Unmanned autonomous cluster flight control method based on bionic warning mechanism

Abstract

The invention discloses an unmanned autonomous cluster flight control method based on a bionic alert mechanism, which is applied to the field of unmanned autonomous cluster control, and is characterized in that a W-MSR algorithm and a dynamic weighted bionic alert mechanism are fused, the W-MSR filters and eliminates extreme values of neighbor individuals, the dynamic weighted bionic alert mechanism strengthens input from informed individuals, and the dynamic weighted bionic alert mechanism is used for improving the robustness of the unmanned autonomous cluster flight control. And meanwhile, the influence of suspicious individuals is inhibited, so that the local information aggregation degree is effectively improved, and group splitting can be prevented. Different from a scheme depending on explicit attacker recognition, the method can improve the motion accuracy and connectivity of the unmanned autonomous cluster in a confrontation environment by probabilistically suppressing the influence of suspicious individuals and amplifying a consistent signal, provides a new idea for improving the security of a swarm intelligence system, and has a wide application prospect. Damage of malicious individuals to the cluster is effectively defended, and robustness and recovery capability of the cluster in a complex environment are improved.

Description

Technical Field

[0001] This invention relates to the field of unmanned autonomous swarm control technology, specifically an unmanned autonomous swarm flight control method based on a biomimetic warning mechanism. Background Technology

[0002] Unmanned swarms achieve decentralized collaborative control by simulating the local interaction mechanisms of biological swarms (such as bird migration and ant foraging). This biologically inspired architecture (such as the Boid, Vicsek, and Cousin models) endows unmanned swarms with high scalability, environmental adaptability, and fault tolerance, giving them significant advantages in dynamic scenarios such as disaster relief and military reconnaissance.

[0003] However, despite being widely considered to have strong fault tolerance, unmanned swarms still face significant challenges in practical deployment. Some individuals within the swarm may fail due to malicious external interference or internal malfunctions. For example, malicious attackers can disrupt node navigation and the state updates of neighboring nodes by injecting bounded minor deviations (such as GPS spoofing, LiDAR interference, visual deception, etc.); hackers may also directly hijack some robots, manipulating them to perform misleading behaviors. Such interference disrupts the swarm's collaborative mechanisms and can even lead to overall mission failure.

[0004] Therefore, how to effectively improve the robustness of unmanned swarms in the presence of faulty or malicious individuals, and ensure that they can still stably and reliably complete collaborative tasks in complex adversarial environments, has become a core issue that needs to be addressed to move this technology out of the laboratory and into practical applications.

[0005] Existing fault detection and defense technologies against malicious individuals have the following limitations: (1) Threshold-based fault detection method: relies on preset abnormal thresholds (such as speed variance thresholds), and cannot detect malicious attackers whose deviations are below the threshold.

[0006] (2) Multi-agent safety and consistency control method: This method relies on strict network topology conditions, such as the weighted mean subsequence reduction algorithm W-MSR, which requires the system to continuously satisfy (F+1, F+1) robustness. In actual clusters, limited communication distance and dynamic obstacle avoidance behavior lead to frequent topology changes, making it difficult to guarantee this condition. More importantly, such algorithms remove extreme input values ​​that differ significantly from their own state in descending order, which cannot effectively avoid filtering out small deviations, causing interference to accumulate continuously.

[0007] (3) While encryption and economic mechanisms (such as blockchain-based verification methods) can suppress malicious behavior, they bring huge computational overhead, thus limiting their applicability to lightweight, resource-constrained unmanned clusters.

[0008] (4) Machine learning-based fault detection methods: These methods, especially supervised methods such as logistic regression and gradient-enhanced decision trees, have shown promise in detecting anomalies in swarm systems by classifying faulty agents based on observed features. However, they typically require pre-collected datasets and extensive training, and may struggle to generalize against stealth or adaptive attacks. Furthermore, their dependence on global features and computational intensity often makes them unsuitable for real-time execution in fully distributed systems. Summary of the Invention

[0009] To address the shortcomings of the existing technologies, this invention provides an unmanned autonomous swarm flight control method based on a biomimetic warning mechanism, which can effectively defend against malicious individuals from damaging the swarm and improve the swarm's robustness and resilience in complex environments.

[0010] To achieve the above objectives, this invention provides an unmanned autonomous swarm flight control method based on a biomimetic warning mechanism, wherein the unmanned autonomous swarm has at least one informed participant, which is capable of obtaining target direction information in advance. The intelligent agent is the leader, and the other intelligent agents in the unmanned autonomous cluster are followers. The robust cooperative control method for unmanned swarms includes the following steps: Step 1: Obtain the number of agents in the unmanned autonomous cluster. The interaction radius of the intelligent agent The separation radius of the intelligent agent constant velocity of the agent And let the agent iterate the parameters ; Step 2, Search for intelligent agents The set of all neighboring smart agents within the interaction area And filter intelligent agents based on extreme values. Remove the set of agents To obtain the intelligent agent The remaining set of neighboring smart agents And calculate the value of each agent in the remaining set of neighboring agents. Dynamic weights ; Step 3, Search for intelligent agents Set of agents within the exclusion region and judge Is it true or false? If so, search intelligent agent The set of cooperative neighboring intelligent agents within the interaction area and outside the exclusion area And based on the set of collaborative neighboring intelligent agents Dynamic weight update of each agent in the agent After determining the direction vector of motion, proceed to step 4; Otherwise, based on the set of intelligent agents Each agent in the process updates the agent. After determining the direction vector of motion, proceed to step 5; Step 4, determine the intelligent agent Whether or not someone was aware of the situation: If so, based on target direction information Update the intelligent agent again After determining the direction vector of motion, proceed to step 5; Otherwise, proceed to step 5; Step 5, based on intelligent agents Update the motion direction vector and update the agent. The position vector, and determine Is it true or false? If so, let Then return to step 2; Otherwise, proceed to step 6; Step 6: Determine whether the termination conditions for coordinated control have been met. If so, terminate collaborative control; Otherwise, return to step 1.

[0011] Compared with the prior art, the present invention has the following beneficial technical effects: 1. This invention integrates the W-MSR algorithm with a dynamic weighted bionic vigilance mechanism. W-MSR filters out extreme values ​​from neighboring individuals, while the dynamic weighted bionic vigilance mechanism strengthens input from informed individuals and suppresses the influence of suspicious individuals. This effectively enhances local information aggregation and prevents group splitting. Unlike schemes that rely on explicit attacker identification, this invention probabilistically suppresses the influence of suspicious individuals while amplifying consistency signals. This improves the movement accuracy and connectivity of unmanned autonomous swarms in adversarial environments, providing a new approach to enhancing the security of swarm intelligence systems. It effectively defends against malicious individuals' attacks on the swarm and improves the swarm's robustness and resilience in complex environments. 2. This invention can be widely applied to decentralized systems such as drone swarms, robot groups, and distributed sensor networks, improving their security and robustness in adversarial environments. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram of the observed feature quantities in an embodiment of the present invention; Figure 2 This is a flowchart of an unmanned autonomous swarm flight control method based on a biomimetic warning mechanism in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the trajectory accuracy of different methods in the embodiments of the present invention; Figure 4 This is a schematic diagram illustrating cluster connectivity using different methods in embodiments of the present invention; Figure 5 This is a schematic diagram of trajectory accuracy under different attack ratios in an embodiment of the present invention; Figure 6 This is a schematic diagram of cluster connectivity under different attack ratios in an embodiment of the present invention; Figure 7 This is a schematic diagram illustrating the trajectory accuracy for different cluster sizes in an embodiment of the present invention; Figure 8 This is a schematic diagram of cluster connectivity at different cluster sizes in an embodiment of the present invention; Figure 9 This is a schematic diagram of ASE with different cluster sizes in an embodiment of the present invention.

[0014] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0016] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0017] This embodiment discloses an unmanned autonomous swarm flight control method based on a biomimetic vigilance mechanism (referred to as "V-WMSR"), which can defend against malicious individuals' damage to the swarm and improve the swarm's robustness and resilience in complex environments. This method combines a weighted mean subsequence reduction (W-MSR) algorithm with a dynamically weighted bio-vigilance mechanism. The W-MSR component filters out extreme values ​​from neighboring individuals, while the dynamically weighted bio-vigilance mechanism reinforces input from informed individuals while suppressing the influence of suspicious individuals.

[0018] Classic swarm models include Boids, Vicsek, Cousin, Cucker-small, and social force models. These models are built on behavioral rules, such as repulsion, alignment, and attraction, which govern local interactions between agents. In this embodiment, the Cousin model is used, a biologically inspired framework specifically designed to simulate the migration behavior of leader-follower groups. This model enables a small group of informed individuals to guide the entire group toward a specific goal without requiring global coordination or centralized communication. This system has been applied in various scenarios, including forest exploration, swarm navigation, and omnidirectional patrolling.

[0019] In the Cousin model, consider including N An unmanned autonomous swarm of agents comprises two types: informed agents and ordinary agents. Each agent has only local interaction capabilities, meaning it can only obtain the position and velocity information of its neighboring agents within its own neighborhood. Each agent's neighborhood is defined by its communication radius. By definition, "neighbor" is interpreted as defined by communication radius. Other intelligent agents within the defined circle.

[0020] Each agent needs to maintain its own minimum movement space (radius is ). This area is called the exclusion region. If a neighboring agent appears within the exclusion region, the target agent moves in a direction away from that neighbor. The movement direction update rule is as follows:

[0021] in, For intelligent agents The updated motion direction vector, i.e., the agent exist The direction vector of motion, , respectively intelligent agents Neighbor Intelligent Agent exist The position vector at time , For intelligent agents exist Constantly exclude sets of agents within the region, that is:

[0022] In the Cousin model, repulsive motion has the highest priority, if in the agent If no neighboring agent is detected within the exclusion zone, then the agent... Towards its local interaction radius The central position of all neighboring agents moves, tending towards the average direction of movement of all neighboring agents. This behavioral rule enables the group to achieve aggregation and velocity matching. At this time, the agents... The alignment and aggregation update rule for motion direction vectors is as follows:

[0023] in, For Neighbor Intelligent Agent exist The direction vector of motion at any moment, For intelligent agents exist A set of cooperative neighboring intelligent agents within the constant interaction area and outside the exclusion area, namely:

[0024] Therefore, intelligent agents exist The set of all neighboring intelligent agents within the instantaneous interaction area for:

[0025] Will Standardize to a unit vector: .

[0026] Assuming some agents (e.g., a proportion of 1 / 3) Information about the target direction can be obtained in advance. The agents with guiding information are called informed agents. Other agents, lacking any prior information and relying solely on interactions with their neighbors to determine their direction of movement, are called followers. Informed agents are weighted... To balance target preferences and neighbor interactions, the behavioral rules are as follows:

[0027] if Then the agent's direction of movement is only affected by neighbor interactions and not by the preferred direction. Irrelevant. If Then the neighbor interaction and preference direction The influence of the agent's direction of motion has the same weight. If Then the preferred direction It has a greater impact on the direction of movement for those who are in the know.

[0028] This behavioral rule mimics the homing or migration behavior of natural clusters, enabling a group of intelligent agents to move toward a common goal under the guidance of informed individuals.

[0029] This embodiment employs a hybrid defense mechanism, V-WMSR, which integrates the W-MSR algorithm with a dynamic weighted strategy based on bio-vigilance to enhance the unmanned swarm's ability to resist interference from malicious individuals and improve the system's resilience.

[0030] The W-MSR algorithm is a multi-agent secure consensus control technique that achieves resilient consensus in a multi-agent system subjected to attacks from a limited number of malicious agents by locally filtering extreme values. Represents intelligent agents At time step state, The agent's set of neighbors (including itself). At most, there may be Several neighbors exhibiting unusual behavior, but unable to be definitively identified. Each step includes: 1) Local information collection, for each normal intelligent agent i At any moment Collect neighbor nodes State value It includes its own state; 2) Extreme value filtering: Sort neighbor state values ​​from largest to smallest; if strictly greater than... If there are fewer than F values, then all values ​​that are strictly greater than the specified value are removed; otherwise, the F largest values ​​are removed. If the values ​​are strictly less than the specified value... If there are fewer than F values, then all values ​​that are strictly less than F are removed; otherwise, the smallest F values ​​are removed. 3) Weighted update: all removed neighbors form a set. Update using the retained neighbor values, i.e.:

[0031] in, Represents intelligent agents At any moment Assigned to neighboring intelligent agents The weights; The W-MSR algorithm, a lightweight mechanism, requires only local information and no centralized coordination. However, it relies on the (F+1,F+1)-robustness condition. In real-world unmanned swarm systems, this condition is often difficult to meet due to dynamic topology, mobility, and communication loss. Under such sparse connectivity conditions, excessive filtering can lead to the isolation of effective nodes, causing group splits. To address this, this embodiment introduces an adaptive vigilance mechanism that enhances the W-MSR algorithm through behavior-aware trust adjustment. In natural groups, animals have evolved decentralized vigilance mechanisms to cope with uncertain and adversarial environments. For example, meerkats adjust their vigilance levels based on environmental stimuli and the behavior of surrounding individuals. Similarly, locusts and pigeons prioritize neighbors with consistent movement in their visual field while ignoring individuals with abnormal movement, thereby maintaining group cohesion and directional consistency.

[0032] In the Cousin-based leader-follower group framework, informed individuals typically occupy forward-facing positions—either because they possess prior information about the target location or because subtle differences in their movement characteristics naturally place them more prominent within the migrating group. For example, empirical studies have shown that in migrating bee colonies, scout bees aware of the target location tend to fly faster and higher than the group average, thus implicitly guiding the overall movement of the group. These observations suggest that individuals in forward positions often have a stronger leadership influence on group movement. Inspired by these biological principles, this embodiment proposes a vigilance-based dynamically weighted strategy that adjusts the influence of forward neighbors based on their relative orientation. This method probabilistically amplifies the influence of forward neighbors (often informed individuals) while suppressing the influence of potentially malicious individuals. The mechanism dynamically adjusts neighbor influence based on three observable features: agent velocity vector... Position vector And relative position vectors, such as Figure 1 As shown. Weights are dynamically assigned based on the relative positions of neighbors to the target agent (determined by their position relative to the target agent's heading), i.e.:

[0033]

[0034] in, For Neighbor Intelligent Agent exist Time relative to the intelligent agent Dynamic weights, For vectors and The angle between them. Representing neighboring intelligent agents Located in the intelligent agent Directly in front, at this time ; Representing neighboring intelligent agents Located in the intelligent agent Directly behind, at this time .

[0035] This directional weighting mechanism precisely prioritizes and focuses on neighbors aligned with the agent's heading, cleverly replicating the natural visual attention filtering characteristics and forward-dominant influence patterns prevalent in biological groups. By retaining input information from all neighbors and dynamically adjusting their influence, this mechanism probabilistically strengthens the guiding role of informed agents (who are typically in the lead); simultaneously, it gently suppresses anomalous inputs—rather than simply eliminating them entirely—thus effectively reducing the potential risk of group fragmentation while maintaining group cohesion.

[0036] The unmanned autonomous swarm flight control method based on a biomimetic warning mechanism in this embodiment mainly includes three stages: extreme value filtering, a weighted strategy based on a biomimetic warning mechanism, and state updating. The state updating is performed according to the Cousin model, at each time step... intelligent agent Far from the repulsion area (radius is) To avoid collisions, the neighbors of the model are considered. If there are no neighbors within the exclusion zone, the agent is limited by its interaction range. The influence of neighbors within, namely:

[0037] Assume that The swarm of intelligent agents moves at a constant speed in a two-dimensional plane. They move in initially random directions, starting from a random position where each agent can detect at least one other agent. (Reference) Figure 2 The unmanned autonomous swarm flight control method based on a biomimetic warning mechanism in this embodiment specifically includes the following steps: Step 1: Obtain the number of agents in the unmanned autonomous cluster. The interaction radius of the intelligent agent The separation radius of the intelligent agent constant velocity of the agent And let the agent iterate the parameters ; Step 2, Search for intelligent agents based on equation (5) The set of all neighboring smart agents within the interaction area And filter intelligent agents based on extreme values. Remove the set of agents Specifically: Get the set of all neighboring smart agents Various intelligent agents motion direction vector And based on the difference in motion direction vectors set of neighboring intelligent agents Sort the left and right agents in the system; If the set of neighboring smart agents The first condition is met If the number of neighboring agents is less than F, then all neighboring agents that satisfy this first condition are added to the removal agent set. Otherwise, add the F neighboring agents with the largest differences in their motion direction vectors to the set of agents to be removed. ; If the set of neighboring smart agents The second condition is met. If the number of neighboring agents is less than F, then all neighboring agents that satisfy this second condition are added to the set of agents to be removed. Otherwise, add the F neighboring agents with the smallest difference in motion direction vectors to the set of agents to be removed. ; Remove the set of agents All agents from the set of all neighboring agents By removing from the middle, we can obtain the intelligent agent. The remaining set of neighboring smart agents And based on Equation (9), calculate the value of each agent in the set of remaining neighbor agents. Dynamic weights ; Step 3, Search for intelligent agents based on equation (2) Set of agents within the exclusion region and judge Is it true or false? If so, search intelligent agents based on equation (4) The set of cooperative neighboring intelligent agents within the interaction area and outside the exclusion area And based on the set of collaborative neighboring intelligent agents Dynamic weight update of each agent in the agent After calculating the direction vector of motion, i.e., formula (10), proceed to step 4; Otherwise, based on the set of intelligent agents Each agent in the process updates the agent. After calculating the direction vector of motion, i.e., formula (1), proceed to step 5; Step 4, determine the intelligent agent Whether or not someone was aware of the situation: If so, based on target direction information Update the intelligent agent again After calculating the direction vector of motion, i.e., formula (6), proceed to step 5; Otherwise, proceed to step 5; Step 5, based on intelligent agents Update the motion direction vector and update the agent. The position vector, and determine Is it true or false? If so, let Then return to step 2; Otherwise, proceed to step 6; Step 6: Determine whether the termination conditions for collaborative control have been met, such as reaching the simulation duration or achieving group consensus convergence of the unmanned autonomous swarm, such as trajectory accuracy. Exceeding the set threshold, etc.: If so, terminate collaborative control; Otherwise, return to step 1.

[0038] It is worth noting that, although this embodiment Figure 2 The steps are shown sequentially as indicated by the arrows, but they are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order in which these steps are performed; they can be executed in other orders. Figure 2 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0039] The following section provides a further explanation of the unmanned autonomous swarm flight control method based on a biomimetic warning mechanism in this embodiment, using specific examples.

[0040] In this example, a malicious individual refers to a node that injects bounded bias into its state updates due to external attacks such as GPS spoofing, sensor interference, hacking, or internal malfunctions. A malicious individual can propagate its polluted state to neighboring nodes through the local communication topology. Its movement rules are as follows:

[0041] in, For intelligent agents The correct direction of movement is determined based on local information. For the introduced bias, As the deviation factor, For an intelligent agent exist The actual direction of motion after introducing deviations at any time.

[0042] The core objective of the cluster model is to successfully achieve cooperative migration to the target location; therefore, three performance metrics are used in the example: trajectory accuracy ( ), group connectivity ( Cluster performance is evaluated using trajectory accuracy and attack suppression efficiency (ASE). Trajectory accuracy quantifies the effectiveness of aligning the group's movement direction with the target direction, group connectivity measures the group's ability to maintain cohesion and prevent splitting during movement, and attack suppression efficiency assesses the system's ability to suppress external interference.

[0043] Trajectory accuracy quantifies the relationship between the cluster and the target orientation over time. The effectiveness of alignment, the direction of cluster motion is defined as the time of the cluster's centroid. Time The displacement vectors between the groups represent the short-term trajectory within a fixed time window. This helps mitigate the impact of short-term noise and better reflects the overall trend of the group's movement. Therefore, trajectory accuracy is achieved by projecting this group displacement vector onto the target direction. Let's calculate it. Its mathematical form is:

[0044] in, express The coordinates of the group's centroid position at any given moment. Characterization from Time's up The group centroid displacement vector at time t, and the vector dot product operation. It reflects the projected components of the group displacement in the target direction during this period. This represents the individual's baseline motion rate. , The group moves completely in the opposite direction to the target. The movement of the group is undirected (random walk pattern). The group movement is perfectly aligned with the target direction.

[0045] Assuming the group is Split into There are clusters, and the number of agents in each cluster is as follows: , … The population connectivity is defined as the probability that two agents are connected by at least one path, and is expressed by the following formula:

[0046] in, For a moment An index of clusters containing informed individuals; group connectivity measures the distribution and connectivity of clusters containing informed individuals. The larger the cluster containing the informed individual, or the fewer the number of clusters, the better. The larger.

[0047] Attack suppression efficiency is the normalized recovery rate of group function in an adversarial environment, i.e.:

[0048] in, This indicates the current system performance after the measures were taken. For the baseline scenario (attacked but no action taken), the group performance is... This indicates the performance level before the attack begins.

[0049] This example provides a comprehensive evaluation of the defensive effectiveness of the method in this embodiment under different scenarios. Simulation analysis examines the performance of the method in this embodiment under the bias factor (…). ), attack ratio ( ) and group size ( The performance of the method under changes in variables such as ( ) is analyzed. Trajectory accuracy, population connectivity, and ASE value are used as the main performance indicators. To verify the effectiveness of the method in this embodiment, the following four comparative models are selected for comparative analysis: Couzin Model (Baseline): The classic Couzin model without a defense strategy, serving as a baseline for attacks; Single-weight model: The dynamic warning model in this embodiment is based on a weighted mechanism. Divide Prevention Model (DFM): A distributed bee colony model with a two-way adaptive feedback mechanism for weights and directions; W-MSR algorithm: Weighted average subsequence reduction algorithm.

[0050] Consider a A group of agents moves on a two-dimensional cyclic plane, with each agent maintaining a constant speed. The unit is per second. Initially, the movement direction and position of each agent are randomly distributed, but it must be ensured that each agent has at least one neighbor to ensure that all agents are in the same network at the initial moment. The attack starts at [time]. (i.e., the 200th second), end time (Total duration 600 seconds). Preliminary experiments verified that this duration is sufficient for the group to recover from the chaotic state caused by the attack to a relatively stable state. Each experiment was repeated 500 times, and the average of the results was taken to minimize the impact of randomness. The main parameters are shown in Table 1.

[0051] Table 1 Simulation Parameters

[0052] To evaluate the effectiveness of different defense methods under different bias factors, the bias factor was used. A series of experiments were conducted, with the attack ratio fixed at 100%. =0.1.

[0053] Figure 3 The trajectory accuracy results are presented. After an attack, both the V-WMSR and W-MSR methods demonstrated strong anti-interference capabilities. Although initial performance declined slightly, both methods recovered quickly and continued to improve. Notably, V-WMSR outperformed the other four methods in all test scenarios after the attack. While W-MSR showed a similar recovery trend, its trajectory accuracy was slightly lower after the attack. In contrast, the single-weighted model and the anti-split model initially exhibited short-term fluctuations before gradually returning to their original values, demonstrating moderate anti-interference capabilities. The baseline method, lacking defensive mechanisms, showed persistent accuracy decay, indicating that even limited hidden biases can lead to cumulative inaccuracies without countermeasures.

[0054] Figure 4 Showing different Changes in group connectivity under a given value. With... As the value increases, connectivity decreases for all methods. The baseline method not only shows a significant decrease but also continues to decline. In contrast, the other four defense methods—single-weight model, anti-split model, W-MSR, and V-WMSR—all demonstrate stronger resistance to interference. Although connectivity initially decreases after the attack begins, it eventually stabilizes at its original level, indicating that the group achieves adaptation and coordination through self-organized interactions. Among them, V-WMSR consistently maintains the highest or near-highest group connectivity, demonstrating strong structural robustness. The slight decrease in group connectivity observed in the defense methods is actually a "self-purification" mechanism—gradually isolating severely disturbed agents, thereby allowing the remaining group to maintain coordinated operation.

[0055] Tables 2 and 3 present the quantitative results of ASE, an index used to measure the degree of standardized improvement of each method relative to the baseline. V-WMSR is present in all... The V-WMSR exhibits the highest trajectory accuracy ASE value (59.4%-82.3%) across the entire range, significantly outperforming other methods. This validates its strong suppression and adaptability against malicious agents (SAs). Regarding connectivity, as shown in Table 3, the V-WMSR's ASE value remains relatively stable, indicating consistent protection of structural integrity.

[0056] Table 2 ASE for Group Precision

[0057] Table 3 ASE of Cluster Connectivity

[0058] Furthermore, the performance of V-WMSR under different attacker ratios was analyzed. All simulations used a bias factor set to... =1 / 4, size is A group of 100. When the attack time reaches... When the threshold is 1000Δt, the system will simultaneously activate all defense strategies and launch a persistent, covert attack. Attacker ratio Gradually increase from 0.1 to 0.5. For example... Figure 5 and Figure 6 As shown, the performance of each method is analyzed from two dimensions: trajectory accuracy and group connectivity.

[0059] like Figure 5 As shown, V-WMSR exhibits strong anti-interference capabilities against stealth attacks. After a brief performance degradation, the system quickly recovers stability, even when the attack ratio... It even surpassed the accuracy before the attack. The V-WMSR not only has the fastest recovery speed, but also... Maintain the highest accuracy within the value range, even when It can still maintain stable performance. Although W-MSR shows a similar recovery trend, its accuracy is consistently lower than that of V-WMSR.

[0060] Conversely, single-weighted models and anti-split models only... It can maintain acceptable performance at times; but when At times, its accuracy drops sharply. It's worth noting that in scenarios with a high attack ratio, the Cousin model, lacking defensive mechanisms, suffers a significant decline in accuracy. As the value increases, accuracy continues to deteriorate until it falls into negative territory, indicating that the group has lost coordination and is being manipulated by an attacker. The anti-split model performs significantly worse than the baseline method—this is because its design primarily focuses on the internal dynamics between informed parties and followers, without considering the impact of external adversarial interference.

[0061] Figure 6 This displays the corresponding group connection trends. (As the attack ratio...) As the value increases, the performance of all methods decreases, but the W-MSR method shows the most significant performance drop. It is worth noting that when... At that time, the connectivity of W-MSR decreased sharply, and in The time is close to zero, indicating that over-filtering leads to network fragmentation. In contrast, V-WMSR in all cases... It maintains the highest cluster connectivity under all conditions because it enhances the weights of trusted agents by applying an adaptive weight filtering mechanism.

[0062] Overall, the V-WMSR hybrid strategy proposed in this embodiment combines the advantages of W-MSR and the adaptive weighting mechanism: W-MSR can effectively eliminate extreme outliers, while the dynamic weighting bionic vigilance mechanism enhances group cohesion by preserving key connections. This synergistic effect enables V-WMSR to effectively suppress covert attacks, maintain group connectivity, and sustain high accuracy even in severe adversarial environments.

[0063] Table 4. ASE Trajectory Accuracy under Different Attack Ratios

[0064] Table 5. Cluster connectivity ASE under different attack ratios

[0065] Tables 4 and 5 show the ASE experimental results, which quantitatively verify the above intuitive observations: even in In the case of [unclear context], the V-WMSR method continues to maintain its advantage in trajectory accuracy and exhibits superior group connectivity performance. In stark contrast, W-MSR [unclear context] in high [unclear context]... The performance degradation under certain conditions exposed the limitations of its hard filtering mechanism, while V-WMSR successfully avoided this defect thanks to its adaptive design.

[0066] To evaluate the scalability of the V-WMSR method, different population sizes were considered. Extensive simulation experiments were conducted using attack parameters of 50, 100, 150, and 200. =1 / 4 and =0.1.

[0067] like Figure 7 As shown, V-WMSR consistently outperforms the baseline method in trajectory accuracy across populations of varying sizes. While the performance of the baseline method gradually declines with increasing agent numbers, V-WMSR maintains stable and accurate trajectory alignment. It is noteworthy that as the number of agents increases... As the value increases, the performance gap between V-WMSR and the benchmark method widens further, indicating that the method has a scale advantage in terms of coordination efficiency.

[0068] Figure 8The study demonstrates the corresponding trend in group connectivity. The baseline method exhibits increased network fragmentation in larger groups. In contrast, V-WMSR effectively maintains network connectivity even as group size increases, demonstrating its ability to maintain a robust connection structure under adverse conditions. This improvement is particularly important for large-scale decentralized systems—where maintaining network connectivity is crucial for achieving collective intelligence.

[0069] Table 6 and Figure 9 This demonstrates the improvement in standardized performance reflected by the ASE metric. The ASE value for trajectory accuracy steadily increases with increasing population size, from... =50, 49.1% increased to The ASE value at 200 is 76.7%, which fully demonstrates the algorithm's superior performance in larger-scale systems. Although the ASE value of group connectivity also improves with increasing group size, the increase is relatively gradual, eventually reaching [a certain value]. = 200, reaching 45.0%.

[0070] Table 6. Trajectory accuracy and cluster connectivity ASE at different cluster sizes

[0071] The above description is only a preferred embodiment of the present invention and does not limit the scope of protection of the present invention. All equivalent structural transformations made under the inventive concept of the present invention using the contents of the present invention specification and drawings, or direct / indirect applications in other related technical fields, are included within the scope of protection of the present invention.

Claims

1. A method for controlling unmanned autonomous swarm flight based on a biomimetic warning mechanism, characterized in that, The unmanned autonomous swarm has at least one informed participant, who is someone who can obtain target direction information in advance. The intelligent agent is the leader, and the other intelligent agents in the unmanned autonomous cluster are followers. The robust cooperative control method for unmanned swarms includes the following steps: Step 1: Obtain the number of agents in the unmanned autonomous cluster. The interaction radius of the intelligent agent The separation radius of the intelligent agent constant velocity of the agent And let the agent iterate the parameters ; Step 2, Search for intelligent agents The set of all neighboring smart agents within the interaction area And filter intelligent agents based on extreme values. Remove the set of agents To obtain the intelligent agent The remaining set of neighboring smart agents And calculate the value of each agent in the remaining set of neighboring agents. Dynamic weights ; Step 3, Search for intelligent agents Set of agents within the exclusion region and judge Is it true or false? If so, search intelligent agent The set of cooperative neighboring intelligent agents within the interaction area and outside the exclusion area And based on the set of collaborative neighboring intelligent agents Dynamic weight update of each agent in the agent After determining the direction vector of motion, proceed to step 4; Otherwise, based on the set of intelligent agents Each agent in the process updates the agent. After determining the direction vector of motion, proceed to step 5; Step 4, determine the intelligent agent Whether or not someone was aware of the situation: If so, based on target direction information Update the intelligent agent again After determining the direction vector of motion, proceed to step 5; Otherwise, proceed to step 5; Step 5, based on intelligent agents Update the motion direction vector and update the agent. The position vector, and determine Is it true or false? If so, let Then return to step 2; Otherwise, proceed to step 6; Step 6: Determine whether the termination conditions for coordinated control have been met. If so, terminate collaborative control; Otherwise, return to step 1.

2. The unmanned autonomous swarm flight control method based on a biomimetic warning mechanism according to claim 1, characterized in that, In step 2, the intelligent agent is selected based on extreme value filtering. Remove the set of agents Specifically: Get the set of all neighboring smart agents Various intelligent agents motion direction vector And based on the difference in motion direction vectors set of neighboring intelligent agents Sort all agents in the system, where... For intelligent agents The direction vector of motion; If the set of neighboring smart agents The first condition is met If the number of neighboring agents is less than F, then all neighboring agents that satisfy this first condition are added to the removal agent set. Otherwise, add the F neighboring agents with the largest differences in their motion direction vectors to the set of agents to be removed. ; If the set of neighboring smart agents The second condition is met. If the number of neighboring agents is less than F, then all neighboring agents that satisfy this second condition are added to the set of agents to be removed. Otherwise, add the F neighboring agents with the smallest difference in motion direction vectors to the set of agents to be removed. .

3. The unmanned autonomous swarm flight control method based on a biomimetic warning mechanism according to claim 1, characterized in that, In step 2, the dynamic weights The calculation process is as follows: in, , respectively intelligent agents exist The direction vector and position vector at any given moment For Neighbor Intelligent Agent exist The position vector at time , For vectors and The angle between them.

4. The unmanned autonomous swarm flight control method based on a biomimetic warning mechanism according to claim 1, characterized in that, In step 3, the set of cooperative neighbor intelligent agents... Dynamic weight update of each agent in the agent The direction vector of motion is specifically: in, For intelligent agents The updated motion direction vector, i.e., the agent exist The direction vector of motion, For intelligent agents exist The position vector at time , , For Neighbor Intelligent Agent exist The direction vector and position vector of motion at any given moment.

5. The unmanned autonomous swarm flight control method based on a biomimetic warning mechanism according to claim 4, characterized in that, In step 4, the target direction information is used as a basis for... Update the intelligent agent again The direction vector of motion is specifically: in, This is the updated motion direction vector. The motion direction vector before the next update. These are the preference weights.

6. The unmanned autonomous swarm flight control method based on a biomimetic warning mechanism according to claim 1, characterized in that, In step 3, the intelligent agent set Each agent in the process updates the agent. The direction vector of motion is specifically: in, For intelligent agents The updated motion direction vector, i.e., the agent exist The direction vector of motion, For intelligent agents exist The position vector at time , For Neighbor Intelligent Agent exist The position vector at any given time.

Images