Multi-agent control method, multi-agent device and multi-agent system
By acquiring and fusing the local and global state deviations of multiple agents, information interaction and collaborative control across neighboring agents are achieved, solving the problems of long task completion time and poor collaborative performance in existing technologies, and improving the efficiency and consistency of the system.
Patent Information
- Application Number
- CN202511679481.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-17
- Publication Date
- 2026-02-06
AI Technical Summary
In existing multi-agent systems, local information interaction between individuals leads to long task completion times and poor group collaboration performance.
By acquiring the deviation between the current state and the desired state of the agent, as well as the state deviation of neighboring agents, the local and global relative state deviations are determined. By fusing local and global control inputs, information interaction and collaborative control across neighboring agents can be achieved.
It reduces the time required for multiple agents to complete system tasks and improves the system's convergence and group cooperation performance.
Smart Images
Figure CN121486767A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of intelligent agent control, and in particular, to a multi-agent control method. Background Technology
[0002] Intelligent agents, as independent entities that autonomously complete system tasks, are finding increasingly wider applications. For example, in systems composed of multiple intelligent agents—such as drones, unmanned ships, and swarms of mobile robots—that autonomously perform spatial mobility tasks, individuals need to collaborate through coordinated control and exchange information via communication networks to achieve the system's overall objectives. Therefore, reliable coordinated control and high-quality communication are crucial for accomplishing system tasks.
[0003] Multi-agent systems often employ mobile ad hoc networks (MANETs) as their networking architecture and utilize routing protocols such as Alignment Distance Vector Routing (AODV), Dynamic Source Routing (DSR), and Optimized Link State Routing (OLSR) for data transmission. Agents can exchange information through single-hop or multi-hop routing.
[0004] Current research mostly focuses on local information interaction between neighboring agents, which results in a long time required for multiple agents to complete system tasks and poor group collaboration performance. Summary of the Invention
[0005] This application provides a multi-agent control method to reduce the time required for multiple agents to complete system tasks.
[0006] The first aspect of this application provides a multi-agent control method, including:
[0007] On either agent's side,
[0008] Obtain the current first state deviation between the agent's current state and the expected state of the agent's system task.
[0009] Obtain the current second state deviation between the current state of each neighboring agent and the expected state of the system task of each neighboring agent.
[0010] Based on the agent's current first state deviation and the current second state deviations of each of its neighboring agents, determine the current local relative state deviations between the agent's current second state deviations and its current first state deviation.
[0011] The current local control input is obtained by summing up the determined deviations of each current local relative state.
[0012] The receiver receives the current third state deviation broadcast by the broadcast agent, wherein the third state deviation characterizes the state deviation between the current state of the broadcast agent and the desired state of the system task of the broadcast agent.
[0013] Based on the received third state deviation and the agent's current first state deviation, the current global relative state deviation between the third state deviation and the agent's current first state deviation is determined to obtain the current global control input.
[0014] The current local control input and the current global control input are combined to serve as the control input for this agent.
[0015] As one possible implementation, the fusion of the current local control input and the current global control input includes:
[0016] The current local control input is weighted using the first weighting coefficient.
[0017] The current global control input is weighted using a second weighting factor.
[0018] The weighted current local control input is summed with the weighted current global control input;
[0019] As one possible implementation, obtaining the current first state deviation between the agent's current state and the desired state of the agent's system task further includes:
[0020] The current first state deviation of the agent is sent to each of the agent's neighboring agents, so that each neighboring agent can obtain the local relative state deviation between itself and the agent based on its current second state deviation and the agent's current first state deviation.
[0021] As one possible implementation, obtaining the current first state deviation between the agent's current state and the desired state of the agent's system task includes:
[0022] For any first time point
[0023] The state of the agent at that first point in time is taken as the current state of the agent.
[0024] Calculate the difference between the current state of the agent and the expected state of the system task of the agent to obtain the first state deviation of the agent at the first time point;
[0025] As one possible implementation, obtaining the current second state deviation between the current state of each neighboring agent and the expected state of the system task of each neighboring agent includes:
[0026] Receive the second state deviation from each of the agent's neighboring agents at the first time point.
[0027] in,
[0028] The second state deviation is determined as follows:
[0029] For each of the intelligent agents' neighboring intelligent agents...
[0030] The state of the agent at the first time point is taken as the current state of the neighboring agent.
[0031] Calculate the difference between the current state of the neighboring agent and the expected state of the system task of the neighboring agent to obtain the second state deviation of the neighboring agent at the first time point.
[0032] As one possible implementation, the current third state deviation is determined in the following manner:
[0033] The broadcast agent will use its state at the second time point as its current state, where the second time point is a time point earlier than the first time point by a broadcast delay.
[0034] The difference between the current state of the broadcast agent and the expected state of the system task of the broadcast agent is calculated to obtain the third state deviation of the broadcast agent at the second time point, which is used as the global state deviation at the first time point.
[0035] As one possible implementation, the broadcast agent is any one of multiple agents;
[0036] The local relative state deviation is determined in the following manner:
[0037] For any agent and any neighboring agent.
[0038] Based on the second state deviation received by the agent from the neighboring agent at the first time point, the difference between the second state deviation and the first state deviation of the agent at the first time point is calculated to obtain the local relative state deviation of the agent relative to the neighboring agent at the first time point.
[0039] As one possible implementation, the global relative state deviation is determined in the following manner:
[0040] For any intelligent agent
[0041] Based on the third state deviation received by the agent from the broadcast agent at the first time point, the difference between the third state deviation and the first state deviation of the agent at the first time point is calculated to obtain the global relative state deviation of the agent relative to the broadcast agent at the first time point.
[0042] As one possible implementation, the broadcast agent is determined in the following manner:
[0043] If the number of agents is less than a set first threshold and / or the state change is less than a set change threshold, a static assignment strategy is used to determine the broadcast agents.
[0044] When the capabilities of the agent are limited, a round-robin assignment strategy is used to determine the broadcasting agent.
[0045] If the number of agents exceeds a set second threshold and / or the routing delay exceeds a set delay threshold, an autonomous election strategy is adopted to determine the broadcast agents.
[0046] As one possible implementation, the step of determining the broadcast agent using a static assignment strategy includes:
[0047] In a multi-agent system, one agent is pre-designated as a fixed broadcast agent.
[0048] The method of determining the broadcast agent using a round-robin assignment strategy includes:
[0049] At set time intervals, one agent is sequentially designated as the broadcast agent in a multi-agent network.
[0050] The broadcast agent is determined using an autonomous election strategy, including:
[0051] In a multi-agent system, each agent calculates the maximum number of hops from itself to all other agents based on the routing information it maintains.
[0052] The maximum number of hops calculated for interactions between agents is used to select the agent with the smallest maximum number of hops among all interactions as the broadcast agent.
[0053] As one possible implementation, the multi-agent network is composed of mobile ad hoc networks based on routing protocols.
[0054] The intelligent agent is a mobile intelligent agent that autonomously completes spatial movement tasks.
[0055] The state deviation includes positional offset.
[0056] The desired state includes the desired position.
[0057] A second aspect of this application provides an intelligent agent device, the intelligent agent device comprising:
[0058] The state module is used to obtain the current first state deviation between the agent's current state and the expected state of the agent's system task.
[0059] The local control module is used to acquire the current second state deviation between the current state of each neighboring intelligent agent and the expected state of the system task of each neighboring intelligent agent; based on the current first state deviation of the intelligent agent and the current second state deviation of each neighboring intelligent agent, it determines each current local relative state deviation between the current second state deviation of each neighboring intelligent agent and the current first state deviation of the intelligent agent; and accumulates the determined current local relative state deviations to obtain the current local control input.
[0060] The global control module controls the reception of the broadcast agent's current third state deviation, whereby the third state deviation characterizes the state deviation between the broadcast agent's current state and the desired state of the system task. Based on the received third state deviation and the agent's current first state deviation, the module determines the current global relative state deviation between the third state deviation and the agent's current first state deviation to obtain the current global control input.
[0061] The fusion module is used to fuse the current local control input and the current global control input as the control input for the agent.
[0062] A third aspect of this application provides a multi-agent system, which includes two or more of the aforementioned agent devices.
[0063] The multi-agent control method provided in this application receives state deviations from the broadcast agent among multiple agents, enabling the agent to obtain state deviation information across neighboring agents, which is beneficial for global information interaction. By obtaining the global relative state deviation of the agent relative to the broadcast agent and fusing this global relative state deviation as a global control input with the local control input, the broadcast agent becomes the leader, and the other agents become followers, while maintaining consistency among neighboring agents. This helps reduce the time required for multiple agents to complete system tasks and improves system convergence and group cooperation performance. Attached Figure Description
[0064] Figure 1 This is a schematic diagram illustrating one embodiment of the multi-agent control method of this application.
[0065] Figure 2 This is a flowchart illustrating a multi-agent control method according to an embodiment of this application.
[0066] Figure 3 This is a schematic diagram of the network topology of a multi-agent system.
[0067] Figure 4 This is a schematic diagram illustrating the simulation results of the collaborative formation control in this embodiment.
[0068] Figure 5 This is a schematic diagram illustrating the simulation results of the system convergence speed in this embodiment.
[0069] Figure 6 This is a schematic diagram showing the comparison between the convergence speed of the system in this embodiment and the convergence speed of the system in the prior art.
[0070] Figure 7 This is a schematic diagram of an intelligent agent device according to an embodiment of this application.
[0071] Figure 8 This is another schematic diagram of the intelligent agent device according to an embodiment of this application. Detailed Implementation
[0072] To make the objectives, technical means, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings.
[0073] This application proposes a multi-agent control method. A broadcast agent, used for information broadcasting, broadcasts its current state deviation, enabling all other agents in the multi-agent system except the broadcast agent to obtain the broadcast state deviation and determine the control input based on the state deviation. This allows each agent to follow the broadcast agent, thereby reducing the time required for the multi-agent system to complete the system task and improving the system convergence.
[0074] See Figure 1 As shown, Figure 1 This is a schematic diagram illustrating a multi-agent control method according to an embodiment of this application. The method includes: on the side of any agent in the multi-agent group,
[0075] Step 101: Obtain the current first state deviation between the agent's current state and the expected state of the agent's system task.
[0076] As an example, for any first time point, the state of the agent at that first time point is taken as the current state of the agent. The difference between the current state of the agent and the expected state of the system task of the agent is calculated to obtain the first state deviation of the agent at that first time point.
[0077] As an example, the step further includes: sending the agent's current first state deviation to each of the agent's neighboring agents, so that each neighboring agent obtains the local relative state deviation between itself and the agent based on its current second state deviation and the agent's current first state deviation.
[0078] Step 102: Obtain the current second state deviation between the current state of each neighboring agent and the expected state of the system task of each neighboring agent.
[0079] As an example, the agent receives the second state deviation from each of its neighboring agents at the first time point.
[0080] in,
[0081] The second state deviation is determined as follows:
[0082] For each of the intelligent agents' neighboring intelligent agents...
[0083] The state of the agent at the first time point is taken as the current state of the neighboring agent.
[0084] Calculate the difference between the current state of the neighboring agent and the expected state of the system task of the neighboring agent to obtain the second state deviation of the neighboring agent at the first time point.
[0085] Step 103: Based on the agent's current first state deviation and the current second state deviations of each of the agent's neighboring agents, determine the current local relative state deviations between the agent's current second state deviations and the agent's current first state deviation.
[0086] As an example, the local relative state deviation is determined as follows:
[0087] For any agent and any neighboring agent.
[0088] Based on the second state deviation received by the agent from the neighboring agent at the first time point, the difference between the second state deviation and the first state deviation of the agent at the first time point is calculated to obtain the local relative state deviation of the agent relative to the neighboring agent at the first time point.
[0089] Step 104: Accumulate the determined current local relative state deviations to obtain the current local control input.
[0090] Step 105: Receive the current third state deviation broadcast by the broadcast agent, wherein the third state deviation is used to characterize the state deviation between the current state of the broadcast agent and the desired state of the system task of the broadcast agent.
[0091] As an example, broadcast agents are determined in the following way:
[0092] When the number of agents is less than a set first quantity threshold and / or the state change is less than a set change threshold, a static assignment strategy is adopted to determine the broadcast agent, that is, one agent is pre-assigned as a fixed broadcast agent among multiple agents.
[0093] When the capabilities of the agents themselves are limited, a round-robin assignment strategy is adopted to determine the broadcast agent, that is, at a set time interval, one agent is assigned as the broadcast agent in turn among multiple agents.
[0094] If the number of agents exceeds a set second threshold and / or the routing delay exceeds a set delay threshold, an autonomous election strategy is adopted to determine the broadcast agent. That is, each agent in the multi-agent group calculates the maximum number of hops from itself to other agents based on the routing information it maintains. The maximum number of hops calculated by the interaction between agents is used to select the agent corresponding to the smallest maximum number of hops among the interaction maximum numbers as the broadcast agent.
[0095] As an example, the third state deviation is determined as follows:
[0096] The broadcast agent uses its state at a second time point as its current state. The second time point is earlier than the first time point by a broadcast delay. This broadcast delay can be determined based on the routing protocol used during the broadcast process. The broadcast delays to different agents can be different or the same to reduce complexity.
[0097] The difference between the current state of the broadcast agent and the expected state of the system task of the broadcast agent is calculated to obtain the third state deviation of the broadcast agent at the second time point, which is used as the global state deviation at the first time point.
[0098] Step 106: Based on the received third state deviation and the agent's current first state deviation, determine the current global relative state deviation between the third state deviation and the agent's current first state deviation to obtain the current global control input.
[0099] As an example, the global relative state deviation is determined as follows:
[0100] For any intelligent agent
[0101] Based on the third state deviation received by the agent from the broadcast agent at the first time point, the difference between the third state deviation and the first state deviation of the agent at the first time point is calculated to obtain the global relative state deviation of the agent relative to the broadcast agent at the first time point.
[0102] Step 107: Accumulate the current local control input and the current global control input to use as the control input for the agent.
[0103] As an example, the current local control input is weighted using a first weighting coefficient.
[0104] The current global control input is weighted using a second weighting factor.
[0105] The weighted current local control input is summed with the weighted current global control input.
[0106] In the above steps, steps 101-104 and steps 105-106 can be executed simultaneously.
[0107] The multi-agent control method provided in this application enables the broadcast agent to play a guiding role among multiple agents by using the state deviation of the broadcast agent across neighbor agents caused by communication routing. This is beneficial for achieving cooperative control and reduces the time for multiple agents to complete system tasks, thereby accelerating system convergence.
[0108] To facilitate understanding of the embodiments of this application, the following description uses a mobile intelligent agent that completes a spatial movement task autonomously and a system task that is spatial displacement as an example. It should be understood that the intelligent agent in the embodiments of this application is not limited to a mobile intelligent agent, and the system task is not limited to spatial displacement.
[0109] See Figure 2 As shown, Figure 2 This is a flowchart illustrating a multi-agent control method according to an embodiment of this application. The control method includes:
[0110] Step 201: Based on the routing protocol, initialize the communication routes of each agent used to complete the system tasks, so as to configure the neighbor agents of each agent and a broadcast agent.
[0111] In multi-agent systems, mobile ad hoc networks are often used as the network architecture. Agents can exchange local or global information through routing protocols such as AODV, DSR, and OLSR in a single-hop or multi-hop forwarding manner.
[0112] See Figure 3 As shown, Figure 3This is a schematic diagram of the network topology of a multi-agent system. The system comprises N agents, where N is a natural number greater than or equal to 2. One agent acts as a broadcast agent for global information exchange; this broadcast agent may also have neighboring agents. Dashed lines represent neighbor connections between agents, used for local information exchange between neighboring agents. Solid lines represent broadcast connections between the broadcast agent and all other agents, used for global information broadcasting.
[0113] Different broadcast agent selection strategies can be adopted for different scenarios to select the most suitable agent from the multi-agent system as the broadcast source, so as to ensure that its broadcast behavior can cover the entire network efficiently and reliably.
[0114] Broadcast agent selection strategies include one of the following:
[0115] (1) Static specification strategy
[0116] For scenarios with small network size and / or low agent mobility, a static assignment strategy can be adopted. As an example, before networking, an agent can be pre-assigned as a fixed broadcast agent, such as the agent with the lowest agent number, or an agent with a specific identity, such as a leader agent.
[0117] This strategy is simple and easy to implement, does not generate additional communication overhead, and is suitable for small networks with relatively stable communication conditions.
[0118] (2) Strategy assignment in turn
[0119] For energy-constrained scenarios, such as wireless sensor networks, a turn-based designation strategy can be employed. As an example, at set time intervals, a specific agent is designated as the broadcast agent in ascending order of its agent number. This strategy effectively avoids broadcast interruptions caused by a single broadcast agent consuming too much energy or malfunctioning.
[0120] (3) Autonomous election strategy
[0121] For scenarios with large network scale and / or low latency tolerance, an autonomous election strategy can be adopted to elect an agent located at the center of the network topology as the broadcast agent, thereby reducing the routing overhead and broadcast latency of the broadcast process.
[0122] As an example, each agent calculates the maximum number of hops from itself to all other agents based on its own maintained routing table. For instance, the maximum number of hops from agent n to all other agents is M. nEach agent exchanges its calculated maximum hop count information. Based on the exchanged maximum hop count information, the agent with the smallest exchanged maximum hop count is elected as the broadcast agent. For example, the exchanged maximum hop count information is M1, M2...M... n …M N The agent corresponding to the smallest value among them is selected as the broadcast agent.
[0123] For each agent in a multi-agent system, configure the neighboring agents of each agent according to the allowed communication distance between agents.
[0124] Step 202, for each agent,
[0125] The system calculates a first position offset between the agent's current position and its desired position, as a first state deviation, and receives second position offsets from each of the agent's neighboring agents. These second position offsets characterize the position offset between a neighboring agent's current position and its desired position, and are used as second state deviations.
[0126] Based on the first position offset and the second position offsets from each neighboring agent, the current local relative position offset between the first position offset and each second position offset is calculated to obtain the local relative position offset of the agent relative to each neighboring agent, which is used as the local relative state deviation.
[0127] The current local relative position offsets of the agent are accumulated, and the accumulated current local relative position offsets are weighted using a first weighting coefficient to obtain the current local control input of the agent.
[0128] The mathematical expression for the current local control input is:
[0129]
[0130] in, This represents the local control input of agent i at the first time point, for example, time t. Let be the first position offset of agent i at time t. Let i be the desired position for the system task of agent i, for example, in a formation control task, the desired position of the agent in the final formation. Let be the second position offset of agent j of agent i at time t. Let j be the desired location for the system task of agent j. Let be the local relative position offset between agent i and agent j at time t. This represents the neighbor connection state between agent i and agent j. When agent j is a neighbor agent of agent i, then... The value is 1, meaning that agent j is not a neighboring agent of agent i. The value is 0; is the first weighting coefficient, and N is the total number of agents.
[0131] As an example, each agent sends its first location offset to all its neighboring agents so that each neighboring agent can calculate its own current local control input. As another example, using the OLSR routing protocol, the first location offset is periodically sent to single-hop neighbors via HELLO messages.
[0132] Step 203: The broadcast agent broadcasts the third position offset to all other agents except the broadcast agent. The third position offset is used to characterize the position offset between the broadcast agent's current position and the broadcast agent's desired position, as a third state deviation.
[0133] Due to the time delay in the routing process, all agents except the broadcast agent will receive the broadcast agent's message at time t, for example, at time 2. The third position offset sent at time, where, This refers to the broadcast delay generated during the broadcasting process of the third position offset.
[0134] As an example, broadcast agents periodically broadcast via TC messages in multiple hops or single hops.
[0135] Step 204, for each agent,
[0136] The agent calculates its current global control input based on its current global relative position offset from the broadcast agent.
[0137] The mathematical expression for the current global control input is:
[0138]
[0139] in, Let be the global control input of agent i at time t. Let be the third position offset received by agent i at time t, that is, the position offset of the broadcast agent at time t. Location at any moment The desired location of the system task with the broadcast agent The third position offset between, Let be the global relative position offset of agent i with respect to the broadcast agent at time t. This is the second weighting coefficient.
[0140] Step 205: Each agent updates its current position based on its current local control input and current global control input.
[0141] Expressed mathematically as follows:
[0142]
[0143] in, Let i be the position updated by agent i at time t.
[0144] In this embodiment, by using the global relative position offset as the global control input, each agent can further superimpose the global control input on top of its local control input. This allows the broadcast agent to become the leader, while the other agents become followers, avoiding the limitations of relying solely on local control input. This improves system convergence and enables more efficient and reliable multi-agent cooperative control.
[0145] To illustrate the technical effects of this embodiment, the control process is simulated.
[0146] See Figure 4 As shown, Figure 4 This diagram illustrates the simulation results of the cooperative formation control in this embodiment. It demonstrates the effectiveness of the cooperative control process, showing the position trajectories of the multi-agent system and the resulting formation shape. It shows the process of multiple agents gradually forming a formation from their initial positions. Solid lines represent the movement trajectories of the agents, and red pentagrams represent the final formation shape of the multi-agent system.
[0147] See Figure 5 As shown, Figure 5 This diagram illustrates the simulation results of the system convergence speed in this embodiment. It shows the process of the exponential convergence speed change of each agent under cooperative control. The solid line represents the exponential convergence speed curve of each agent, and the dashed line represents the upper bound of the theoretically analyzed system exponential convergence speed. It can be seen that the convergence speed of each agent gradually approaches the upper bound of the theoretical exponential convergence speed.
[0148] See Figure 6 As shown, Figure 6 This diagram illustrates a comparison between the convergence speed of the system in this embodiment and the convergence speed of the system in the prior art. It is evident that the convergence speed of the system in this embodiment is improved compared to existing methods.
[0149] See Figure 7 As shown, Figure 7 This is a schematic diagram of an intelligent agent device according to an embodiment of this application. The device includes:
[0150] The state module is used to obtain the current first state deviation between the agent's current state and the expected state of the agent's system task.
[0151] The local control module is used to acquire the current second state deviation between the current state of each neighboring intelligent agent and the expected state of the system task of each neighboring intelligent agent; based on the current first state deviation of the intelligent agent and the current second state deviation of each neighboring intelligent agent, it determines each current local relative state deviation between the current second state deviation of each neighboring intelligent agent and the current first state deviation of the intelligent agent; and accumulates the determined current local relative state deviations to obtain the current local control input.
[0152] The global control module receives the current third state deviation broadcast by the broadcast agent. This third state deviation characterizes the state deviation between the broadcast agent's current state and the desired state of the system task. Based on the received third state deviation and the agent's current first state deviation, the module determines the current global relative state deviation between the third state deviation and the agent's current first state deviation to obtain the current global control input.
[0153] The fusion control module is used to accumulate the current local control input and the current global control input to serve as the control input for the agent.
[0154] See Figure 8 As shown, Figure 8 This is another schematic diagram of an intelligent agent device according to an embodiment of this application. The device includes a memory and a processor. The memory stores a computer program, and the processor is configured to execute the computer program to implement the steps of the multi-agent control method according to an embodiment of this application.
[0155] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.
[0156] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0157] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the multi-agent control method of this application.
[0158] For the device / network-side equipment / storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and relevant parts can be referred to in the description of the method embodiments.
[0159] In this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, without necessarily requiring or implying any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0160] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A multi-agent control method, characterized in that, include: On either agent's side, Obtain the current first state deviation between the agent's current state and the expected state of the agent's system task. Obtain the current second state deviation between the current state of each neighboring agent and the expected state of the system task of each neighboring agent. Based on the agent's current first state deviation and the current second state deviations of each of its neighboring agents, determine the current local relative state deviations between the agent's current second state deviations and its current first state deviation. The current local control input is obtained by summing up the determined deviations of each current local relative state. The receiver receives the current third state deviation broadcast by the broadcast agent, wherein the third state deviation characterizes the state deviation between the current state of the broadcast agent and the desired state of the system task of the broadcast agent. Based on the received third state deviation and the agent's current first state deviation, the current global relative state deviation between the third state deviation and the agent's current first state deviation is determined to obtain the current global control input. The current local control input and the current global control input are combined to serve as the control input for this agent.
2. The method as described in claim 1, characterized in that, The fusion of current local control input and current global control input includes: The current local control input is weighted using the first weighting coefficient. The current global control input is weighted using a second weighting factor. The weighted current local control input is summed with the weighted current global control input; The step of obtaining the current first state deviation between the agent's current state and the expected state of the agent's system task further includes: The current first state deviation of the agent is sent to each of the agent's neighboring agents, so that each neighboring agent can obtain the local relative state deviation between itself and the agent based on its current second state deviation and the agent's current first state deviation.
3. The method as described in claim 1 or 2, characterized in that, The step of obtaining the current first state deviation between the agent's current state and the expected state of the agent's system task includes: For any first time point The state of the agent at that first point in time is taken as the current state of the agent. Calculate the difference between the current state of the agent and the expected state of the system task of the agent to obtain the first state deviation of the agent at the first time point; The step of obtaining the current second state deviation between the current state of each neighboring agent and the expected state of the system task of each neighboring agent includes: Receive the second state deviation from each of the agent's neighboring agents at the first time point. in, The second state deviation is determined as follows: For each of the intelligent agents' neighboring intelligent agents... The state of the agent at the first time point is taken as the current state of the neighboring agent. Calculate the difference between the current state of the neighboring agent and the expected state of the system task of the neighboring agent to obtain the second state deviation of the neighboring agent at the first time point.
4. The method as described in claim 3, characterized in that, The current third state deviation is determined in the following manner: The broadcast agent will use its state at the second time point as its current state, where the second time point is a time point earlier than the first time point by a broadcast delay. The difference between the current state of the broadcast agent and the expected state of the system task of the broadcast agent is calculated to obtain the third state deviation of the broadcast agent at the second time point, which is used as the global state deviation at the first time point.
5. The method as described in claim 4, characterized in that, The broadcast agent can be any one of the multiple agents; The local relative state deviation is determined in the following manner: For any agent and any neighboring agent. Based on the second state deviation received by the agent from the neighboring agent at the first time point, the difference between the second state deviation and the first state deviation of the agent at the first time point is calculated to obtain the local relative state deviation of the agent relative to the neighboring agent at the first time point. The global relative state deviation is determined in the following manner: For any intelligent agent Based on the third state deviation received by the agent from the broadcast agent at the first time point, the difference between the third state deviation and the first state deviation of the agent at the first time point is calculated to obtain the global relative state deviation of the agent relative to the broadcast agent at the first time point.
6. The method as described in claim 5, characterized in that, The broadcast agent is determined in the following manner: If the number of agents is less than a set first threshold and / or the state change is less than a set change threshold, a static assignment strategy is used to determine the broadcast agents. When the capabilities of the agent are limited, a round-robin assignment strategy is used to determine the broadcasting agent. If the number of agents exceeds a set second threshold and / or the routing delay exceeds a set delay threshold, an autonomous election strategy is adopted to determine the broadcast agents.
7. The method as described in claim 6, characterized in that, The method of determining the broadcast agent using a static assignment strategy includes: In a multi-agent system, one agent is pre-designated as a fixed broadcast agent. The method of determining the broadcast agent using a round-robin assignment strategy includes: At set time intervals, one agent is sequentially designated as the broadcast agent in a multi-agent network. The broadcast agent is determined using an autonomous election strategy, including: In a multi-agent system, each agent calculates the maximum number of hops from itself to all other agents based on the routing information it maintains. The maximum number of hops calculated for interactions between agents is used to select the agent with the smallest maximum number of hops among all interactions as the broadcast agent.
8. The method as described in claim 1, characterized in that, The multi-agent network is composed of mobile ad hoc agents based on routing protocols. The intelligent agent is a mobile intelligent agent that autonomously completes spatial movement tasks. The state deviation includes positional offset. The desired state includes the desired position.
9. An intelligent agent device, characterized in that, The intelligent agent device includes: The state module is used to obtain the current first state deviation between the agent's current state and the expected state of the agent's system task. The local control module is used to acquire the current second state deviation between the current state of each neighboring intelligent agent and the expected state of the system task of each neighboring intelligent agent; based on the current first state deviation of the intelligent agent and the current second state deviation of each neighboring intelligent agent, it determines each current local relative state deviation between the current second state deviation of each neighboring intelligent agent and the current first state deviation of the intelligent agent; and accumulates the determined current local relative state deviations to obtain the current local control input. The global control module controls the reception of the broadcast agent's current third state deviation, whereby the third state deviation characterizes the state deviation between the broadcast agent's current state and the desired state of the system task. Based on the received third state deviation and the agent's current first state deviation, the module determines the current global relative state deviation between the third state deviation and the agent's current first state deviation to obtain the current global control input. The fusion module is used to fuse the current local control input and the current global control input as the control input for the agent.
10. A multi-agent system, characterized in that, The system includes two or more intelligent agent devices as described in claim 9.