Cooperative defense and attack behavior cognition method for multi-robot system
By constructing a collaborative defense game model and distributed equilibrium solution in a multi-robot system, the problems of load imbalance and unstable defense of the multi-robot system under denial-of-service attacks are solved. This enables interpretable cognition and prediction of attack behavior and improves the system's collaborative task latency robustness and defense effectiveness.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SUN YAT SEN UNIVERSITY SHENZHEN
- Filing Date
- 2026-01-29
- Publication Date
- 2026-05-08
AI Technical Summary
Existing multi-robot systems lack the ability to recognize and predict attack behavior when facing denial-of-service attacks, resulting in insufficient autonomous collaborative optimization decision-making capabilities. Furthermore, existing defense solutions are difficult to adapt to the characteristics of multi-source traffic, heterogeneous computing power, dynamic topology, and time-varying task load, leading to unbalanced load and unstable defense.
By defining defense nodes, edge filtering nodes, and cloud filtering nodes in a multi-robot system, a collaborative defense game model is constructed. A distributed equilibrium solution method is adopted to achieve stable collaborative filtering allocation. Furthermore, by solving the attacker's strategy, early warning and defense resources are proactively configured, thus explicitly modeling the interpretable cognition and prediction of attack behavior.
It achieves stable collaborative filtering allocation under heterogeneous edge computing power and resource sharing conditions, reduces filtering congestion, improves the latency robustness of multi-robot collaborative tasks, reduces communication and computing overhead, is suitable for 5G multi-robot dynamic topology scenarios, and effectively resists distributed denial-of-service attacks.
Smart Images
Figure CN122002293A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of multi-robot cooperative control technology, and in particular to a method for recognizing cooperative defense and attack behaviors in multi-robot systems. Background Technology
[0002] Multi-robot systems (such as drone swarms, ground mobile robot formations, and warehouse logistics robot groups) typically rely on 5G networks to achieve low-latency, high-bandwidth collaborative perception, collaborative decision-making, and collaborative behavior control. They are often combined with edge computing platforms to carry out computationally intensive functions such as task offloading, collaborative localization, map building, and model inference.
[0003] However, the open access and high concurrency of 5G multi-robot systems also make them more vulnerable to denial-of-service attacks. Attackers can use external botnets or compromised robot nodes to continuously send a large number of invalid requests / traffic to the 5G access network, edge nodes, or collaborative control services, causing congestion of critical control links, exhaustion of edge computing resources, increased latency or even interruption of collaborative tasks, thereby leading to formation instability, task failure, or security incidents.
[0004] Existing DDoS defense solutions mostly rely on centralized scrubbing or static threshold rate limiting, which are difficult to adapt to the characteristics of multi-robot systems, such as "multi-source traffic, heterogeneous computing power, dynamic topology, and time-varying task load". At the same time, existing solutions often lack interpretable cognition and prediction mechanisms for attack behavior, which also limits the autonomous decision-making ability of multi-robot systems in complex adversarial environments. Summary of the Invention
[0005] In view of this, in order to address the technical problem that existing multi-robot systems lack a process for recognizing and predicting attack behavior, which hinders autonomous collaborative optimization decision-making in multi-robot systems, this invention proposes a collaborative defense and attack behavior recognition method for multi-robot systems. This method includes the following steps: System modeling involves defining a set of defense nodes, a set of edge filtering nodes, and a set of cloud filtering nodes within a multi-robot system; obtaining the filtering capabilities of each edge filtering node and the cloud transmission latency coefficient; and statistically analyzing the suspicious traffic intensity of each defense node.
[0006] Collaborative defense game modeling: Define a collaborative filtering strategy for each defense node; define the filtering delay of edge filtering nodes and construct the individual cost function of defense nodes; each defense node forms a selfish filtering game with the goal of minimizing its own cost, and obtain the Nash equilibrium filtering strategy.
[0007] Distributed load balancing solution: Utilizing the property of consistent filtering delay among edge filtering nodes under load balancing conditions, a common filtering delay parameter is introduced and a distributed iterative algorithm is designed; each defense node only needs to update local variables based on its own traffic intensity and broadcast information to converge to Nash equilibrium within a finite number of iterations, thereby achieving collaborative filtering with load balancing. Attacker strategy solution: obtaining a balanced defense strategy Then, an optimization model is constructed to maximize the total filtering cost of the system under budget constraints. and in the attack feasible domain ,pass It enables the recognition and prediction of attack intensity, target preferences, and attack types, providing early warning and proactive allocation of defense resources for multi-robot systems.
[0008] Based on the above scheme, this invention provides a collaborative defense and attack behavior cognition method for multi-robot systems. Under heterogeneous edge computing power and resource sharing conditions, it achieves stable collaborative filtering allocation, significantly reduces filtering congestion, and improves the latency robustness of multi-robot collaborative tasks. It employs distributed equilibrium solution to reduce dependence on the global network state, lowering communication and computational overhead, making it suitable for 5G multi-robot dynamic topology scenarios. Through explicit attacker optimization modeling, it achieves interpretable cognition and prediction of attack behavior, which is beneficial for autonomous collaborative optimization decision-making and behavioral cognition research in adversarial environments. This invention performs excellently in extensive simulation experiments, achieving the lowest filtering cost compared to benchmark algorithms, demonstrating that its defense method can effectively resist distributed denial-of-service attacks in networks. Attached Figure Description
[0009] Figure 1 This is a flowchart of the steps of a collaborative defense and attack behavior recognition method for multi-robot systems according to the present invention; Figure 2 This is a schematic diagram illustrating the application scenario of the present invention; Figure 3 This is a schematic diagram of the game modeling structure in a specific embodiment of the present invention; Figure 4 This is a schematic diagram comparing the total filtering cost of the present invention with other methods under different cloud transmission latency coefficients; Figure 5 This is a schematic diagram comparing the total filtering cost of the present invention with other methods under different total filtering capacities of edge servers; Figure 6 This is a schematic diagram showing the filtering costs of different types of defenders under different cloud transmission latency coefficients according to the present invention; Figure 7 This is a schematic diagram showing the filtering costs of different types of defenders under different total filtering capacities of edge servers according to the present invention. Figure 8 This is a schematic diagram comparing the number of iterations required for the algorithm to converge under different defense node sizes. Detailed Implementation
[0010] In addition to the problems mentioned in the background, existing solutions often overlook the attacker's strategy and the resource competition and self-interest among defense nodes, leading to issues such as load imbalance, local optima, or unstable defense in resource-sharing scenarios. This invention, without relying on centralized global optimization, models the multi-robot-edge-cloud collaborative filtering process as a collaborative attack-defense game. It obtains a stable collaborative defense strategy through distributed equilibrium solving and further solves for the attacker's optimal attack strategy under budget constraints, which is used for attack behavior recognition and prediction.
[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0012] It should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this application can be combined with each other.
[0013] It should be understood that the terms "system," "apparatus," "unit," and / or "module" used in this application are a method of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.
[0014] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "a," and / or "the" are not specifically singular and may include the plural. Generally, the terms "comprising" and "including" only indicate the inclusion of expressly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements. An element defined by the phrase "comprising an..." does not exclude the presence of other identical elements in the process, method, product, or apparatus that includes the element.
[0015] In the description of the embodiments of this application, "a plurality of" refers to two or more. The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.
[0016] Furthermore, flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, the steps can be processed in reverse order or simultaneously. Additionally, other operations can be added to these processes, or one or more steps can be removed from them.
[0017] Reference Figure 1 This is a flowchart illustrating an optional example of the collaborative defense and attack behavior recognition method for multi-robot systems proposed in this invention. The method can be applied to computer devices, and the method proposed in this embodiment may include, but is not limited to, the following steps: Step S1: Obtain network and computing resource information of the multi-robot system, and define defense nodes, edge filtering nodes, and cloud filtering nodes; Step S1 further includes obtaining the filtering capability parameters of each edge filtering node and obtaining the transmission latency coefficient of a unit traffic uploaded to the cloud filtering node.
[0018] Step S2: Estimate the intensity of suspicious traffic to the defense nodes; Step S3: Construct collaborative filtering policy variables for each defense node; Step S4: Calculate the filtering delay of each edge filtering node based on the collaborative filtering strategy variables, and construct the individual filtering cost function of the defense node; The filtering latency is determined by the total amount of suspicious traffic and filtering capacity allocated to the edge filtering nodes by each defense node, and the individual filtering cost function includes edge-side filtering cost and cloud-side filtering cost. Step S5: Under constraints, each defense node forms a selfish filtering game with the goal of minimizing its own individual filtering cost function. Taking advantage of the load balancing property of the consistency of filtering delay of edge filtering nodes under equilibrium conditions, a common filtering delay parameter is introduced. The network management entity and each defense node interact to calculate the Nash equilibrium collaborative filtering strategy in a distributed iterative manner. Step S6: According to the Nash equilibrium collaborative filtering strategy, the edge filtering node and the cloud filtering node work together to perform suspicious traffic diversion and filtering; Step S7: After obtaining the Nash equilibrium collaborative filtering strategy, construct an optimization model for the attacker to maximize the total filtering cost of the system under the attack budget constraint and solve it to obtain the optimal attack strategy. Step S8: Based on the optimal attack strategy, output the behavioral perception results and early warning information on attack intensity, attack target preference and attack type.
[0019] In some feasible embodiments, steps S1 and S2, in this embodiment, are geared towards a multi-robot autonomous collaborative system based on 5G communication, such as... Figure 1 As shown. The system includes: robot terminals (drones, ground mobile robots, or warehouse robots, etc.), 5G base stations and core networks, edge computing nodes, and cloud-based security cleaning / analysis services. The robot terminals typically access the network as 5G user equipment, running collaborative sensing and decision-making applications, and carrying low-latency control messages and high-throughput sensing data through control and data links respectively. In the defense system, the set of defense nodes is defined as... The set of edge filtering nodes is The cloud node is denoted as Edge filtering node It has filtering capabilities This represents the maximum traffic volume that can be used to complete deep packet inspection, feature extraction, or malicious request identification per unit time. Cloud transmission latency coefficient. Used to describe the transmission and queuing latency introduced when a unit of traffic is uploaded to the cloud via the 5G core network / backhaul link. Defense Node Generate or receive suspicious traffic intensity that needs to be detected / filtered This includes normal traffic. With malicious traffic ,Right now .
[0020] Malicious traffic The estimation includes: using traffic measurement information from the 5G access network or core network and end-side communication status information to calculate features such as connection growth rate, packet loss rate, retransmission rate, queue length, request interval distribution, or source address distribution.
[0021] In some feasible embodiments, steps S3 and S4 specifically include: Edge filtering node The filtering delay is defined as: Among them, the molecule is the node. The total suspicious traffic carried within the current time window, with its filtering capacity as the denominator; For defense nodes The cost of edge-side filtering is: The cost of cloud-side filtering is: Therefore, defense nodes The total cost is: Each defense node forms a selfish filtering game with the goal of minimizing its own cost, and seeks the Nash equilibrium filtering strategy. .
[0022] The cost function can be interpreted as follows: edge-side costs mainly reflect congestion latency caused by shared filtering resources, while cloud-side costs mainly reflect transmission and queuing latency caused by the upload link. For the differences in different services (control messages, state synchronization, video / laser point cloud backhaul) in a multi-robot system, it can be... Traffic can be segmented into multiple categories, and different weighting coefficients can be assigned to different categories, or different weighting coefficients can be assigned to different categories. Priority is given to the control link. The above extensions do not change the solution framework of this invention.
[0023] In some feasible embodiments, in step S5, the load balancing properties include: Nash equilibrium collaborative filtering strategy Under these conditions, the filtering delay of each edge filtering node is equal, that is, for any... All meet ,in The common filtering delay is denoted as .
[0024] This invention models the interaction between the defender and the attacker as a two-phase dynamic game, such as Figure 3 As shown. The defender's objective: To effectively mitigate DDoS attacks, the goal of the multi-bot system is to minimize the overall filtering cost in the system, i.e., the traffic filtering cost generated by all defenders. The optimization problem is modeled as follows: Attacker Objective: To disrupt the robotic system and computing resources in the edge cloud network, the attacker aims to maximize the total filtering cost across the system by sending malicious traffic. The attacker's own malicious traffic has a cap. Define the total system cost as The attacker's optimization problem can then be modeled as follows: Cooperative attack and defense game and equilibrium solution: Given an attack strategy Subsequently, each defense node minimizes its own cost. Choose for the goal This leads to a selfish filtering game. A Nash equilibrium is a strategy combination where each defender adopts its optimal response strategy. Under a Nash equilibrium, no defender will unilaterally change its strategy.
[0025] Prove that this selfish filtering game has the important property of load balancing: the filtering latency of all edge servers is equal, denoted as . Based on this, we further proved With Nash equilibrium There is a one-to-one correspondence, and the Nash equilibrium is unique.
[0026] Based on the above properties, this embodiment designs a distributed Nash equilibrium calculation method, which specifically includes the following steps: Each edge server broadcasts its filtering capability parameters to the defense nodes in the network. After acquiring the filtering capabilities of all edge servers, each defense node... Based on its total suspicious flow Calculate its initial time delay impact parameters. And send the parameter to the network management entity.
[0027] The network management entity bases its decisions on the initial impact parameters of each defense node. The defense nodes are sorted, and the initial unified filtering latency of the system is calculated accordingly. The filtered latency value is then broadcast to all defense nodes. Based on this, the system enters an iterative update process.
[0028] Let the first The system filtering latency of round iteration is When satisfied At this point, the algorithm enters a loop, where It is a pre-set threshold.
[0029] In each iteration, the defense nodes update their impact on system filtering latency sequentially according to the aforementioned sorting order. Specifically, for any defense node... In its first In the round of iteration, the current filtering delay is used as a basis. Calculate intermediate variables .
[0030] Subsequently, the defense node updates its latency impact parameters according to the following piecewise function: The defense node will be updated Send to the network management entity.
[0031] After receiving the update information from the defense nodes, the network management entity combines the impact parameters of the defense nodes that have been updated and those that have not yet been updated to correct the system filtering latency. The formula for correcting the system filtering delay is as follows: in Indicates a defense node Position in the sort.
[0032] Once the iterative process converges, each defense node uses the final obtained filtering delay. and based on With Nash equilibrium The meaning mapping relationship is used to calculate the traffic filtering ratio on the cloud and each edge server, thereby obtaining the corresponding Nash equilibrium filtering strategy, and based on this, the traffic filtering operation is executed collaboratively on the edge server and the cloud.
[0033] In some feasible embodiments, step S6, the execution of the Nash equilibrium collaborative filtering strategy includes: The network management entity allocates a proportion of cloud resources to each defense node. Convert to 5G-side traffic offloading or routing rules to redirect traffic to cloud filtering nodes. And assign a proportion to the edge of each defense node. Traffic forwarding rules are converted to edge-side filtering instances to direct traffic to the corresponding edge filtering nodes. .
[0034] In some feasible embodiments, steps S7-S8 specifically include: To disrupt computing resources in robotic systems and edge cloud networks, attackers aim to maximize the total filtering cost across the system by sending malicious traffic. This is achieved through a balanced defense strategy. Below, the total system cost is defined as The attacker maximizes within the feasible region. Obtain the optimal attack strategy .
[0035] This embodiment demonstrates the existence of an optimal attack strategy characterized by making the total traffic received by each node as "uniform" as possible. That is, a defender with low normal traffic before the attack will still have low total traffic after the attack. Therefore, in practical implementation, this maximization problem can be decomposed into several subproblems: enumerating a finite number of key combinations related to the ranking of the defender's total traffic, each subproblem becomes a sequential quadratic programming problem (SQP) under a fixed combination, and selecting the solution from all subproblems that maximizes the total traffic received by each node. The largest solution is .
[0036] Based on the overall process of the above method, this invention also provides specific simulation examples: To verify the collaborative anti-denial-of-service defense effect of this invention in a 5G multi-robot autonomous collaborative scenario, this embodiment constructs a simulation environment of "robot terminal - 5G access network - edge node - cloud filtering center". Each robot terminal acts as a defense node, generating suspicious traffic and routing it according to the collaborative filtering strategy calculated by this invention; edge filtering nodes share limited filtering capabilities and generate queuing / processing latency; cloud filtering nodes introduce transmission latency coefficients related to 5G backhaul. Simulation evaluation indicators include: 1) total system filtering cost (the sum of filtering costs of all defense nodes), used to measure overall defense efficiency; 2) parameter sensitivity, exploring the relationship between filtering cost and system parameters. and The relationship between the following: 3) The number of iterations and convergence accuracy of the distributed equilibrium solution algorithm are used to measure real-time response capability and scalability.
[0037] Parameter settings and coverage: 1) Baseline comparison experiment (verifying overall superiority): Set the number of defense nodes and edge filtering nodes to 10 (corresponding to 10 collaborative robots or 10 robot clusters / edge nodes). The suspicious traffic intensity of each defense node within the time window follows a uniform distribution of [1000, 2000] to cover different robot task loads and anomaly injection intensities. The filtering capability of the edge filtering nodes... Randomly generated within the range [0,200] to simulate heterogeneous computing power and load fluctuations, and latency coefficients transmitted in the cloud. The value is selected within the range of s / bit to cover different 5G backhaul conditions of "better / poorer cloud access". Furthermore, in a fixed... At s / bit, the total filtering capacity of the system will be... The range is [0, 5000] to cover edge-side deployment scales from resource-scarce to resource-abundant; 2) Parameter sensitivity experiment: The number of defense nodes is set to 10, and the total traffic of each defense node is respectively bits. We fix them separately. bits / s change and fixed s / bit change To assess the impact of these two parameters; 3) Scalability verification experiment (verifying real-time performance): at different defense node scales The simulation is repeated (the total traffic to each defense node still follows a uniform distribution of [1000, 2000]), the filtering capacity of each edge filtering node is set to 200, and the cloud transmission latency coefficient is set to... s / bit, convergence tolerance set to The process was repeated 10 times independently for each scale, and the mean and fluctuation range of the number of iterations were statistically analyzed to evaluate the convergence speed of the distributed equilibrium solution.
[0038] Comparison Method: To demonstrate the advantages of this invention, four comparative baselines were set in the simulation: 1) Non-cooperative - Partial Filtering (NCP): Each defense node only allocates resources proportionally between its own edge and the cloud, unable to utilize the remaining filtering capabilities of other edge nodes; 2) Non-cooperative - Non-partial Filtering (NCNP): Each defense node can only choose between its own edge and the cloud for full filtering; 3) Greedy - Partial Filtering (GP): Allows each defense node to partially allocate resources among any computing nodes, but minimizes its own cost sequentially, making it susceptible to the influence of order and instantaneous resource status; 4) Greedy - Non-partial Filtering (GNP): Allows each defense node to select one computing node for full filtering, and minimizes its own cost sequentially. The above baselines represent typical solutions that "lack collaborative resource sharing capabilities," "lack of load balancing flexibility," and "lack of stable equilibrium guarantees," respectively.
[0039] Experimental Results and Effects: 1) Overall Superiority: such as Figure 4 and Figure 5 As shown, the transmission latency coefficient in the cloud Within a range where cloud backhaul costs are higher and edge computing has a greater advantage, the distributed collaborative defense mechanism of this invention achieves the lowest total system filtering cost, outperforming baseline methods such as NCP, NCNP, GP, and GNP. This is because this invention can fully utilize the total filtering capacity of all edge filtering nodes under balanced conditions and make the filtering latency of each edge node tend to be consistent, thereby avoiding single-point congestion and reducing overall queuing latency. When taking a smaller value (lower cloud backhaul cost, more advantages of cloud cleaning), the non-cooperative but cloud-biased baseline method may be close to optimal in terms of total filtering cost; however, this invention can still achieve suboptimal or even near-optimal total filtering cost, and maintains more stable results under different random filtering capability samples, demonstrating robustness to "time-varying load and heterogeneous computing power" of multi-robot systems; 2) Parameter sensitivity and deployment guidance: such as Figure 6 and Figure 7 As shown, with As the system's total filtering capacity increases, simulations show that more defense nodes transition to edge-only filtering or increase the edge filtering ratio under balanced conditions, demonstrating that this invention can adapt to changes in 5G backhaul quality; As the edge filtering capacity increases, the overall filtering cost of defense nodes decreases, and more nodes tend to complete filtering at the edge, indicating that increasing edge filtering capabilities can directly improve collaborative defense performance and reduce cloud dependence. These patterns provide a quantitative basis for the deployment scale of edge nodes and cloud-based linkage strategies in multi-robot systems under different network conditions; 3) Real-time performance and scalability: such as Figure 8 As shown, at different defense node sizes The distributed equilibrium solution algorithm of the present invention can converge within a finite number of iterations, and the average number of iterations increases with... Slow growth and less than Furthermore, the results showed that the invention maintained a small fluctuation range in 10 independent repeated experiments. This demonstrates that the present invention can maintain a rapid response even when the scale of multi-robot swarms expands, meeting the requirements of 5G multi-robot cooperative control for "short time windows and rapid reconfiguration".
[0040] In summary, the simulation settings cover key factors such as cloud transmission latency, total edge computing power, and the number of nodes. The simulation results consistently verify that the present invention has lower overall cost, more stable defense effect, and better scalability compared to existing baseline solutions in typical 5G multi-robot collaborative scenarios.
[0041] A collaborative defense and attack behavior recognition system for multi-robot systems includes: The modeling module is used to execute step S1; The flow awareness module is used to execute steps S2 and S3; The parameter acquisition module is used to perform the filter delay calculation in step S4; The game modeling module is used to perform cost function modeling in step S4. The equilibrium solution module is used to execute step S5; The rule distribution and filter execution module is used to execute step S6; The behavioral cognition module is used to execute step S7.
[0042] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.
[0043] A collaborative defense and attack behavior recognition device for multi-robot systems: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the collaborative defense and attack behavior recognition method for multi-robot systems as described above.
[0044] The content of the above method embodiments is applicable to the device embodiments. The specific functions implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.
[0045] A storage medium storing processor-executable instructions, which, when executed by a processor, are used to implement a collaborative defense and attack behavior recognition method for multi-robot systems as described above.
[0046] The content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.
[0047] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.
Claims
1. A method for recognizing collaborative defense and attack behaviors in multi-robot systems, characterized in that, Includes the following steps: Acquire resource information for the multi-robot system and define defense nodes, edge filtering nodes, and cloud filtering nodes; The intensity of suspicious traffic at the defense nodes is estimated, and collaborative filtering strategy variables are constructed. The filtering latency of the edge filtering node is calculated based on the collaborative filtering strategy variables, and the individual filtering cost function of the defense node is constructed. The defense node aims to minimize its own individual filtering cost function. By utilizing the load balancing property of the consistent filtering delay of edge filtering nodes under balanced conditions, a common filtering delay parameter is introduced, and interactive calculation is performed in a distributed iterative manner to obtain the Nash equilibrium collaborative filtering strategy. According to the Nash equilibrium collaborative filtering strategy, the edge filtering node and the cloud filtering node collaboratively perform suspicious traffic diversion and filtering. Based on the Nash equilibrium collaborative filtering strategy, an optimization model is constructed to maximize the total filtering cost of the system under the attack budget constraint, and the optimal attack strategy is obtained by solving the model. Based on the optimal attack strategy, the system outputs behavioral perception results and early warning information regarding attack intensity, target preference, and attack type.
2. The method for recognizing collaborative defense and attack behavior in multi-robot systems according to claim 1, characterized in that, The formula for calculating the intensity of the suspicious flow is as follows: in, Indicates a defense node Suspicious traffic intensity Indicates a defense node Normal traffic, Indicates a defense node Malicious traffic.
3. The method for recognizing collaborative defense and attack behavior in multi-robot systems according to claim 2, characterized in that, The collaborative filtering strategy variables are represented as follows: in, Represents the collaborative filtering strategy variable. Indicates assignment to edge filter nodes proportion, This indicates the proportion allocated to the cloud. This represents the set of edge filtering nodes.
4. The method for recognizing collaborative defense and attack behavior in multi-robot systems according to claim 3, characterized in that, The formula for the individual cost function is expressed as follows: in, Including the defender Filtering strategies for other defenders, This indicates the attacker's attack strategy. Represents edge filtering nodes Filtering delay, , This represents the cloud transmission latency coefficient.
5. The method for recognizing collaborative defense and attack behavior in multi-robot systems according to claim 2, characterized in that, The process of interactive computation using a distributed iterative approach specifically includes: Each edge filtering node broadcasts its filtering capability parameters to the defense nodes; The defense node calculates the initial latency impact parameters based on the intensity of its suspicious traffic and sends them to the network management entity; The network management entity sorts the defense nodes according to the initial delay impact parameters of each defense node and calculates the initial common filtering delay of the system. In the During rounds of iteration: The network management entity broadcasts the current common filtering delay to the defense nodes that are currently awaiting updates; The defense node calculates intermediate variables based on the current common filtering latency, updates its latency impact parameters according to the piecewise function based on the intermediate variables, and then sends them back to the network management entity. After receiving the update information from the defense nodes, the network management entity corrects the common filtering delay by combining the delay impact parameters of the updated and unupdated defense nodes, and obtains the common filtering delay for the next round. The process iterates until the co-filtering delay in the next round meets the convergence condition, then stops and outputs the Nash equilibrium co-filtering strategy.
6. The method for recognizing collaborative defense and attack behavior in multi-robot systems according to claim 5, characterized in that, The update formula for the time delay impact parameter is expressed as follows: in, Indicates a defense node Arrived The time delay of the round iteration affects the parameters. Indicates intermediate variables. Indicates a defense node Normal traffic, Indicates a defense node malicious traffic, Represents edge filtering nodes Filtering capacity Indicates cloud transmission latency coefficient Indicates the first System filtering latency during round iteration Indicates the first The time delay of round iteration affects parameters.
7. A collaborative defense and attack behavior recognition system for multi-robot systems, characterized in that, include: The modeling module is used to acquire resource information of a multi-robot system and define defense nodes, edge filtering nodes, and cloud filtering nodes. The traffic awareness module is used to estimate the intensity of suspicious traffic at the defense node and construct collaborative filtering strategy variables; The parameter acquisition module calculates the filtering delay of the edge filtering node based on the collaborative filtering strategy variables; The game modeling module is used to construct the individual filtering cost function of the defense node; The load balancing module is used to make the defense nodes aim to minimize their own individual filtering cost function. It utilizes the load balancing property of the consistency of filtering delay of edge filtering nodes under the equilibrium condition, introduces a common filtering delay parameter, and performs interactive calculation in a distributed iterative manner to obtain the Nash equilibrium collaborative filtering strategy. The rule distribution and filtering execution module is used to coordinate the edge filtering node and the cloud filtering node to perform suspicious traffic diversion and filtering according to the Nash balance collaborative filtering strategy. The behavioral cognition module, based on the Nash equilibrium collaborative filtering strategy, constructs an optimization model for the attacker to maximize the total filtering cost of the system under the attack budget constraint, and solves for the optimal attack strategy; based on the optimal attack strategy, it outputs behavioral cognition results and early warning information on attack intensity, attack target preference, and attack type.
8. A device for recognizing collaborative defense and attack behavior in multi-robot systems, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the collaborative defense and attack behavior recognition method for multi-robot systems as described in any one of claims 1-6.