prevent The combined force of vortex obstacle avoidance approaches infinity, and the direction decision factor... ,in This represents the displacement vector from the UAV to the assigned capture point. The direction of the tangential force of the vortex obstacle avoidance force is different when the value is different. Figure 8 Directional decision factors At that time, the tangential force of the vortex obstacle avoidance force is in a clockwise direction. Figure 9 Directional decision factors At that time, a schematic diagram of the tangential force of the vortex obstacle avoidance force in the counterclockwise direction is shown. Figure 10 The diagram shows that the tangential force of the vortex obstacle avoidance force is 0 when the direction decision factor is used. Figure 11 For including radial repulsion and tangential force Vortex obstacle avoidance force The diagram simulates time, time, time, Time and The radial repulsive force, tangential force, and obstacle avoidance trajectory at any given moment.

[0069] Step 3-1-4: Calculate the attraction force at the capture point ; The capture point tendency force is the main component force for achieving precise capture, used to drive the UAV rapidly toward the corresponding capture point calculated by the decision-making layer. To address the oscillations caused by velocity inertia overshoot during dynamic capture, an adaptive damping mechanism is implemented to enable the UAV to rapidly approach the capture point when it is far away and to land stably when it is close to eliminate oscillations. This is achieved by generating a pulling force at the capture point position and using damping to prevent velocity overshoot after reaching the position. The specific expression is: (19) in, The value is a scalar, representing the currently planned drone. Current location With assigned capture points Spatial distance; For current planning of drones The coordinates of the assigned capture point; For current planning of drones Current position coordinates; For current planning of drones Current velocity vector. , The current planned drones about The tensile and damping coefficients, and the physical meaning of the damping mechanism used in this invention can be found in the physical diagram. Figure 12 (a) in the diagram is a schematic diagram of the physical mechanism of the tension coefficient and damping coefficient when the UAV is far from the capture point. Figure 12 (b) is a schematic diagram of the physical mechanism of the tension coefficient and damping coefficient when the UAV is close to the capture point; Specifically, regarding Tensile coefficient and damping coefficient , The formula is as follows: (20) (twenty one) in, This is the deceleration boundary distance; This represents the magnitude of the traction force before deceleration. This refers to the magnitude of the traction force after deceleration. This represents the magnitude of the damping force before deceleration. The magnitude of the damping force after deceleration, where The repulsion coefficient at close range. For current planning of drones Current location With assigned capture points Spatial distance. Specifically, the damping coefficient. The growth rate increases exponentially as the drone approaches the target acquisition point, such as... Figure 13 As shown. This damping mechanism slows the UAV as it approaches the target acquisition point, thereby achieving precise locking and stable landing while preventing oscillations. In this example, we take... .

[0070] In particular, the damping coefficient As the drone approaches the target acquisition point, its speed increases exponentially. The damping mechanism slows down the drone as it approaches the target acquisition point coordinates, thereby achieving precise locking and stable landing and preventing oscillations.

[0071] Step 3-2: Sum the four component forces obtained in Step 3-1 with weights to calculate the resultant force and velocity vector of the UAV; After obtaining the force field components, the drone synthesizes them. The final resultant force at any moment : (twenty two) in, , , , Target traction force mutual repulsion of drones Vortex obstacle avoidance force Capture point tendency force The weights are determined by applying different weights to different forces to achieve smooth motion in complex environments. In this example, we take... The resultant force diagram is as follows: Figure 14 As stated above.

[0072] From the basic formula of kinematics , It can be known that ,in, The net force acting on the object. For the mass of the object, Let be the acceleration of the object's motion. It is the change in the velocity of the object. Let be the time of the object's motion. For each drone, the final net force acting on it has been calculated as follows: The mass of the drone is known to be The unit movement time of the drone is Calculate the time of a drone per unit of movement The velocity change vector within the space is taken in this example. , velocity change vector The calculation formula is as follows: (twenty three) To better reflect practical applications, the cutoff speed is set. As a practical constraint, this example takes If the speed of the drone is... If the speed exceeds the cutoff speed, maintain the desired speed direction and change its magnitude. For final speed The details are as follows: (twenty four) That is, each drone in Always follow the calculated velocity vector Perform real-time displacement.

[0073] In summary, the UAV solves for the velocity vector at each moment according to this invention, as shown in the schematic diagram below. Figure 2 As shown.

[0074] Following steps 1-3, algorithm simulation was conducted. The algorithm simulated 12 drones surrounding and capturing three randomly moving targets in a complex, random obstacle environment. The algorithm took 80 seconds. Figure 15 (a) in the middle is Real-time animation simulation; Figure 15 (b) in the middle is Real-time animation simulation; Figure 15 (c) in the middle is Real-time animation simulation; Figure 15 (d) in the middle is The animation simulation was conducted in real time. The effectiveness of the encirclement was refined into three evaluation indicators: encirclement uniformity, average drone movement distance, and escape distance. Figure 16 Display over time The varying values ​​of the uniformity of the enclosure. Figure 17 Display over time The changing average distance traveled by the drone. Figure 18 Display over time The changing target escape distance values ​​indicate that the invention can capture targets faster and more steadily while maintaining formation stability and reducing energy consumption, thereby reducing the target escape rate and improving capture efficiency.

[0075] The above-described embodiments are merely illustrative of the implementation methods of the present invention, but should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the protection scope of the present invention.

Claims

1. A method for coordinated drone swarm capture based on a predictive hierarchical gene regulatory network, characterized in that, The method for capturing and swarming drones includes the following steps: Step 1: Use long-term and short-term networks to predict the target trajectory, and based on the predicted target and obstacle locations, use a gene regulatory network to obtain a predicted encirclement pattern; specifically: Step 1-1: Predict the target trajectory using a Long Short-Term Memory (LSTM) network to obtain the predicted target position at the next moment; Steps 1-2: Use the predicted target position and obstacle position at the next moment as input to the gene regulatory network to generate the original morphogenetic concentration. Process the original morphogenetic concentration to calculate the comprehensive morphogenetic gradient and obtain the encirclement mode. Step 2: Generate a uniform sequence of capture points in the encirclement mode and assign tasks to the drones to obtain the target capture point determined for each drone, specifically: Step 2-1: Generate capture points evenly distributed on each capture pattern according to the perimeter ratio and length interval mechanism of the capture pattern; Step 2-2: Given the sequence of drone capture points, solve the allocation relationship between drones and capture points by combining optimization problem models, and add a stickiness factor to achieve more stable drone task allocation; Step 3: After the UAV determines its assigned capture point in Step 2, it is driven to move towards the assigned capture point using an improved artificial potential field method incorporating vortex obstacle avoidance force and damping mechanism. Specifically: Step 3-1: Calculate the four components that affect the movement of the UAV; the four components are the target traction force, the UAV mutual repulsion force, the vortex obstacle avoidance force, and the acquisition point tendency force. Step 3-2: Sum the four component forces obtained in Step 3-1 with weights to calculate the resultant force and velocity vector of the UAV.

2. The method for coordinated drone swarm capture based on a predictive hierarchical gene regulatory network according to claim 1, characterized in that, Specifically, step 1-1 is as follows: Assuming in At any given time, LSTM will store the target's position coordinates over H historical time steps. Instantaneous velocity and the relative displacement vector with respect to nearby obstacles As input, output directly Time to Continuous time velocity vector of the target Define the target's velocity vector. ,…, ;in, for The position coordinates of the target at any given time. for The position coordinates of the target at any given time. for The velocity vector of the target at any given moment. for The velocity vector of the target at any given moment. for The relative displacement vector of the target to the nearest obstacle at any given time. for The relative displacement vector of the target to the nearest obstacle at any given time. for The velocity vector of the target at any given moment. for The velocity vector of the target at any given moment. For unit time intervals, for The position coordinates of the target at any given time. for The position coordinates of the target at any given time. for The target's position coordinates at any given time; for The target's position coordinates at any given time; right The predicted velocity vectors of each step are linearly weighted to obtain the original prediction. velocity vector at time As shown in the following formula: （1） in, for The weighting factor, take This aims to balance the accuracy of near-term forecasts with the guidance of long-term trends; for The velocity vector of the target at any given moment; ,express At that moment; Introducing a temporal coherence smoothing mechanism, by... Smooth predictions are obtained by performing weighted recursive processing. velocity vector at time As shown in the following formula: （2） in, It is a smoothing factor; According to the kinematic formula ,in, Represents the coordinates of the object's final position. Indicates the initial position coordinates of the object. Represents the velocity vector of an object. Indicates the time it takes for an object to move; describes the time it takes for a target to move. The time predicted Time coordinates position This yields the predicted target position for the next moment, as shown in the following formula: （3） in, express Time coordinates; This indicates the unit travel time of the drone.

3. The method for coordinated drone swarm capture based on a predictive hierarchical gene regulatory network according to claim 2, characterized in that, Specifically, steps 1-2 are as follows: Step 1-2-1: Generate the original morphogenetic concentration within the region based on the obstacle location and the predicted target location. The region refers to the limited space of the encirclement action. During the encirclement mode generation process, firstly, the UAV uses exponential diffusion to simulate the concentration diffusion process in cells based on the location of obstacles and the predicted target location, and calculates the original morphogenetic concentration of obstacles and predicted targets in the area. The original morphogenetic concentration of obstacles and predicted targets obtained at each moment. As shown below: （4） （5） in, Indicates by the first An obstacle or predicted target in The morphogen concentration induced by time, of which , The total number of targets or obstacles; index This is the concentration diffusion factor, used to adjust the mapping relationship between distance and morphogen concentration; and They are Real-time horizontal and vertical distances of the drone to obstacles or predicted targets; Indicates a drone or obstacle; Generate the original morpheme concentration of the target or obstacle based on all detected targets or obstacles. The original morphogenetic concentration of the obstacle is denoted as The concentration of the target original form element is denoted as ; Step 1-2-2: Determine the original morphogenetic concentration of the obstacle. and target original morphogen concentration The process involves processing to generate the final coupled integrated morphological gradient field and trapping mode. The specific steps are as follows: （6） （7） （8） （9） In the formula, , and As Intermediate regulatory genes at specific time points are used to hierarchically regulate the concentration of primitive morphogens. From the target original morphogen concentration First activation threshold With regulatory factors Calculation generation, From the original morphogenetic concentration of the obstacle Second activation threshold With regulatory factors Calculation generation, As coupling and Third activation threshold With regulatory factors The gradient field of the comprehensive morphology occurs; For the sigmoid function, , and These are the parameters in the sigmoid function. It is the concentration of morphogens; It is the activation threshold; As a regulatory factor; Gradient field generated by taking the comprehensive shape Medium concentration is the preset threshold The contour lines form an adaptive closed manifold, which is called... A real-time predicted capture pattern, wherein the same preset threshold There are multiple capture modes.

4. The method for coordinated drone swarm capture based on a predictive hierarchical gene regulatory network according to claim 3, characterized in that, Specifically, step 2-1 is as follows: Step 2-1-1: Allocate the number of capture points in the encirclement mode according to the proportional fairness principle; First, calculate the number of capture points in each capture mode; for multiple capture modes with different perimeters in a multi-object scenario, use the proportional fairness criterion to calculate the number of capture points in each capture mode, i.e., allocate them to the [number of capture points]. Number of drones in each encirclement mode Based on its manifold perimeter Dynamic calculation, as shown in the following formula: （10） in, The total number of drones; For the first The manifold perimeter of a single encirclement pattern; For the number of encirclement and capture modes, Indicates the first Each encirclement mode is used sequentially. ; Given the number of capture points in each encirclement mode, for each encirclement mode, randomly select capture point positions with equal arc lengths according to the number of capture points, and all capture point positions constitute the initial capture point sequence. Step 2-1-2: Determine the capture point sequence by using clockwise sorting and defining a global reference vector; The capture points in each encirclement pattern are counted clockwise. First, the signed area formed by the capture points in each encirclement pattern in the initial capture point sequence is calculated using the Gaussian area formula. Then, the area is... Signed area formed by each encirclement pattern The following formula provides an explanation: （11） in, To be allocated to the first The number of drones in each encirclement and capture mode; Indicates the first The first encirclement mode One capture point; For capture point The x-coordinate; For capture point The ordinate; For capture point The x-coordinate; For capture point The ordinate; when hour, The value is 1, which is the capture point. Location and capture point coincide; pass The sign of the target is used to determine the order of capture points in each capture mode, unifying the order to a clockwise direction to optimize the order of the initial capture point sequence; a global reference vector is defined. The easternmost point of the location and capture mode is used as the absolute starting point; finally, the capture point sequence is determined by clockwise sorting and defining a global reference vector. Step 2-1-3: Set up a capture point smoothing mechanism to obtain a spatially continuous sequence of capture points that varies over time; Based on the number, order, and absolute starting point position of the capture points allocated in each capture mode, the coordinate positions of all capture points are calculated. A capture point smoothing mechanism is set to ensure that the spatial movement of each capture point in the capture point sequence changes continuously and smoothly over time. The capture point smoothing mechanism is shown in the following formula: （12） in, for The actual location of the point is captured at any given moment; for The actual location of the point is captured at any given moment; According to The ideal capture point is calculated based on the global reference vector and equal arc length sampling at any given moment; It is a smoothing factor to balance the actual capture point at the previous time step with the capture point calculated at this time step.

5. The method for coordinated drone swarm capture based on a predictive hierarchical gene regulatory network according to claim 4, characterized in that, Step 2-2 specifically includes: Step 2-2-1: Establish a model for the UAV task allocation and combination optimization problem; Known A drone, A capture point, defined The time allocation matrix is ,when At that time, the current planning of drones Assigned to the capture point ,in, For drone indexing, For the capture point index; the objective of the task allocation problem is to minimize the cost function of the average obstacle avoidance distance of the drone swarm. As shown in the following formula: （13） in, For obstacle avoidance distance matrix The elements include currently planned drones With capture point The Euclidean distance between them is also integrated with a penalty operator based on obstacle location to pre-exclude physically unreachable paths; The binary decision variable represents the currently planned drone. Whether assigned to the capture point 0 indicates no, and 1 indicates yes; For the current planning of drone index, , This refers to the total number of drones and the total number of capture points; For the capture point index, ; At the same time, the following allocation constraints must be met: each drone is assigned only one capture point, and each capture point is occupied by only one drone, as shown in the following formula: （14） Solving the constrained combinatorial optimization problem shown in formula (14), we obtain... Time allocation matrix ; Step 2-2-2: Add a sticky factor to the task allocation combinatorial optimization problem model to obtain a stable UAV task allocation; By setting a viscosity factor, the obstacle avoidance distance matrix in the cost function is reset, where... Obstacle avoidance distance at any time Received Time allocation status The impact, ultimately Obstacle avoidance distance matrix with viscosity factor at all times This yields the corrected obstacle avoidance distance matrix. The inertial decision-making mechanism is shown in the following formula: （15） in, It is a viscosity factor. The closer the value is to 0, the less likely it is to trigger task redistribution; Finally, based on the corrected obstacle avoidance distance matrix, the Hungarian algorithm is used to solve the global minimum weight complete matching problem, resulting in a stable UAV task allocation.

6. The method for coordinated drone swarm capture based on a predictive hierarchical gene regulatory network according to claim 5, characterized in that, Specifically, step 3-1 is as follows: Step 3-1-1: Calculate the target traction force ; （16） in, This is the concentration value at the drone's current location; It sets a threshold value, ranging from 0.65 to 0.95; The location coordinates of the drone. These are the position coordinates of the corresponding target; Step 3-1-2: Calculate the mutual repulsion force of the drones ,at this time, Indicates drone; （17） in, Indicates the mutual exclusion strength coefficient; From drones Pointing to the currently planned drone vector, drone In addition to the currently planned drones Other drones besides those; It is the vector length. This indicates that from the drone Pointing to the currently planned drone , unit vector; It is a preset radius; Step 3-1-3: Calculate the vortex obstacle avoidance force ; The design incorporates radial repulsion. and tangential force composite model The formula for the composite model is as follows: （18） in, Radial repulsion weight; The obstacle avoidance sensing radius is used; beyond this radius, the obstacle repulsive force will not affect the obstacle. The shortest Euclidean distance from the edge of the drone to the edge of the obstacle; It is a constant; For vectors Orthogonal unit normal vectors, where, This represents the displacement vector from the obstacle to the drone; The weight of the tangential force; For smoothing operators; As a direction decision factor, it determines whether the direction of the tangential force is clockwise or counterclockwise; For vectors Orthogonal tangential vectors; Step 3-1-4: Calculate the attraction force at the capture point ; （19） in, The variable is a scalar, representing the currently planned drone. Current location With assigned capture points Spatial distance; For current planning of drones The coordinates of the assigned capture point; For current planning of drones Current position coordinates; For current planning of drones Current velocity vector; , The current planned drones about The tensile coefficient and damping coefficient, where, For current planning of drones Current location With assigned capture points Spatial distance.

7. The method for coordinated drone swarm capture based on a predictive hierarchical gene regulatory network according to claim 6, characterized in that, In step 3-1: In step 3-1-2, the preset radius The range is arrive ; In step 3-1-3, the direction decision factor ,in This represents the displacement vector from the UAV to the assigned capture point; In step 3-1-4, regarding Tensile coefficient and damping coefficient , The formula is as follows: （20） （21） in, This is the deceleration boundary distance; This represents the magnitude of the traction force before deceleration. This refers to the magnitude of the traction force after deceleration. This represents the magnitude of the damping force before deceleration. The magnitude of the damping force after deceleration, where This is the repulsive force coefficient at close range.

8. The method for coordinated drone swarm capture based on a predictive hierarchical gene regulatory network according to claim 7, characterized in that, Specifically, step 3-2 is as follows: After obtaining the force field components, the drone synthesizes them. The final resultant force at any moment : （22） in, , , , Target traction force mutual repulsion of drones Vortex obstacle avoidance force Capture point tendency force The weights are applied to different forces to achieve smooth motion in complex environments. From the basic formula of kinematics , It can be known that ,in, The net force acting on the object. For the mass of the object, Let be the acceleration of the object's motion. It is the change in the velocity of the object. Let be the time of the object's motion; for each drone, the final net force acting on it has been calculated as... The mass of the drone is known to be The unit movement time of the drone is Calculate the time of a drone per unit of movement velocity change vector within, velocity change vector The calculation formula is as follows: （23） Set cutoff speed As an actual speed constraint; if the speed of the drone is calculated based on the resultant force... Exceeding the deadline speed If the desired direction remains unchanged, its size is limited to... If its size Not exceeding If the original speed remains unchanged, then the final speed can be obtained. The details are as follows: （24） That is, each drone in Always follow the calculated velocity vector Perform real-time displacement.

Images