Distributed control method and system based on agent cluster

Abstract

The application relates to the technical field of intelligent control, and discloses a distributed control method and system based on an agent cluster. The method comprises the following steps: acquiring the distances between agents, and constructing an adjacency matrix; determining the connection weights of the agents based on the adjacency matrix and signal propagation characteristics, and generating a weight distribution matrix; if the communication quality between the agents deviates from a preset range, dynamically adjusting the weight distribution matrix; performing weighted fusion processing on the control parameters of the agents, obtaining a fusion control parameter vector, extracting key parameter components from the fusion control parameter vector, analyzing the change trend and convergence characteristics of the key parameter components, and optimizing the control parameters of the agents; predicting the position state of the agents at the next moment, updating the position index of the agents in a space grid based on the position state, distributing the control parameters to the agents corresponding to the position index, and forming a distributed adaptive adjustment system. The application improves the communication reliability of a multi-agent system in a dynamic environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control, in particular to a distributed control method and system based on agent cluster. BACKGROUND

[0002] In a multi-agent system, the contradiction between the communication quality fluctuation caused by dynamic complex environment and the difficulty of real-time optimization of control parameters becomes the core technical problem that restricts the efficiency and robustness of collaborative control. In actual business scenarios, such as unmanned aerial vehicle swarm collaborative search task execution or robot cluster cooperative handling in complex terrain, multi-agent needs to maintain efficient collaboration in dynamic network environment. However, signal attenuation or interruption caused by environmental changes (such as obstacles, electromagnetic interference) makes the communication quality between nodes unstable, which is difficult to accurately represent the adjacency relationship and weight distribution, and further affects the real-time generation and optimization of coordination control parameters. At the same time, the dynamic change of agent position increases the complexity of space grid update, and the traditional method is difficult to quickly predict and adjust under limited computing resources, resulting in poor convergence of control parameters and decreased collaborative efficiency. In addition, the real-time and accuracy of parameter distribution are limited by communication quality fluctuations, which easily causes some agents to receive outdated or incorrect parameters, further exacerbating the deviation of the overall system collaboration. The above problems are interwoven, which constitutes the core contradiction of multi-agent system in dynamic environment that is difficult to achieve efficient and robust collaborative control. Especially in high dynamic and high confrontation scenarios, this contradiction is more prominent, and it is urgent to solve the dynamic coupling problem between communication quality, parameter optimization and position prediction to ensure the stability and collaborative performance of the system in complex environment. SUMMARY

[0003] To solve the above technical problems, the present application provides a distributed control method and system based on agent cluster.

[0004] In a first aspect, the present application provides a distributed control method based on agent cluster, which comprises:

[0005] Step S1: obtaining the real-time position coordinates of each agent in the multi-agent system, calculating the distance between agents, and constructing the adjacency matrix between agents based on the distance;

[0006] Step S2: determining the connection weight of each agent based on the adjacency matrix and signal propagation characteristics, generating a weight distribution matrix, continuously monitoring the communication quality between agents, and if the communication quality deviates from the preset range, dynamically adjusting the weight distribution matrix to obtain a corrected weight distribution matrix;

[0007] Step S3: the control parameters of each agent are weighted and fused by using the modified weight distribution matrix to obtain a fused control parameter vector, a key parameter component in the fused control parameter vector is obtained, a change trend and convergence characteristics of the key parameter component are analyzed, and the control parameters of the agent are optimized;

[0008] Step S4: a relative displacement vector between each agent is calculated, a position state of the agent at the next moment is predicted based on the relative displacement vector, a position index of the agent in a space grid is updated based on the position state, the optimized control parameters are distributed to the agent corresponding to the position index, and a distributed adaptive adjustment system is formed.

[0009] Compared with the prior art, the beneficial effects of the present application are at least as follows:

[0010] The present application is aimed at the problem of communication quality fluctuation and control parameter real-time optimization difficulty caused by dynamic environment in the multi-agent system, and realizes efficient cooperation through the technical means of fusion positioning, communication monitoring and parameter optimization. Firstly, the present application calculates the Euclidean distance and constructs the adjacency matrix by real-time obtaining the three-dimensional coordinates and signal strength of the agent, accurately representing the connectivity between nodes; generates a weight distribution matrix based on the signal propagation attenuation characteristics, dynamically adjusts the weight to reflect the real-time communication quality; generates a coordinated control parameter vector by weighted fusion, and ensures the convergence through parameter increment analysis and optimization; predicts the position of the agent by matrix sparsification, and updates the space grid; finally, the optimized parameters are distributed to each agent through the parameter distribution confirmation mechanism, forming a distributed adaptive adjustment system. The present application significantly improves the cooperation efficiency and robustness of the multi-agent system in complex environment through dynamic coupling optimization of communication quality and control parameters, and is suitable for distributed control tasks in dynamic network scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0012] Figure 1 An embodiment of the distributed control method based on agent cluster in the embodiment of the present application is shown in the figure;

[0013] Figure 2 A prediction flowchart of the agent position in the embodiment of the present application is shown in the figure;

[0014] Figure 3A comparison chart of the effects of the traditional method and the method of the present application on multiple key performance indicators in the embodiments of the present application;

[0015] Figure 4 An embodiment diagram of the distributed control system based on the agent cluster in the embodiments of the present application. DETAILED DESCRIPTION

[0016] The embodiments of the present application provide a distributed control method and system based on an agent cluster. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0017] For the sake of understanding, the specific flow of the embodiments of the present application is described below. Please refer to Figure 1 An embodiment of the distributed control method based on the agent cluster in the embodiments of the present application includes:

[0018] Step S1: Obtain the real-time position coordinates of each agent in the multi-agent system, and calculate the distance between the agents. Based on the distance, an adjacency matrix between the agents is constructed.

[0019] Specifically, the multi-agent system refers to a system composed of multiple agents, all of which work together to complete tasks or solve problems through certain rules and collaboration mechanisms. For example, in a drone swarm, each drone is an independent agent that collaborates through communication to complete tasks such as photography, patrol, search, etc. However, as the distance between agents changes, the quality of communication is affected by various factors (such as signal strength, network latency, environmental changes, etc.), so it is necessary to obtain the position coordinates of the agents in the environment. The present application obtains the real-time position coordinates of each agent through sensors or positioning systems and represents them using three-dimensional coordinates. The Euclidean distance between each pair of agents is calculated to obtain the distance between them. Based on the distance, the connection state between them is determined, and a neighbor matrix is set based on the connection state. Each row of the neighbor matrix represents the connection information of an agent, and each column of the matrix represents the connection state with other agents. If the distance between the two is within the effective communication range, the corresponding matrix element is assigned a weight value representing the strength of the connection. If the distance exceeds the effective communication range, the matrix element is assigned a value of zero, indicating that there is no effective connection between the two. The weight value can be set according to specific requirements and is usually adjusted based on factors such as signal strength attenuation, transmission delay, etc. The specific setting method will be described later. By obtaining the position coordinates of the agents in real time and calculating the distance between them, the connection state between the agents can be accurately reflected, effectively ensuring the quality of communication and collaboration efficiency. Building a neighbor matrix based on distance helps to dynamically adjust the collaboration relationship between agents and optimize the coordination and communication reliability of the system.

[0020] Step S2: Based on the neighbor matrix and signal propagation characteristics, determine the connection weight of each agent, generate a weight distribution matrix, and continuously monitor the communication quality between agents. If the communication quality deviates from the preset range, dynamically adjust the weight distribution matrix to obtain a corrected weight distribution matrix.

[0021] Specifically, in order to ensure the effective cooperation of each agent in the multi-agent system, it is necessary to adjust the connection weight between agents in real time according to the connection condition and communication quality between agents. The connection weight refers to the strength or influence of the communication connection between two agents in the multi-agent system. It not only indicates "whether to connect", but also quantifies the quality of communication, such as signal strength, transmission delay, interference level, etc. Specifically, first, the connection information between each pair of agents is extracted from the adjacency matrix. If the distance between two agents is close and the signal quality is good, a higher connection weight is given. According to the signal propagation characteristics combined with the distance between agents, signal attenuation and other factors, the signal propagation attenuation model is used to adjust the connection weight, obtain the weighted influence coefficient, and obtain the weight distribution matrix based on the weighted influence coefficient. Each item of the matrix corresponds to the weighted connection relationship between agent i and agent j. The detailed calculation is described below. After generating the weight distribution matrix, the system needs to continuously monitor the communication quality between agents, including data transmission delay, signal strength and other communication quality indicators. When the communication quality between a pair of agents deviates from the preset range, the system will detect this abnormal situation. In order to ensure system stability and reliability, the system will start the dynamic adjustment mechanism to correct the weight coefficients in the weight distribution matrix. For example, assume that the communication quality between agent i and agent j changes, resulting in an increase in signal attenuation. The system will adjust the weight at the corresponding position in the matrix according to the new signal attenuation value, so that the connection strength between the two agents is reduced. Through the dynamic adjustment process of the connection weight coefficients between agents, the system can respond in time when the communication quality fluctuates, avoid the overall performance decline caused by the communication problem between a single agent, and correct the weight distribution matrix to reflect the latest communication quality state, providing optimized data support for subsequent agent cooperation, control decision and information transmission, ensuring that the system can adapt to changes in the dynamic environment.

[0022] Step S3: using the corrected weight distribution matrix to perform weighted fusion processing on the control parameters of each agent to obtain a fused control parameter vector, obtaining a key parameter component in the fused control parameter vector, analyzing the change trend and convergence characteristics of the key parameter component, and optimizing the control parameters of the agent.

[0023] Specifically, the control parameters of the agents include position control parameters, speed control parameters, acceleration control parameters, etc., and the control parameters of each agent are weighted and averaged according to the weight coefficients in the adjacency matrix and the control parameters of the neighbor nodes to obtain the fused control parameters. The fused control parameters of all agents are combined in the order of node number to form a whole fused control parameter vector, which represents the coordinated control strategy of the whole system. The fused control parameter vector includes fused position control parameter components, fused speed control parameter components, and fused acceleration control parameter components. The specific fusion process will be described later. Through the weighted fusion method, the control strategy of each agent not only considers its own situation, but also combines the state of the neighbor agents, thereby enhancing the collaboration of the whole system. Secondly, key parameter components are extracted from the fused control parameter vector, such as position control gain, speed synchronization coefficient, and communication weight factor. The position control gain is used to adjust the response speed and accuracy of the agent when performing position adjustment tasks. The speed synchronization coefficient is used to control how multiple agents coordinate their speeds in collaborative tasks, so that they can move synchronously. The communication weight factor determines the importance or priority of information exchange between agents. After extracting the key parameter components, their change trends and convergence characteristics are analyzed. Specifically, the increment of each key parameter component within a continuous time step is calculated to determine whether it is stable, thereby determining whether it has converged. For example, the change trend of the position control gain can be obtained by calculating the difference between the current position and the position at the previous time. If the difference gradually decreases, it means that the parameter tends to be stable and may have converged. The specific analysis process will be described later. If the increment of a certain parameter exceeds the preset convergence threshold, the system will start the optimization process to adjust and optimize the control parameters of the agents. Methods such as gradient descent, adaptive gain adjustment, or weighted averaging are used to correct the control parameters, and finally an optimized control parameter set is generated to make the multi-agent system maintain stability and fast convergence characteristics during collaboration. The detailed optimization method will be described later. Adjusting the control parameters according to the current environmental changes not only improves the stability and collaboration efficiency of the system, but also automatically adjusts the behavior according to the environmental changes, and ensures that the multi-agent system can still operate coordinately under different conditions.

[0024] Step S4: Calculate the relative displacement vectors between the agents, predict the position state of the agents at the next time based on the relative displacement vectors, and update the position index of the agents in the spatial grid based on the position state. Distribute the optimized control parameters to the agents corresponding to the position index to form a distributed adaptive adjustment system.

[0025] Specifically, by obtaining the difference between the position coordinates of each agent at the current time and the position coordinates at the last time, the relative displacement vector of the agent is obtained, which represents the motion direction and speed of the agent in unit time. Based on the calculated relative displacement vector, the system predicts the position of the agent at the next time by linear extrapolation. The agent activity area is divided into a spatial grid, and each grid cell has a unique row and column index. According to the predicted agent position, it is judged whether it exceeds the current grid boundary. If it exceeds, the grid index position of the agent is recalculated, and the spatial index table is updated. Each agent always corresponds to a unique spatial index, representing its relative position in the environment. According to the updated spatial grid index, the optimized control parameters are distributed to the corresponding agent. Since the agent is constantly moving, if the distribution continues according to the initial position, it is easy to cause parameter distribution error or delay. Therefore, through the position index mechanism, the system can ensure that the optimized parameters are always delivered to the correct target node. When all agents receive the optimized control parameters corresponding to their position index, the system forms a distributed adaptive adjustment system that can dynamically update parameters based on real-time position, motion trend and neighbor relationship, thereby ensuring that the multi-agent system has stability, robustness and high-efficiency collaboration ability during task execution. The system can adjust the control strategy in real time according to the current position, motion trend and environmental changes of the agent, and realize efficient collaboration and adaptive adjustment among agents.

[0026] In a specific embodiment, the distance between agents is calculated, and the adjacency matrix between agents is constructed based on the distance, which includes the following steps:

[0027] Each agent is taken as the center, and other agents with a distance less than or equal to the effective communication distance are selected as neighbor nodes. The connectivity of each agent's neighbor nodes is verified, and the nodes that pass the verification are defined as the effective neighbor nodes of the agent. The adjacency matrix reflecting the connectivity between nodes is generated based on the effective neighbor nodes, wherein the adjacency matrix takes each agent node as the index, and the matrix element value represents the distance data and connection state between agents.

[0028] Specifically, each agent is centered, and other agents with a distance less than or equal to the effective communication distance from the agent are screened as neighbor nodes, and the effective communication distance is set in advance according to the communication capability of the agent and the environmental conditions, usually considering factors such as communication power, signal attenuation, and environmental impact, for example, assuming that the distance between agent 1 and agent 2 is 5 meters, and the effective communication distance is set to 10 meters, agent 2 will be regarded as a neighbor node of agent 1. For all agents, the neighbor node set of each agent is determined based on this method. Subsequently, the connectivity of each agent's neighbor node is verified, and the communication between nodes is determined to be effective and connected by monitoring whether the actual communication quality indicators meet the communication quality requirements, including data transmission success rate, signal attenuation degree, and network delay parameters. The neighbor nodes that pass the connectivity verification are marked as effective neighbor nodes of the agent, and the adjacency matrix is generated based on the effective neighbor nodes. The adjacency matrix takes the agent node as the index, and the matrix element is used to reflect the connection state and distance relationship between nodes. For example, in a three-dimensional space, assuming that there are three agents, agent 1, agent 2, and agent 3, the distance between agent 1 and agent 2 is 5 meters, the distance between agent 1 and agent 3 is 12 meters, and the distance between agent 2 and agent 3 is 7 meters, if the effective communication distance is set to 10 meters, agent 1 and agent 2, agent 2 and agent 3 can communicate, and agent 1 and agent 3 cannot communicate, and the final adjacency matrix is represented by the following form: wherein, represents the connection strength between agent 1 and agent 2, represents that there is no effective connection between agent 1 and agent 3. The adjacency matrix constructed by the above steps can accurately reflect the topological relationship between agents in a real-time dynamic environment, avoiding false connectivity caused by single distance judgment or unstable communication. Since the communication quality verification mechanism is introduced in the process of screening neighbor nodes, it can effectively eliminate invalid connections with low signal quality or high delay, thereby reducing the interference of redundant edges and unreliable edges on overall coordinated control.

[0029] In a specific embodiment, based on the adjacency matrix and signal propagation characteristics, the connection weight of each agent is determined, and a weight distribution matrix is generated, which specifically includes the following steps:

[0030] The distance data of the agent node pair is obtained from the adjacency matrix, and the distance data is input into a preset signal propagation attenuation function to calculate the signal attenuation degree between the agent node pair, and the reciprocal of the signal attenuation degree is defined as the spatial influence strength;

[0031] The importance indicators of each agent node are obtained, and the importance indicators include the degree centrality indicator, the betweenness centrality, and the neighborhood density coefficient. The connection weight of each agent node is obtained based on the importance indicators by using a weighted average method.

[0032] The product of the spatial influence intensity and the connection weight is taken as the weighted influence intensity of the agent node. The weighted influence intensity is taken as the element value of the matrix, and the weight distribution matrix is generated.

[0033] Specifically, first, the distance data between the agent node pairs are obtained from the adjacency matrix, and a signal propagation attenuation function is established. The function adopts an exponential attenuation model, and a mathematical mapping relationship between the signal strength attenuation degree and the propagation distance is established. The attenuation coefficient is determined according to the communication frequency and the environmental medium characteristics. The distance data of each agent node pair is substituted into the signal propagation attenuation function, and the signal attenuation degree between each pair of nodes is calculated one by one. The spatial influence intensity value is obtained by reciprocal transformation of the attenuation degree. The influence intensity value reflects the quality of the communication between the nodes. The node pairs with shorter distances obtain higher influence intensity values, and the node pairs with longer distances obtain lower influence intensity values. The importance indicators of each agent node are obtained, including the degree centrality indicator, the betweenness centrality, and the neighborhood density coefficient. The degree centrality indicator reflects the direct connection ability of the node in the network. The node with more connection edges has a higher degree centrality value. The betweenness centrality reflects the bridge role of the node in the network, that is, the importance of the node in the shortest path. The neighborhood density coefficient measures the distribution density of the neighbor nodes around the node. The node in the high-density area has strong network connection ability. The calculation method of the importance indicators is described in detail later. The degree centrality, the betweenness centrality, and the neighborhood density coefficient are combined, and the connection weight of each node is calculated by using a weighted average method. The coefficients in the weight calculation formula are adjusted according to the network size and the application scenario, so that the weight distribution can accurately reflect the importance of the node in the network topology. The connection weight value is normalized, so that the sum of the weights of all nodes is equal to the total number of network nodes. Finally, the spatial influence intensity and the connection weight of each node are multiplied to obtain the weighted influence intensity of the agent node. The weighted influence intensity considers the physical distance factor and the network topology factor, and can more accurately reflect the actual influence relationship between the nodes.

[0034] A weight distribution matrix of n*n dimension is constructed, where n is the total number of nodes in the agent network, the row index and the column index of the matrix correspond to the number of the sending node and the receiving node respectively, and the element position (i, j) of the matrix stores the weighted influence strength value from node i to node j. During the initialization process of the matrix, the diagonal elements are set to 0, indicating that the influence strength of the node on itself is zero. The weighted influence strength of each pair of agent nodes is taken as the element value of the matrix. For the node pairs identified as non-connected in the adjacency matrix, the corresponding weight distribution matrix elements are directly set to 0. Further, a weight decay coefficient is introduced to adjust the time dimension of the weighted influence strength. The weight decay coefficient is calculated based on the historical communication quality data between nodes. Node pairs with stable communication quality obtain smaller decay coefficients, and node pairs with larger communication quality fluctuations obtain larger decay coefficients. After multiplying the decay coefficient and the weighted influence strength, the corresponding position of the weight distribution matrix is filled in to complete the construction process of the initial weight distribution matrix. The weight distribution matrix generated in the above manner can effectively reflect the actual connection relationship between agents, taking into account the physical distance factor, and also considering the signal propagation characteristics and the importance of agent nodes, thereby providing accurate support for the dynamic adjustment and optimization of the system.

[0035] In a specific embodiment, obtaining the importance indicator of each agent node specifically includes the following steps:

[0036] The number of connection edges of each agent node in the adjacency matrix is counted, which is defined as the degree centrality indicator of the agent node.

[0037] The shortest path between all pairs of agent nodes is obtained, and the number of times each agent node appears in the shortest path is counted, which is defined as the betweenness centrality of the agent node.

[0038] The number of neighbor nodes within a predetermined radius range is counted with each agent node as the center, and the distance of the neighbor nodes is weighted processed by a Gaussian kernel function to obtain a neighborhood density coefficient.

[0039] Specifically, the degree centrality is a measure of the degree of direct connection of each agent node in the network, and for each agent node i, the number of connected edges in the adjacency matrix is counted, and the number of connected edges directly reflects the degree centrality of the node. The higher the degree centrality, the more direct connections between the agent node and other nodes, and the more important the node is in the network. The calculation of the betweenness centrality is to count the shortest paths between all pairs of agent nodes in the network, and to evaluate the indirect connection ability of each agent node in the network. Specifically, for each pair of agent nodes i and j, the shortest path from i to j is calculated, and the number of times that agent node k appears in the path is recorded. The more times agent node k appears in all shortest paths, the more important it is as a bridge in the network, and the higher the betweenness centrality. Finally, the neighborhood density coefficient measures the geographical aggregation ability of the node in the network by counting the distribution density of the neighbor nodes around each agent node. In actual multi-agent systems, nodes often exhibit characteristics of aggregated distribution, with higher node density in some areas and lower node density in some areas. The calculation of the neighborhood density coefficient uses an adaptive radius strategy. Specifically, for areas with high node density, the neighborhood radius is set to a small value around each agent node to avoid interference with density calculation by too many neighbor nodes. For areas with low node density, the neighborhood radius is set to a larger value to ensure that enough neighbor node information is captured. Then, the distance of these neighbor nodes is weighted by a Gaussian kernel function to obtain the neighborhood density coefficient. The Gaussian kernel function is used to give greater weight to neighbor nodes with closer distances and smaller weight to nodes with farther distances, thereby reflecting the tightness of the nodes in space. Nodes with higher neighborhood density coefficients indicate that they are in dense areas in the network and may play a greater role in information propagation and collaboration. By calculating the importance indicators of each agent node and evaluating the relative importance of the nodes in the network based on these indicators, the importance indicators will provide a basis for subsequent control parameter adjustment, helping the system to prioritize key nodes when performing multi-agent collaboration, and ensuring the stability and efficiency of the overall network.

[0040] In a specific embodiment, the weight distribution matrix is dynamically adjusted to obtain a corrected weight distribution matrix, which includes the following steps:

[0041] If the communication quality between the agents deviates from the preset range, the agent node pair is defined as an abnormal node pair, the matrix element position of the abnormal node pair is located in the weight distribution matrix, and the weighted influence strength at the element position is obtained as the weight coefficient;

[0042] The communication quality indicators between the agents include data transmission delay and signal attenuation degree, a deviation value of the data transmission delay and the signal attenuation degree is calculated, a correction coefficient is calculated by linear interpolation based on the deviation value, the weight coefficients in the weight distribution matrix are dynamically adjusted by using an exponential decay function and the correction coefficient to obtain a corrected weight distribution matrix.

[0043] Specifically, when dynamically adjusting the weight distribution matrix, the communication quality indicators between the agents, especially the data transmission delay and the signal attenuation degree, need to be monitored first. If the communication quality between the agents deviates from the preset normal range, the system defines the agent node pair as an abnormal node pair and locates the matrix element positions of these abnormal node pairs in the weight distribution matrix. The system obtains the weighted influence strength of these abnormal node pairs, i.e. locates the weight coefficients corresponding to the element in the matrix. When calculating the correction coefficient, the deviation value of the data transmission delay and the signal attenuation degree is calculated first. The data transmission delay deviation is the difference between the current measured delay value and the preset normal delay range. The signal attenuation deviation is the difference between the actual measured attenuation value and the expected attenuation value. A mapping relationship between the quality deviation and the weight adjustment amount is established. The specific weight correction coefficient for the current quality deviation degree is calculated by linear interpolation method. The correction coefficient reflects the real-time change of the communication link reliability. The numerical range of the correction coefficient is between 0 and 1. When the communication quality is good, the correction coefficient is close to 1. When the communication quality significantly decreases, the correction coefficient tends to 0, thereby reflecting the adjustment amplitude of the weight under different communication states.

[0044] Further, the weight coefficient is updated by combining the correction coefficient with an exponential decay function. The input of the exponential decay function is the communication deviation value, and the output is a decay factor that gradually decreases with the increase of the deviation value, which is used to reflect the rapid weakening effect of the communication condition deterioration on the weight. Thus, the dynamic adjustment formula is: on the basis of the original weight coefficient, multiply the product of the correction coefficient and the decay factor to obtain the corrected weight value. For example, in an unmanned aerial vehicle cluster communication scenario, if the original weight coefficient is 0.8, when the communication delay and signal attenuation deviation values are detected to be 0.6 and 0.5 respectively, the system calculates the correction coefficient to be 0.7, and the decay factor output by the exponential decay function is 0.5, then the corrected weight is 0.8*0.7*0.5=0.28, which reflects the weight weakening caused by the deterioration of the link quality. Through the above adjustment process, the system can make the link weight affected more by the communication quality decrease rapidly under the premise of ensuring the overall stability of the matrix, while the link weight maintaining stable communication quality is maintained or slowly recovered, thereby realizing the dynamic adaptive update of the weight distribution matrix. In the weight adjustment process, the system sets the upper and lower boundary limits of the weight adjustment to ensure that the adjustment amplitude does not exceed the preset percentage of the original weight value, and the adjusted weight value always remains within the effective range, thereby preventing the mutation of the weight from impacting the stability and coordination of the system, and ensuring that the adjusted weight value meets the requirements of the network topology. Finally, the corrected weight coefficient is written back to the corresponding position of the weight distribution matrix, and the corrected weight distribution matrix can reflect the changes of the communication quality between agents in real time, providing more accurate data support for subsequent multi-agent cooperation and control decision-making.

[0045] In a specific embodiment, the control parameters of each agent are weighted and fused to obtain a fused control parameter vector, which includes the following steps:

[0046] The weight distribution matrix is normalized to obtain the weight coefficients of each agent node in the normalized weight distribution matrix, and the control parameters of the effective neighbor nodes of each agent are obtained, including position control parameters, velocity control parameters and acceleration control parameters. The control parameters of the effective neighbor nodes are numerically weighted according to the weight coefficients, and each type of control parameter of all agents is fused according to the node number sequence to obtain a fused control parameter vector.

[0047] Specifically, in the weighted fusion processing of the control parameters of each agent, the weight distribution matrix is first normalized. In the normalization process, the sum of all non-zero elements in each row of the weight distribution matrix is calculated to obtain the sum Si, and then each element in the row is divided by the sum Si to obtain the normalized weight value. The weight coefficient corresponding to each agent node in the normalized weight distribution matrix is obtained, and the control parameters of the effective neighbor nodes of each agent and the weight coefficient are obtained. For example, for an unmanned aerial vehicle formation system including 8 agents, the neighbor node set of agent 1 is {2, 3, 8}, and the corresponding normalized weights are 0.4, 0.35, and 0.25. The position control parameters (coordinates) of agent 1 are [10.2, 15.6, 8.3], and the position control parameters of the neighbor nodes are [9.8, 16.1, 8.1], [10.5, 15.2, 8.5], and [10.1, 15.8, 8.2], respectively. According to the formula The fused position control parameters of agent 1 are calculated as [10.15, 15.67, 8.28]. For each control parameter of all agents, the fusion is performed in the order of node number. For the position control parameter, the system accumulates the weighted position control parameters of all neighbor nodes to obtain the fused position control parameters of the agent. By performing weighted fusion on the control parameters of each agent, the fused position control parameter component, the fused speed control parameter component, and the fused acceleration control parameter component are obtained. The fused control parameter vector of each agent contains the optimal control strategy of the agent and considers the cooperation information of the neighbor nodes, ensuring that the system can maintain efficient and stable cooperation in a dynamic environment.

[0048] In a specific embodiment, the variation trend and convergence characteristics of the key parameter components are analyzed, and the control parameters are optimized, specifically including the following steps:

[0049] The position control gain, the speed synchronization coefficient, and the communication weight factor are extracted from the fused control parameter vector as key parameter components.

[0050] The parameter increment sequence is obtained by calculating the difference between the current parameter value and the previous parameter value of each key parameter component in the continuous time step.

[0051] The variance and mean ratio of the parameter increment sequence is calculated, and the corresponding convergence test threshold is set based on different key parameter components.

[0052] Based on the variance and mean ratio and the convergence test threshold, it is judged whether the key parameter component has convergence. When it is detected that the parameter increment exceeds the convergence test threshold, the parameter re-optimization process is started.

[0053] According to the adjustment result of the parameter re-optimization process, a weighted average method is used to fuse the original parameter value and the optimized adjustment value to generate a new control parameter value.

[0054] Specifically, the position control gain, the speed synchronization coefficient and the communication weight factor are extracted from the fusion control parameter vector as key parameter components, wherein the position control gain adjusts the response speed of the agent to the target position, the speed synchronization coefficient ensures the speed coordination between the agents, and the communication weight factor measures the communication quality between the agents. These parameter components represent the control behavior of each agent in the multi-agent system. After extracting these parameter components, the variation trend and convergence characteristics of each parameter component are analyzed by a parameter increment calculation method. The parameter increment calculation adopts a first-order difference formula, i.e. the parameter increment at the current time t is equal to the key parameter component of the parameter at time t minus the key parameter component at time t-1, forming an increment data set reflecting the dynamic variation characteristics of the parameter. The convergence analysis adopts a variance convergence criterion. When the variance of the parameter increment in consecutive multiple time windows shows a decreasing trend and the variance value is less than a preset convergence test threshold, it is determined that the parameter has convergence characteristics. The convergence test threshold is set to 5% of the initial value of the parameter as a baseline threshold. According to the sensitivity characteristics of different parameter components, the threshold is adjusted differently. For example, the test threshold of the position control gain is set to 1.2 times the baseline threshold, the test threshold of the speed synchronization coefficient is set to 0.8 times the baseline threshold, and the test threshold of the communication weight factor is set to 1.5 times the baseline threshold. By comparing the absolute value of the increment of each parameter component with the size relationship of the corresponding test threshold, the parameter components whose increments exceed the threshold range are identified and marked as parameter objects that need to be re-optimized.

[0055] The parameter re-optimization process first analyzes the reasons for the excessive parameter increment, including the increase of inter-agent communication delay, the drastic change of relative position, or the sudden change of the number of neighbor nodes. For the parameter fluctuation caused by communication delay, the influence weight of distant nodes is reduced by increasing the decay coefficient of the communication weight factor. For the case of drastic position change, the response sensitivity parameter of the position control gain is adjusted, and the adaptive gain adjustment mechanism is used to dynamically adjust the gain value range according to the position change rate. According to the adjustment results of the parameter re-optimization process, the weighted average method is used to fuse the original parameter value and the optimized adjustment value to generate a new parameter value. The fusion weight is determined according to the degree of parameter increment exceeding the threshold. The greater the increment exceeds the threshold, the higher the weight of the optimized adjustment value. For the position control gain parameter, when the increment exceeds the threshold by more than 50%, the weight of the optimized adjustment value is set to 0.7, and the weight of the original parameter value is set to 0.3. For the speed synchronization coefficient, when the increment exceeds the threshold by more than 30%, the weight of the optimized adjustment value is set to 0.6, and the weight of the original parameter value is set to 0.4. The re-optimized parameter components are reassembled to form an optimized control parameter set. The parameter consistency test ensures the coordination and compatibility between the parameter components. The parameter consistency test includes parameter value range rationality check and parameter constraint relationship verification to ensure that the optimized parameter set can meet the stability requirements of multi-agent system coordinated control. Through this optimization process, the system can dynamically adjust the control parameters of each agent to ensure that all agents in the multi-agent system maintain effective cooperation and efficient operation in a changing environment.

[0056] In a specific embodiment, predicting the position state of the agent at the next time based on the relative displacement vector, and updating the position index of the agent in the spatial grid based on the position state specifically includes the following steps:

[0057] Identifying the matrix elements in the corrected weight distribution matrix whose values are below the sparsification threshold, and performing sparsification processing on the connection relationship between the agent nodes corresponding to the matrix elements to obtain a sparse weight distribution matrix;

[0058] Calculating the effective number of neighbors of each agent node in the sparse weight distribution matrix, and counting the connection degree distribution of each agent node in the compressed network topology;

[0059] Calculating the difference between the current time agent position coordinates and the previous time position coordinates to obtain the relative displacement vector of the agent, predicting the position state of each agent at the next time based on the displacement vector data of multiple consecutive time steps, and correcting the position state prediction result according to the motion speed and acceleration characteristics of the agent;

[0060] The activity area of the agent is divided into multiple spatial segmentation grids, each spatial segmentation grid corresponds to a position index, and when the predicted position state of the agent exceeds the boundary of the current belonging grid unit, the position index of the agent in the spatial grid is updated.

[0061] Specifically, as Figure 2As shown, the prediction flowchart of the agent position identifies the matrix elements in the corrected weight distribution matrix that are below the sparsification threshold value, and sparsifies the connection relationship between the agent nodes corresponding to these matrix elements. The sparsification threshold value is preset according to the communication range of the agents in the system and the signal strength attenuation characteristics, and is usually set to a small value, such as 0.05. When the weight coefficient of an element in the matrix is less than the sparsification threshold value, it indicates that the contribution of the connection is low and the impact on the overall coordinated control of the system is weak, so the value of these elements will be set to zero to form a sparse matrix. Then, the number of effective neighbors of each agent node in the sparse weight distribution matrix is calculated, and the degree distribution of each agent node in the compressed network topology is counted. This statistical result reflects the degree of change of the network topology after sparsification processing. The difference between the current time agent position coordinates and the previous time position coordinates is calculated to obtain the relative displacement vector of the agent. The instantaneous motion trend of each agent is obtained through position difference. The linear extrapolation method is used to fit the relative displacement vector at the previous time steps as a linear model to predict the position of the agent at the next time. On this basis, the position state prediction result is corrected according to the motion speed and acceleration characteristics of the agent. The motion speed and acceleration characteristics of the agent are key factors affecting the accuracy of the prediction result. The system obtains the speed of the agent by calculating the modulus of the displacement vector, and obtains the acceleration by calculating the change of the speed between adjacent time steps. If the speed of the agent exceeds the preset maximum moving speed, the predicted displacement vector is limited in amplitude to ensure that the predicted position does not exceed the maximum motion range set by the system. In addition, when the motion trajectory of the agent suddenly changes direction or stops abnormally, the prediction result is smoothed by analyzing the acceleration and curvature change of the motion trajectory to reduce the prediction error caused by sudden events. Then, the agent activity area is divided into multiple spatial segmentation grids, each grid element has a unique index identifier, which is used to represent the distribution state of the agent in space. The size of each grid is usually determined according to the communication radius of the agent to ensure that the agents in adjacent grid elements can establish effective communication connection. When the predicted position of the agent exceeds the boundary of the current grid element, the grid index of the agent will be recalculated, and the position index of the agent in the spatial grid will be updated. If the index is not updated, the parameter may be transmitted to the wrong node, resulting in ineffective control. Index updating ensures that parameters can match the correct agent, realize real-time adaptive adjustment, and ensure the stability of agent cooperation in dynamic environment, and will not cause distributed control failure due to incorrect topology information.

[0062] In a specific embodiment, the optimized control parameters are distributed to the corresponding agents according to the weight proportion to form a distributed adaptive adjustment system, which specifically includes the following steps:

[0063] A parameter distribution confirmation mechanism is established, and according to the updated spatial segmentation grid information, the optimized control parameters are distributed to each agent node according to the weight proportion, the position prediction error and the parameter receiving state are monitored, and a distributed parameter adjustment system based on the spatial influence relationship is generated, wherein the distributed parameter adjustment system dynamically updates the control parameters of each agent through the feedback mechanism.

[0064] Specifically, in the distribution of the optimized control parameters to the corresponding agents in proportion to the weights to form a distributed adaptive adjustment system, first, a parameter distribution confirmation mechanism is established, the key of which lies in ensuring that the optimized control parameters can be accurately and reliably delivered to each agent. In specific implementation, the system sends a distribution package containing the optimized control parameters to each agent node, and embeds a confirmation identification code in the distribution package, which is composed of the unique identifier of the sending node and the timestamp, used to verify the integrity and reliability of the distribution data. The receiving node calculates the check value of the identification code to ensure that the distribution package has not been tampered with after receiving the distribution package, and if the check is successful, it sends an acknowledgement signal; if the check fails, it triggers a retransmission request. Next, according to the updated spatial partition grid information, the system distributes the optimized control parameters to each agent node in proportion to the weights, and the spatial partition grid determines the position of each agent and its relative position by dividing the area. First, query the spatial grid position index of each agent, calculate the receiving proportion of each node according to the weight coefficients stored in the adjacency matrix, and the weight coefficients reflect the connection strength between the agent and the neighbor nodes. For example, if the weight coefficient of an agent is 0.4, it receives 40% of the total optimized parameters. Based on this proportion, the system distributes the corresponding control parameters to the target node through grid index routing, which can reduce invalid transmission, improve distribution efficiency, and at the same time use spatial grid to reduce the transmission path, thereby improving the accuracy and real-time performance of communication. Subsequently, specific adjustments are made in terms of monitoring the position prediction error and parameter reception status. The monitoring steps include comparing the predicted position of the agent with the actual coordinates, calculating the prediction error, and if the error exceeds a predetermined threshold (e.g. 0.1 meters), it is recorded as an anomaly. In addition, the parameter reception status of each node is checked to confirm whether each agent has successfully received and processed the distributed control parameters. If a node fails to successfully receive the parameters within a predetermined time, the system will trigger a retransmission mechanism to ensure that the control parameters of each node are updated in a timely manner. On this basis, the system generates a distributed parameter adjustment system based on spatial influence relationships, dynamically updates the control parameters of each agent through a feedback mechanism. In specific implementation, first, based on the monitored error and parameter reception status, the system calculates the spatial influence strength between agents, which reflects the physical distance and communication quality between nodes. According to the spatial influence strength, the system establishes a feedback loop mechanism, and each node feeds back the local error to the neighbor nodes. After aggregating the feedback information, the control parameters are updated. For example, if the communication quality between agent A and agent B fluctuates and the control parameters of agent A exceed the threshold within the error range, agent A will feed back the error information to agent B so that agent B can adjust its control strategy. Through this feedback mechanism, the system can form adaptive adjustment to ensure that the multi-agent system always maintains coordination and stability in a dynamically changing environment.In this way, the entire multi-agent system can be dynamically adjusted according to real-time feedback data, and the control parameters of each agent can be continuously optimized to ensure rapid adaptation in complex tasks and environmental changes. This distributed adaptive adjustment system based on the feedback mechanism not only improves the stability of the system, but also effectively improves the overall efficiency of the multi-agent system, ensuring that the agent group always maintains efficient collaboration in high-dynamic and high-uncertainty environments.

[0065] As shown in Figure 3 The figure shows the comparison of the effects of the traditional method and the method of the application on multiple key performance indicators. The traditional method usually relies on simple distance measurement and fixed weight allocation rules for multi-agent system control. The method of the application proposes a distributed control method based on agent clusters, which can more effectively adapt to changing environments and improve the coordination ability of multi-agent systems by dynamically adjusting communication weights and agent control parameters. The figure includes weight change stability comparison, network efficiency comparison, weight distribution balance comparison, and comprehensive performance score comparison of the traditional method and the method of the application. As can be seen from the figure, the weight change is relatively stable under the method of the application, while the weight change fluctuates greatly under the traditional method, showing lower stability. Due to the dynamic adjustment of the relative position between nodes, signal strength, and communication quality, the communication efficiency is effectively improved, which is significantly better than the traditional method. The method of the application maintains high weight distribution balance, while the entropy value of the traditional method is low, which may cause some nodes to have excessive influence, causing some agents in the system to be in a less favorable state. The method of the application is superior to the traditional method in terms of comprehensive performance, and the traditional method has large fluctuations in performance, especially in network efficiency and stability. The application ensures the improvement of overall performance through dynamic adjustment, position prediction, and communication quality optimization. In summary, the method of the application significantly improves the stability, efficiency, and balance of weight allocation of the system by introducing dynamic adjustment mechanisms, signal propagation attenuation models between nodes, importance index calculation, and dynamic optimization. These advantages enable multi-agent systems to collaborate more efficiently, especially in complex and changing environments, to better adapt to network topology changes and communication quality fluctuations, thereby improving the overall performance of the system.

[0066] The above describes a distributed control method based on agent clusters in an embodiment of the application. The following describes a distributed control system based on agent clusters in an embodiment of the application. Please refer to Figure 4 An embodiment of a distributed control system based on agent clusters in an embodiment of the application includes:

[0067] The acquisition module acquires the real-time position coordinates of each agent in the multi-agent system and calculates the distance between the agents. Based on the distance, an adjacency matrix between the agents is constructed.

[0068] The correction module determines the connection weight of each agent based on the adjacency matrix and signal propagation characteristics, generates a weight distribution matrix, continuously monitors the communication quality between agents, and dynamically adjusts the weight distribution matrix if the communication quality deviates from the preset range to obtain a corrected weight distribution matrix.

[0069] The optimization module performs weighted fusion processing on the control parameters of each agent using the corrected weight distribution matrix to obtain a fusion control parameter vector, acquires key parameter components in the fusion control parameter vector, analyzes the change trend and convergence characteristics of the key parameter components, and optimizes the control parameters of the agent.

[0070] The control module calculates the relative displacement vector between each agent, predicts the position state of the agent at the next time based on the relative displacement vector, updates the position index of the agent in the spatial grid based on the position state, distributes the optimized control parameters to the agent corresponding to the position index, and forms a distributed adaptive adjustment system.

[0071] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, system and unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0072] The integrated unit, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the embodiments of the method of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0073] The above and the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A distributed control method based on an intelligent agent cluster, characterized in that, The method includes: Step S1: Obtain the real-time position coordinates of each agent in the multi-agent system, calculate the distance between agents, and construct an adjacency matrix between agents based on the distance; Step S2: Based on the adjacency matrix and signal propagation characteristics, determine the connection weights of each agent, generate a weight allocation matrix, continuously monitor the communication quality between agents, and if the communication quality deviates from the preset range, dynamically adjust the weight allocation matrix to obtain a corrected weight allocation matrix. Step S3: Use the modified weight allocation matrix to perform weighted fusion processing on the control parameters of each agent to obtain a fused control parameter vector, obtain the key parameter components in the fused control parameter vector, analyze the changing trend and convergence characteristics of the key parameter components, and optimize the control parameters of the agent. The weighted fusion processing of the control parameters of each agent to obtain the fused control parameter vector includes: normalizing the weight allocation matrix, obtaining the weight coefficients corresponding to each agent node in the normalized weight allocation matrix, obtaining the control parameters of the effective neighbor nodes of each agent, including position control parameters, velocity control parameters and acceleration control parameters, performing numerical weighted calculation of the control parameters of the effective neighbor nodes according to the weight coefficients, and fusing each type of control parameter of all agents according to the node number order to obtain the fused control parameter vector. The process of analyzing the changing trends and convergence characteristics of the key parameter components and optimizing the control parameters includes: extracting position control gain, speed synchronization coefficient, and communication weight factor as key parameter components from the fused control parameter vector; obtaining a parameter increment sequence by calculating the difference between the current parameter value and the parameter value of each key parameter component within a continuous time step; calculating the ratio of variance to mean of the parameter increment sequence and setting corresponding convergence test thresholds based on different key parameter components; determining whether the key parameter components have convergence based on the ratio of variance to mean and the convergence test thresholds; initiating a parameter re-optimization process when the parameter increment exceeds the convergence test threshold; and generating new control parameter values ​​by fusing the original parameter values ​​and optimized adjustment values ​​using a weighted average method based on the adjustment results of the parameter re-optimization process. Step S4: Calculate the relative displacement vector between each agent, predict the position state of the agent at the next moment based on the relative displacement vector, update the position index of the agent in the spatial grid based on the position state, and distribute the optimized control parameters to the agents corresponding to the position indices to form a distributed adaptive adjustment system.

2. The method according to claim 1, characterized in that, Calculating the distance between agents and constructing an adjacency matrix between agents based on the distance includes: Centered on each agent, other agents whose distance is less than or equal to the effective communication distance are selected as neighbor nodes. The connectivity of each agent's neighbor nodes is verified. Nodes that pass the verification are defined as the agent's effective neighbor nodes. An adjacency matrix reflecting the connectivity between nodes is generated based on the effective neighbor nodes. The adjacency matrix is ​​indexed by each agent node, and the matrix element values ​​represent the distance data and connection status between agents.

3. The method according to claim 1, characterized in that, Based on the adjacency matrix and signal propagation characteristics, the connection weights of each agent are determined, and a weight allocation matrix is ​​generated, including: The distance data of agent node pairs is obtained from the adjacency matrix. The distance data is input into a preset signal propagation attenuation function to calculate the signal attenuation degree between agent node pairs. The reciprocal of the signal attenuation degree is defined as the spatial influence intensity. The importance index of each agent node is obtained, including degree centrality, betweenness centrality and neighborhood density coefficient. The connection weight of each agent node is obtained by weighted averaging based on the importance index. The product of the spatial influence intensity and the connection weight is used as the weighted influence intensity of the agent node. Using the agent node pair as the index and the weighted influence intensity as the element value of the matrix, a weight allocation matrix is ​​generated.

4. The method according to claim 3, characterized in that, The importance metrics for each agent node include: The number of connecting edges for each agent node in the adjacency matrix is ​​counted and defined as the degree centrality index of that agent node. Find the shortest path between all pairs of agent nodes, and count the number of times each agent node appears in the shortest path. Define the betweenness centrality of the agent node. Centered on each agent node, the number of neighboring nodes is counted within a preset radius. The neighborhood density coefficient is obtained by weighting the distances of neighboring nodes using a Gaussian kernel function.

5. The method according to claim 1, characterized in that, The weight allocation matrix is ​​dynamically adjusted to obtain the corrected weight allocation matrix, which includes: If the communication quality between agents deviates from the preset range, the agent node pair is defined as an abnormal node pair. The matrix element position of the abnormal node pair is located in the weight allocation matrix, and the weighted influence intensity at the element position is obtained as the weight coefficient. The communication quality indicators between the agents include data transmission delay and signal attenuation. The deviation values ​​of data transmission delay and signal attenuation are calculated. Based on the deviation values, a correction coefficient is calculated by linear interpolation. The weight coefficients in the weight allocation matrix are dynamically adjusted using an exponential decay function and the correction coefficient to obtain the corrected weight allocation matrix.

6. The method according to claim 1, characterized in that, Predicting the agent's position state at the next moment based on the relative displacement vector, and updating the agent's position index in the spatial grid based on the position state, includes: Identify matrix elements in the corrected weight allocation matrix whose values ​​are below the sparsity threshold, and perform sparsification processing on the connection relationships between the agent nodes corresponding to the matrix elements to obtain a sparse weight allocation matrix. Calculate the number of effective neighbors for each agent node in the sparse weight allocation matrix, and statistically analyze the connectivity distribution of each agent node in the compressed network topology. The relative displacement vector of the agent is obtained by calculating the difference between the current position coordinates and the previous position coordinates. Based on the displacement vector data of multiple consecutive time steps, the position state of each agent at the next time step is predicted. The position state prediction results are corrected according to the agent's motion velocity and acceleration characteristics. The agent's activity area is divided into multiple spatial grids, each corresponding to a position index. When the agent predicts a position state that exceeds the boundary of its current grid cell, the agent's position index in the spatial grid is updated.

7. The method according to claim 1, characterized in that, The optimized control parameters are distributed to the corresponding agents according to their weights, forming a distributed adaptive adjustment system, including: A parameter distribution confirmation mechanism is established. Based on the updated spatial segmentation grid information, the optimized control parameters are distributed to each agent node according to the weight ratio. The position prediction error and parameter reception status are monitored to generate a distributed parameter adjustment system based on spatial influence relationship. The distributed parameter adjustment system dynamically updates the control parameters of each agent through a feedback mechanism.

8. A distributed control system based on an agent cluster, used to implement the distributed control method based on an agent cluster as described in any one of claims 1-7, characterized in that, The system includes: The acquisition module acquires the real-time position coordinates of each agent in the multi-agent system, calculates the distance between agents, and constructs an adjacency matrix between agents based on the distance. The correction module determines the connection weights of each agent based on the adjacency matrix and signal propagation characteristics, generates a weight allocation matrix, continuously monitors the communication quality between agents, and dynamically adjusts the weight allocation matrix if the communication quality deviates from a preset range to obtain a corrected weight allocation matrix. The optimization module uses the modified weight allocation matrix to perform weighted fusion processing on the control parameters of each agent to obtain a fused control parameter vector. It then obtains the key parameter components in the fused control parameter vector, analyzes the changing trends and convergence characteristics of these key parameter components, and optimizes the control parameters of the agents. Specifically, the weighted fusion processing of the control parameters of each agent to obtain the fused control parameter vector includes: normalizing the weight allocation matrix; obtaining the weight coefficients corresponding to each agent node in the normalized weight allocation matrix; obtaining the control parameters of each agent's effective neighbor nodes, including position control parameters, velocity control parameters, and acceleration control parameters; numerically weighting the control parameters of the effective neighbor nodes according to the weight coefficients; and fusing each type of control parameter of all agents according to node number order to obtain a fused control parameter vector. The process of optimizing the control parameters involves: extracting position control gain, speed synchronization coefficient, and communication weight factor from the fused control parameter vector as key parameter components; obtaining a parameter increment sequence by calculating the difference between the current parameter value and the previous parameter value of each key parameter component within a continuous time step; calculating the variance-to-mean ratio of the parameter increment sequence and setting corresponding convergence test thresholds based on different key parameter components; determining whether the key parameter components have convergence based on the variance-to-mean ratio and the convergence test thresholds; initiating a parameter re-optimization process when the parameter increment exceeds the convergence test threshold; and generating new control parameter values ​​by fusing the original parameter values ​​and optimized adjustment values ​​using a weighted average method based on the adjustment results of the parameter re-optimization process. The control module calculates the relative displacement vector between each agent, predicts the position state of the agent at the next moment based on the relative displacement vector, updates the position index of the agent in the spatial grid based on the position state, and distributes the optimized control parameters to the agents corresponding to the position indices, forming a distributed adaptive adjustment system.

Images