Search

CN-121995772-A - Multi-agent distributed coordination control method driven by heterogeneous perception neural network

CN121995772ACN 121995772 ACN121995772 ACN 121995772ACN-121995772-A

Abstract

The invention provides a multi-agent distributed coordination control method driven by a heterogeneous perception neural network, which relates to the technical field of multi-agent distributed coordination control, and comprises the steps of defining a heterogeneous distributed nonlinear system formed by a plurality of heterogeneous agents, establishing a distributed heterogeneous multi-agent system model by establishing a heterogeneous intelligent dynamics equation and a communication topological structure, establishing an environment-aware heterogeneous depth controller, carrying out fusion processing on multi-source heterogeneous information by dynamically adjusting neighbor weights, outputting intelligent agent state information and environment perception information, dynamically adjusting communication connection by utilizing the intelligent agent state information and the environment perception information output by the controller, reconstructing a communication topology, realizing distributed coordination control of the heterogeneous multi-agents by utilizing the reconstructed stable communication topology, optimizing a control strategy, establishing a convergence theoretical framework, uniformly processing multiple constraints, simplifying the controller design, and realizing zero collision coordination movement of heterogeneous formation in a complex obstacle environment.

Inventors

  • YANG YANG
  • Luan Tianyun
  • XU ZIHAN
  • LI MINGQIU
  • HUANG FUZHONG

Assignees

  • 长春理工大学

Dates

Publication Date
20260508
Application Date
20260408

Claims (10)

  1. 1. The multi-agent distributed coordination control method driven by the heterogeneous perception neural network is characterized by comprising the following steps of: Step 1, defining a heterogeneous distributed nonlinear system composed of a plurality of heterogeneous intelligent agents, and establishing a distributed heterogeneous multi-intelligent system model by establishing a heterogeneous intelligent agent dynamics equation and a communication topological structure; Step 2, constructing an environment-aware heterogeneous depth controller, and dynamically adjusting neighbor weights to fuse multi-source heterogeneous information and output intelligent body state information and environment-aware information; Step 3, dynamically adjusting communication connection and reconstructing communication topology by using intelligent body state information and environment sensing information output by the controller; And 4, realizing distributed coordination control of heterogeneous multi-agent by utilizing the reconstructed stable communication topology, and optimizing a control strategy.
  2. 2. The heterogeneous sensory neural network driven multi-agent distributed coordination control method according to claim 1, wherein in the step 1, an agent dynamics equation at time t+1 is expressed as: ; Wherein, the To include subsystem Node set of its neighbor, each subsystem at time t+1 State vector of (2) is The control input at time t is , For the unknown process noise at time t, For heterogeneous parameter sets, different subsystems i have different qualities Radius of geometry Maximum thrust Coefficient of resistance ; The global system dynamics equation formed by all subsystems at time t+1 is as follows: ; Wherein, the state vectors of the global system at the time t and the time t+1 are respectively as follows 、 The control input of the global system at the moment t is The unknown process noise and heterogeneous parameter sets of the global system at the moment t are respectively 、 。
  3. 3. The heterogeneous sensory neural network driven multi-agent distributed coordination control method according to claim 2, wherein in the step 2, the agents Is defined as a dynamic system: ; ; ; Wherein, the For the t moment intelligent agent Is provided with a sensing information of (a), In order to be a heterogeneous perceptual information fusion function, Agent for t time and t+1 time Is provided with a controller internal state of the controller, As a state vector for agent j, Is an intelligent body Is used to update the state updating function of (c), Is an intelligent body Is used as a control function of the (c), Is a set of obstacle environments.
  4. 4. The heterogeneous sensory neural network driven multi-agent distributed coordination control method of claim 3, wherein the agents are defined Layered perceptual information of (2) The method comprises the following steps: , individual tracking errors, neighbor relative states, formation configuration deviations, and context awareness information are represented, respectively.
  5. 5. The heterogeneous perception neural network driven multi-agent distributed coordination control method of claim 4, wherein the heterogeneous perception control network of the whole heterogeneous depth controller adopts a distributed architecture, and each agent only needs to communicate with neighbors: ; The controller updates the equation: ; Wherein the method comprises the steps of As the relative state information, For the t moment intelligent agent And Heterogeneous weights between.
  6. 6. The heterogeneous sensory neural network driven multi-agent distributed coordination control method according to claim 5, wherein in the step 3, the network topology is controlled based on the environmental sensory information Dynamically adjusting to adapt to environmental constraints; Environmental weighting The method comprises the following steps: ; communication feasibility function The method comprises the following steps: ; Is an intelligent body And At the moment of Is a function of the environmental weight of the system; is a safety constraint function; Is an intelligent body And Is a position vector of (2); Is a set of obstacle environments; Is an intelligent body And LOS is a sight line reachability judgment function; Is an attenuation parameter; as a function of obstacle distance; based on the security constraints, a security constraint function is defined: ; The Euclidean distance between two intelligent agents; Is an additional safety distance; is a smoothing parameter of the security constraint.
  7. 7. The heterogeneous sensory neural network driven multi-agent distributed coordination control method of claim 6, wherein a topology reconfiguration process of the heterogeneous multi-agent system in an environment is considered, and the topology reconfiguration process maintains system connectivity if heterogeneous weights meet the following conditions: connectivity maintenance conditions: ; for the second small characteristic value to be the second small characteristic value, Is the minimum characteristic value; Heterogeneous compatibility conditions: ; For the purposes of the heterogeneous metrics, Is a critical value; reconstruction smoothness conditions: ; Network topology for time t+1 Network topology at time t Is used for the topology change rate of (a), Is a rate of change threshold; Environmental adaptation conditions: ; as a result of the minimum weight threshold value, And converge to an environmentally adapted optimal topology The method meets the following conditions: 。
  8. 8. The heterogeneous sensory neural network driven multi-agent distributed coordination control method of claim 7, wherein a heterogeneous sensory distributed state estimator is designed for each agent Maintaining local state estimates: ; Wherein the method comprises the steps of A posterior state estimate of agent i based on observations at time t is represented, For the heterogeneous kalman estimation function, For the a priori state estimation, Is an observation vector; fusion of neighbor state estimation by heterogeneous weights : ; Fusion weights And (3) adaptively adjusting according to the estimation precision: ; Wherein, the A posterior state estimation of the intelligent agent j based on the observation at the moment t; Priori covariance matrix Is used for the track of (a), Is a numerical stable term.
  9. 9. The heterogeneous sensory neural network driven multi-agent distributed coordination control method of claim 8, wherein the heterogeneous multi-agent coordinated learning objective function The definition is as follows: ; Wherein, the In order to integrate the performance index of the product, Is a heterogeneous collaborative enhancement item; Converging to a globally optimal solution : ; For a limited time Such that: ; Wherein the accurate convergence rate is 。
  10. 10. The heterogeneous perception neural network driven multi-agent distributed coordination control method of claim 9, wherein coordination control of a heterogeneous multi-agent system is established, multi-level constraint is processed uniformly, and a comprehensive constraint function is constructed : ; Wherein each constraint is numbered as By using To represent the first The constraint weight is as follows 。

Description

Multi-agent distributed coordination control method driven by heterogeneous perception neural network Technical Field The invention relates to the technical field of multi-agent distributed coordination control, in particular to a multi-agent distributed coordination control method driven by a heterogeneous perception neural network. Background With the rapid development of artificial intelligence technology, each intelligent body in the heterogeneous multi-intelligent body system has different capability characteristics, dynamics characteristics and task roles, however, the coordination control of the heterogeneous multi-intelligent body system faces unprecedented challenges, the capability difference among the intelligent bodies makes the traditional control method based on isomorphic assumption invalid, and a new heterogeneous modeling and control framework needs to be established. Multiple constraint control systems in complex dynamic environments are required to have strong environmental adaptation capability and real-time decision capability. The safety key application puts strict requirements on the stability and reliability of a control system, and the traditional control theory has the problems of difficult modeling, high computational complexity and the like when processing a high-dimensional nonlinear system. Disclosure of Invention In order to explain the technical problems, the invention provides a multi-agent distributed coordination control method driven by a heterogeneous perception neural network, which comprises the following steps: Step 1, defining a heterogeneous distributed nonlinear system composed of a plurality of heterogeneous intelligent agents, and establishing a distributed heterogeneous multi-intelligent system model by establishing a heterogeneous intelligent agent dynamics equation and a communication topological structure; Step 2, constructing an environment-aware heterogeneous depth controller, and dynamically adjusting neighbor weights to fuse multi-source heterogeneous information and output intelligent body state information and environment-aware information; Step 3, dynamically adjusting communication connection and reconstructing communication topology by using intelligent body state information and environment sensing information output by the controller; And 4, realizing distributed coordination control of heterogeneous multi-agent by utilizing the reconstructed stable communication topology, and optimizing a control strategy. Preferably, in the step 1, the equation of the dynamics of the agent at time t+1 is expressed as: ; Wherein, the To include subsystemNode set of its neighbor, each subsystem at time t+1State vector of (2) isThe control input at time t is,For the unknown process noise at time t,For heterogeneous parameter sets, different subsystems i have different qualitiesRadius of geometryMaximum thrustCoefficient of resistance; The global system dynamics equation formed by all subsystems at time t+1 is as follows: ; Wherein, the state vectors of the global system at the time t and the time t+1 are respectively as follows 、The control input of the global system at the moment t isThe unknown process noise and heterogeneous parameter sets of the global system at the moment t are respectively、。 Preferably, in the step 2, the agentIs defined as a dynamic system: ; ; ; Wherein, the For the t moment intelligent agentIs provided with a sensing information of (a),In order to be a heterogeneous perceptual information fusion function,Agent for t time and t+1 timeIs provided with a controller internal state of the controller,As a state vector for agent j,Is an intelligent bodyIs used to update the state updating function of (c),Is an intelligent bodyIs used as a control function of the (c),Is a set of obstacle environments. Preferably, an agent is definedLayered perceptual information of (2)The method comprises the following steps: , individual tracking errors, neighbor relative states, formation configuration deviations, and context awareness information are represented, respectively. Preferably, the heterogeneous awareness control network of the whole heterogeneous depth controller adopts a distributed architecture, and each intelligent agent only needs to communicate with the neighbor: ; The controller updates the equation: ; Wherein the method comprises the steps of As the relative state information,For the t moment intelligent agentAndHeterogeneous weights between. Preferably, in the step 3, the network topology is controlled based on the context awareness informationDynamically adjusting to adapt to environmental constraints; Environmental weighting The method comprises the following steps: ; communication feasibility function The method comprises the following steps: ; Is an intelligent body AndAt the moment ofIs a function of the environmental weight of the system; is a safety constraint function; Is an intelligent body AndIs a position vector of (2); Is a set