Multi-agent elastic control method based on multi-step neighbor information difference screening

Abstract

The present invention discloses a multi-agent elastic control method based on multi-step neighbor information difference screening, which belongs to the field of multi-agent security control technology, and includes the following steps: first, construct a multi-agent network model under malicious attack, and give a corresponding multi-agent network communication topology structure according to the attack model. Secondly, the agents in the network exchange information with neighbor agents through the corresponding communication topology, and store the acquired multi-step neighbor information based on the information storage strategy. Then, the agents in the network discriminate and screen the stored neighbor information based on the difference of the multi-step neighbor information mean, and determine the available information set and the discarded information set. Finally, the agents in the network use the information in the available information set and use the elastic control protocol to realize the secure collaboration of the multi-agent network. The method provided by the present invention can effectively resist malicious attacks in the communication topology of the multi-agent network and ensure the secure collaboration of the multi-agent network.

Description

Technical Field

[0001] The present invention belongs to the technical field of multi-agent safety control, and in particular to a multi-agent elastic control method based on multi-step neighbor information difference screening. Background Art

[0002] With the development of large-scale multi-agent systems such as intelligent unmanned system clusters, multi-agent collaborative control strategies have received increasing attention. The distributed network structure of multi-agent systems brings them advantages such as high flexibility, easy operation and strong robustness. However, since multi-agent systems rely on network information interaction to complete collaborative tasks, they are vulnerable to network attacks, causing some agents to be unable to complete tasks according to the established plan, resulting in the failure of collaborative tasks. Therefore, in order to improve the reliability of multi-agent systems under malicious network attacks, it is urgent to carry out research on the security control methods of multi-agent systems. There have been many studies on the security issues of multi-agent systems in the field of computer and control science. However, previous studies have focused more on the defense mechanism of network attacks. After the defense is breached, how to ensure that the unattacked agents are not affected by the attacked agents and perform tasks normally, that is, the research on the network resilience of multi-agent systems is relatively small. Therefore, it is necessary to carry out relevant research on the elastic control methods of multi-agent systems.

[0003] At present, research on the resilient control of multi-agent systems mainly focuses on the study of the properties of network communication topology graphs. Existing results can be divided into resilient control methods for fixed communication topologies and resilient control methods for switching topologies. The existing technology of the former uses a weighted average consensus algorithm based on subsequence deletion to enable a first-order multi-agent system with specific communication topology properties to achieve resilient consensus. See H. J. LeBlanc, H. Zhang, X. Koutsoukos and S. Sundaram, "Resilient Asymptotic Consensus in Robust Networks," in IEEE Journal on Selected Areas in Communications, vol. 31, no. 4, pp. 766-781, April 2013, doi: 10.1109 / JSAC.2013.130413. The related research on the latter is extended on this basis and considers time-varying networks with more practical significance and application value. See G. Wen, Y. Lv, WX Zheng, J. Zhou and J. Fu, "Joint Robustness of Time-Varying Networks and ItsApplications to Resilient Consensus," in IEEE Transactions on Automatic Control, vol. 68, no. 11, pp. 6466-6480, Nov. 2023, doi: 10.1109 / TAC.2023.3237493. However, the communication topology requirements given by existing studies to enable the system to achieve elastic consistency are still difficult to achieve under realistic conditions. Therefore, it is necessary to use multi-step neighbor information on the basis of switching communication topology to further reduce the system's requirements for the relaxation of the time-varying communication graph and reduce the communication burden. Based on this, the present invention proposes a multi-agent elastic control method based on multi-step neighbor information difference screening. Summary of the invention

[0004] The purpose of the present invention is to provide a multi-agent elastic control method based on multi-step neighbor information difference screening, to solve the problem of multi-agent system collaborative control under network attacks existing in the prior art, and to reduce network communication pressure.

[0005] To achieve the above object, the present invention provides a multi-agent elastic control method based on multi-step neighbor information difference screening, comprising the following steps:

[0006] Step 1: Construct a multi-agent network model under malicious attack, and give the corresponding multi-agent network communication topology structure according to the attack model;

[0007] Step 2: The intelligent agents in the network obtained in step 1 interact with neighboring intelligent agents through the corresponding communication topology, and the obtained multi-step neighbor information is stored based on the information storage strategy;

[0008] Step 3: The agents in the network obtained in step 1 are used to discriminate and filter the stored neighbor information based on the difference of the multi-step neighbor information means to determine the usable information set and the discarded information set;

[0009] Step 4: The agents in the network obtained in step 1 use the information in the available information set and utilize the elastic control protocol to achieve secure collaboration of the multi-agent network.

[0010] Preferably, the specific process of constructing a multi-agent network model under malicious attack in step 1 and providing a corresponding multi-agent network communication topology structure according to the attack model is as follows:

[0011] Step 101: Build In a multi-agent network system composed of nodes, the description of nodes updating their states in discrete time is as follows:

[0012] ;

[0013] In the formula, and Respectively represent nodes exist The state and control input at each moment, Representation Node exist The state of the moment;

[0014] Step 102: construct a multi-agent network communication topology model, which is described in detail as follows:

[0015] Depend on The communication topology of a multi-agent network system composed of nodes is recorded as ,in Represents a set of nodes and a set of edges Indicates the communication relationship between nodes. Time Node For Node 's neighbors, then there are , otherwise ,node The neighbor set of ,node exist The neighbor set in the time period is recorded as ;

[0016] Step 103: construct a malicious attack model suffered by the multi-agent network, which is described in detail as follows:

[0017] The nodes that actually execute the same control algorithm as the designed algorithm and send real status information to each neighbor are called normal nodes, and their set is recorded as ; Nodes that actually execute control algorithms that are different from the designed algorithms and send real status information to each neighbor are called malicious nodes, and their set is recorded as ; so that there are no more than The attack model where a node becomes a malicious node is recorded as Overall malicious attack model; make sure that the neighbor set of each normal node does not exceed The attack model where a node becomes a malicious node is recorded as Local malicious attack model;

[0018] Step 104: Use parameters and Define the properties of the multi-agent network communication topology:

[0019] For a given and , if the node set Every pair of non-empty disjoint subsets of has at least one subset containing at least one node that satisfies the following conditions. The multi-agent network communication topology is said to have -Robustness, the specific conditions are as follows: Consider the topological graph , the node has at least r neighbors outside its subset;

[0020] Step 105: Construct a multi-agent network communication topology structure under a malicious attack model:

[0021] against Overall malicious attack model and Local malicious attack model, build Robust communication topology.

[0022] Preferably, in step 2, the intelligent agent in the network obtained in step 1 interacts with the neighbor intelligent agent through the corresponding communication topology, and the specific process of storing the obtained multi-step neighbor information based on the information storage strategy is as follows:

[0023] Step 201: The intelligent agents in the network exchange information with neighboring intelligent agents through the corresponding communication topology. Specifically, according to the network communication topology, each intelligent agent node exist Always send your identity to your neighbor agents ,state and communication time ;

[0024] Step 202: The intelligent agent in the network receives multi-step neighbor information and implements information storage, specifically: Moment, Node Receive each neighbor node Information sent and stored The information sent by each neighbor node during the time period.

[0025] Preferably, in step 3, the agents in the network obtained in step 1 discriminate and filter the stored neighbor information based on the difference of the multi-step neighbor information means, and the specific process of determining the usable information set and the discarded information set is as follows:

[0026] Step 301: Select The most recent status information sent by each neighbor during the time period:

[0027] Select As the neighbor state used by the subsequent algorithm, ,in, Representation Node Receiving neighbor nodes The last moment distance of sending the message The time difference between moments;

[0028] Step 302: Calculate the mean of multi-step neighbor information differences:

[0029] exist moment, for all ,calculate ,in, Represents the mean of multi-step neighbor information differences;

[0030] Step 303: Perform multi-step neighbor information difference mean screening:

[0031] for Arrange the resulting sequence in descending order, if , then the sequence is preceded by The element is assigned a value of 0. , then assign all elements in the sequence to 0, and record the new sequence as , where the elements are recorded as ;

[0032] Step 304: perform parameter normalization and value selection.

[0033] Preferably, the parameter normalization and value selection in step 304 are specifically as follows: Normalize the parameters, the specific expression is as follows:

[0034] ;

[0035] In the formula, for the node ,like ,but Enter the discarded information set, otherwise enter the available information set, where Represents the normalized value of the parameter.

[0036] Preferably, in step 4, the specific process of using the information in the available information set obtained in step 1 by the intelligent agents in the network to realize the secure collaboration of the multi-agent network by using the elastic control protocol is as follows:

[0037] Step 401, calculate the filtered Always participate in the node The node status weight value of the status update;

[0038] Step 402: Use the elastic control protocol to update the status.

[0039] Preferably, in step 401, the Always participate in the node The node status weight values ​​of the status update are as follows:

[0040] For all ,calculate , among which , the state authority value in the available information set is not zero, and the state authority value in the discarded information set is zero, where, Indicates that after filtering Always participate in the node The node status weight value of the status update.

[0041] Preferably, the state update using the elastic control protocol in step 402 is specifically as follows:

[0042] node Calculate the updated state based on the neighbor value and the weight of the filtered neighbor value , achieving safe collaborative control.

[0043] Therefore, the present invention adopts the above-mentioned multi-agent elastic control method based on multi-step neighbor information difference screening, which has the following beneficial effects:

[0044] (1) The present invention utilizes multi-step neighbor information to reduce network connectivity requirements and network communication pressure;

[0045] (2) Based on switching topology communication, the present invention constructs a multi-agent network model under malicious attack and provides the requirements for the multi-agent network communication topology structure to be constructed;

[0046] (3) The present invention provides information identification and screening methods and designs a flexible control protocol to achieve safe collaboration of multi-agent systems.

[0047] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is an overall flow chart of the multi-agent elastic control method based on multi-step neighbor information difference screening of the present invention;

[0049] Figure 2 Schematic diagram of the time-varying communication switching topology structure of the multi-agent system of the present invention, wherein (a) is a schematic diagram of the communication topology structure of the multi-agent system of the present invention at time k=1+5n (n=0,1,2,…); (b) is a schematic diagram of the communication topology structure of the multi-agent system of the present invention at time k=2+5n (n=0,1,2,…); (c) is a schematic diagram of the communication topology structure of the multi-agent system of the present invention at time k=3+5n (n=0,1,2,…); (d) is a schematic diagram of the communication topology structure of the multi-agent system of the present invention at time k=4+5n (n=0,1,2,…); (e) is a schematic diagram of the communication topology structure of the multi-agent system of the present invention at time k=5+5n (n=0,1,2,…);

[0050] Figure 3 It is a schematic diagram of a multi-agent network under constant value attack in a simulation experiment of the present invention;

[0051] Figure 4 Schematic diagram of a multi-agent network under a sine value attack in a simulation experiment of the present invention;

[0052] Figure 5 It is a schematic diagram of a multi-agent network under random value attack in the simulation experiment of the present invention. DETAILED DESCRIPTION

[0053] The following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] See also Figure 1-5 , a multi-agent elastic control method based on multi-step neighbor information difference screening includes the following steps:

[0055] Step 1: Construct a multi-agent network model under malicious attack, and give the corresponding multi-agent network communication topology structure according to the attack model. The specific process is as follows:

[0056] Step 101: Build In a multi-agent network system composed of nodes, the description of nodes updating their states in discrete time is as follows:

[0057] ;

[0058] In the formula, and Respectively represent nodes exist The state and control input at each moment, Representation Node exist The state of the moment;

[0059] Step 102: construct a multi-agent network communication topology model, which is described in detail as follows:

[0060] Depend on The communication topology of a multi-agent network system composed of nodes is recorded as ,in Represents a set of nodes and a set of edges Indicates the communication relationship between nodes. Time Node For Node 's neighbors, then there are , otherwise ,node The neighbor set of ,node exist The neighbor set in the time period is recorded as ;

[0061] Step 103: construct a malicious attack model suffered by the multi-agent network, which is described in detail as follows:

[0062] The nodes that actually execute the same control algorithm as the designed algorithm and send real status information to each neighbor are called normal nodes, and their set is recorded as ; Nodes that actually execute control algorithms that are different from the designed algorithms and send real status information to each neighbor are called malicious nodes, and their set is recorded as ; so that there are no more than The attack model where a node becomes a malicious node is recorded as Overall malicious attack model; make sure that the neighbor set of each normal node does not exceed The attack model where a node becomes a malicious node is recorded as Local malicious attack model;

[0063] Step 104: Use parameters and Define the properties of the multi-agent network communication topology:

[0064] For a given and , if the node set Every pair of non-empty disjoint subsets of has at least one subset containing at least one node that satisfies the following conditions. The multi-agent network communication topology is said to have -Robustness, the specific conditions are as follows: Consider the topological graph , the node has at least r neighbors outside its subset;

[0065] Step 105: Construct a multi-agent network communication topology structure under a malicious attack model:

[0066] against Overall malicious attack model and Local malicious attack model, build Robust communication topology.

[0067] Step 2: The agents in the network obtained in step 1 interact with neighboring agents through the corresponding communication topology, and the obtained multi-step neighbor information is stored based on the information storage strategy. The specific process is as follows:

[0068] Step 201: The intelligent agents in the network exchange information with neighboring intelligent agents through the corresponding communication topology. Specifically, according to the network communication topology, each intelligent agent node exist Always send your identity to your neighbor agents ,state and communication time ;

[0069] Step 202: The intelligent agent in the network receives multi-step neighbor information and implements information storage, specifically: Moment, Node Receive each neighbor node Information sent and stored The information sent by each neighbor node during the time period.

[0070] Step 3: The agents in the network obtained in step 1 are used to discriminate and filter the stored neighbor information based on the difference of the multi-step neighbor information means, and determine the usable information set and the discarded information set. The specific process is as follows:

[0071] Step 301: Select The most recent status information sent by each neighbor during the time period:

[0072] Select As the neighbor state used by the subsequent algorithm, ,in, Representation Node Receiving neighbor nodes The last moment distance of sending the message The time difference between moments;

[0073] Step 302: Calculate the mean of multi-step neighbor information differences:

[0074] exist moment, for all ,calculate ,in, Represents the mean of multi-step neighbor information differences;

[0075] Step 303: Perform multi-step neighbor information difference mean screening:

[0076] for Arrange the resulting sequence in descending order, if , then the sequence is preceded by The element is assigned a value of 0. , then assign all elements in the sequence to 0, and record the new sequence as , where the elements are recorded as ;

[0077] Step 304: perform parameter normalization and value selection:

[0078] based on Normalize the parameters

[0079] , for the node ,like ,but Enter the discarded information set, otherwise enter the available information set, where Represents the normalized value of the parameter.

[0080] Step 4: The agents in the network obtained in step 1 use the information in the available information set and use the elastic control protocol to achieve secure collaboration of the multi-agent network. The specific process is as follows:

[0081] Step 401, calculate the filtered Always participate in the node Node status weight value of status update:

[0082] For all ,calculate , among which , the state authority value in the available information set is not zero, and the state authority value in the discarded information set is zero, where, Indicates that after filtering Always participate in the node The node status weight value of the status update;

[0083] Step 402: Update the status using the elastic control protocol:

[0084] node Calculate the updated state based on the neighbor value and the weight of the filtered neighbor value , achieving safe collaborative control.

[0085] Example

[0086] The algorithm of the present invention is verified by adding malicious nodes in a multi-agent network. Select a switching topology directed communication graph network, such as Figure 2 As shown, select , , the arrow points to the direction of information transmission, Figure 2 (a)-(e) last for one time step each and repeat in a cycle. The topological graph has Robustness. The initial state of the agent is , where agents 1, 2, 3, and 5 are normal nodes, and are updated according to the algorithm given in the present invention. Agent 4 is a malicious node, and the following three attack forms are set for it: one is a constant value attack, that is, ; The second is the sine value attack, that is ; The third is a random value attack, namely For interval The random integer in . Verify the above three situations respectively, such as Figure 3 , Figure 4 , Figure 5 As shown in Figure 2, normal nodes in a multi-agent network can reach a resilient consensus in the presence of malicious nodes and ensure convergence to a safe interval. It is proved that the multi-agent elastic control method based on multi-step neighbor information difference screening proposed in the present invention can ensure the safe coordination of the multi-agent system and is feasible and effective.

[0087] Therefore, the present invention adopts the above-mentioned multi-agent elastic control method based on multi-step neighbor information difference screening, which can effectively resist malicious attacks in the multi-agent network communication topology, solve the problem of collaborative control of multi-agent systems under network attacks, reduce network communication pressure, and ensure the safe collaboration of multi-agent networks.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A multi-agent elastic control method based on multi-step neighbor information difference screening, characterized in that: The following steps are involved: Step 1: Construct a multi-agent network model under malicious attack, and give the corresponding multi-agent network communication topology structure according to the attack model; Step 2: The intelligent agents in the network obtained in step 1 interact with neighboring intelligent agents through the corresponding communication topology, and the obtained multi-step neighbor information is stored based on the information storage strategy; Step 3: The agents in the network obtained in step 1 are used to discriminate and filter the stored neighbor information based on the mean of the multi-step neighbor information difference, and determine the usable information set and the discarded information set; Step 4: The agents in the network obtained in step 1 use the information in the available information set to achieve secure collaboration of the multi-agent network using a flexible control protocol; In step 3, the agents in the network obtained in step 1 are used to discriminate and filter the stored neighbor information based on the mean of the multi-step neighbor information difference. The specific process of determining the usable information set and the discarded information set is as follows: Step 301: Select The most recent status information sent by each neighbor during the time period: Select As the neighbor state used by the subsequent algorithm, ,in, Representation Node Receiving neighbor nodes The last moment distance of sending the message The time difference between moments; Step 302: Calculate the mean of multi-step neighbor information differences: exist moment, for all ,calculate ,in, Represents the mean of multi-step neighbor information differences; Step 303: Perform multi-step neighbor information difference mean screening: for Arrange the resulting sequence in descending order, if , then the sequence is preceded by The element is assigned a value of 0. , then assign all elements in the sequence to 0, and record the new sequence as , where the elements are recorded as ; Step 304: perform parameter normalization and value selection; The parameter normalization and value selection in step 304 are specifically as follows: Normalize the parameters, the specific expression is as follows: In the formula, for the node ,like ,but Enter the discarded information set, otherwise enter the available information set, where Represents the normalized value of the parameter.

2. The multi-agent elastic control method based on multi-step neighbor information difference screening according to claim 1 is characterized in that: In step 1, the multi-agent network model under malicious attack is constructed, and the specific process of giving the corresponding multi-agent network communication topology structure according to the attack model is as follows: Step 101: Build In a multi-agent network system composed of nodes, the description of nodes updating their states in discrete time is as follows: In the formula, and Respectively represent nodes exist The state and control input at each moment, Representation Node exist The state of the moment; Step 102: construct a multi-agent network communication topology model, which is described in detail as follows: Depend on The communication topology of a multi-agent network system composed of nodes is recorded as ,in Represents a set of nodes and a set of edges Indicates the communication relationship between nodes. Time Node For Node Neighbors of , otherwise ,node The neighbor set of ,node exist The neighbor set in the time period is recorded as ; Step 103: construct a malicious attack model suffered by the multi-agent network, which is described in detail as follows: The nodes that actually execute the same control algorithm as the designed algorithm and send real status information to each neighbor are called normal nodes, and their set is recorded as ; The nodes that actually execute control algorithms that are different from the designed algorithms and send real status information to each neighbor are called malicious nodes, and their set is recorded as ; so that there are no more than The attack model where a node becomes a malicious node is recorded as - Overall malicious attack model; make sure that the neighbor set of each normal node does not exceed The attack model where a node becomes a malicious node is recorded as - Local malicious attack model; Step 104: Use parameters and Define the properties of the multi-agent network communication topology: For a given and , if the node set Every pair of non-empty disjoint subsets of has at least one subset containing at least one node that satisfies the following conditions. The multi-agent network communication topology is said to have -Robustness, the specific conditions are as follows: Consider the topological graph , the node has at least r neighbors outside its subset; Step 105: Construct a multi-agent network communication topology structure under a malicious attack model: against -Overall malicious attack model and - Local malicious attack model, building - Robust communication topology.

3. The multi-agent elastic control method based on multi-step neighbor information difference screening according to claim 2 is characterized in that: In step 2, the agents in the network obtained in step 1 interact with neighboring agents through the corresponding communication topology. The specific process of storing the obtained multi-step neighbor information based on the information storage strategy is as follows: Step 201: The intelligent agents in the network exchange information with neighboring intelligent agents through the corresponding communication topology. Specifically, according to the network communication topology, each intelligent agent node exist Always send your identity to your neighbor agents ,state and communication time ; Step 202: The intelligent agent in the network receives multi-step neighbor information and implements information storage, specifically: Moment, Node Receive each neighbor node Information sent and stored The information sent by each neighbor node during the time period.

4. The multi-agent elastic control method based on multi-step neighbor information difference screening according to claim 3 is characterized in that: In step 4, the agents in the network obtained in step 1 use the information in the available information set to achieve secure collaboration of the multi-agent network using the elastic control protocol. The specific process is as follows: Step 401, calculate the filtered Always participate in the node The node status weight value of the status update; Step 402: Use the elastic control protocol to update the status.

5. The multi-agent elastic control method based on multi-step neighbor information difference screening according to claim 4 is characterized in that: Step 401 calculates the filtered Always participate in the node The node status weight values ​​of the status update are as follows: For all ,calculate , among which , the state authority value in the available information set is not zero, and the state authority value in the discarded information set is zero, where, Indicates that after filtering Always participate in the node The node status weight value of the status update.

6. The multi-agent elastic control method based on multi-step neighbor information difference screening according to claim 5 is characterized in that: In step 402, the state update using the elastic control protocol is specifically as follows: node Calculate the updated state based on the neighbor value and the weight of the filtered neighbor value , achieving safe collaborative control.

Images