Multi-agent two-part tracking control method based on neural network coupled disturbance estimator
Through the neural network coupled cascade uncertain perturbation estimator method, the robust two-part tracking control problem in hybrid order heterogeneous multi-agent system is solved, the system robustness and control accuracy are improved, the application scenario is expanded, and the solution is solved. Problems of insufficient application in the prior art.
Patent Information
- Application Number
- CN202210851725.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-07-20
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2042-07-20
AI Technical Summary
The prior art has failed to effectively solve the problem of robust two-part tracking control in hybrid order heterogeneous multiagent systems, especially inadequate application in mobile target capture.
Using the neural network coupled cascade uncertain perturbation estimator (NN-CUDE) method, a fractional and integer-order model between leaders and followers was established, and a robust two-part tracking control protocol was designed to estimate external perturbations through neural networks. Model uncertainty is used and compensated using cascading UDE algorithm to improve system robustness.
Improve the control robustness of multi-agent system, reduce the consumption of algorithm computing resources, expand the application scenario, solve the problems of mismatch uncertainty and disturbance, reduce the vibration phenomenon, and realize the heterogeneous multi-agent Robust two-part tracking control.
Smart Images

Figure CN115097734B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of cooperative control of multi-agent systems, and in particular relates to a multi-agent two-part tracking control method of a neural network coupled disturbance estimator. Background Art
[0002] The swarm intelligence of multi-agent systems (MAS) is usually reflected in cluster behaviors, which can be expressed in various forms. Convergence is more common in collaboration, but in nature, dispersion is also common, such as fish schools dispersing when encountering obstacles and ant colonies separating on the way to food. Similarly, for multi-agent systems, such as covering, formation, and encirclement, the dispersed behavior of agents plays an important role, and related research has important application value in both civil and military applications. Bipartite division of agent networks is an effective way to solve the dispersion of agents. Bipartite consistency is a multi-agent behavior based on a structurally balanced symbolic network. In the bipartite consistency problem, the order of agents in current research has been extended from first order to second order or even higher order, and its dynamic model has gradually extended from linear to nonlinear. In addition, the leader-follower collaboration structure has also been extended to symmetric divergent behavior, namely the bipartite tracking problem, which has attracted widespread attention for its potential application in mobile target capture and group confrontation.
[0003] On the other hand, swarm intelligence occurs not only between homogeneous agents, but also between heterogeneous agents. In air-ground collaboration scenarios, the motion of drones and ground vehicles on viscoelastic surfaces (grass or muddy roads) or in special weather (dust rain, sand, snow, storms) exhibits different dynamic characteristics, the latter of which is expressed in the form of fractional differential equations. It is difficult to accurately model moving objects, and agents are always disturbed by the environment in actual tasks. This inevitable uncertainty leads to the heterogeneity of the group, and the above factors are likely to affect the stability of the system.
[0004] Many existing studies have attempted to solve the problem of robust bipartite tracking control and have innovatively proposed various methods to reduce or weaken the impact of uncertain conditions on system stability and control accuracy. However, previous studies have not fully considered the robust bipartite tracking control of mixed-order heterogeneous multi-agents, which has important application value in mobile target capture. Summary of the invention
[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a heterogeneous multi-agent two-part tracking control method based on a neural network coupled cascade uncertain disturbance estimator (NN-CUDE), which can improve the robustness of system control and reduce the resource consumption of algorithm calculation and expand the application scenarios of the algorithm.
[0006] The object of the present invention is achieved through the following technical solution: A multi-agent two-part tracking control method of a neural network coupled disturbance estimator comprises the following steps:
[0007] S1. Establish a fractional-order model of the leader agent and an integer-order model of the follower agent in the multi-agent system to form a heterogeneous multi-agent system.
[0008] S2. For multi-agent systems, a fusion framework of neural network coupled cascade uncertain disturbance estimator is designed to estimate external disturbance and model uncertainty in two-part tracking control respectively;
[0009] S3, define the two-part tracking error, compensate the estimated value obtained in S2 into the controller, and then design a robust two-part tracking control protocol;
[0010] S4, designing a parameter adaptive law for the control protocol established in S3, and completing the design of the adaptive parameters of the neural network;
[0011] S5. Apply the control protocol to the dynamic model of the follower agent, so that the multi-agent under the symbolic network completes the two-part division, one group of agents tracks the trajectory of the leader, and the other group moves away from the leader's trajectory. The trajectory is symmetric with the leader's trajectory about the x-axis, completing the two-part division of the followers and the two-part tracking of the leader.
[0012] Furthermore, the step S1 includes the following sub-steps:
[0013] S11. There is a leader and n followers in a multi-agent system, which are connected through a symbolic network G. The fractional-order model of the leader in the multi-agent system is established as follows:
[0014]
[0015] in, represents the fractional order model, the value range of the fractional order is 0<q≤2, f0(r0(t),t) is an unknown smooth function that other agents cannot obtain, and r0(t) is the position variable of the leader agent;
[0016] S12. The integer-order model of n follower agents is as follows:
[0017]
[0018] in denote the position, velocity and control input of the ith agent respectively, represents an n-dimensional real vector, They represent the first-order derivatives of position and velocity with respect to time; f i (ri ,t) and g i (r i ,v i , t) represent the combined input of model uncertainty and external disturbance in the position loop and velocity loop respectively; the follower cannot obtain the model information of the leader, but can obtain the leader's position information r0(t);
[0019] S13. To realize the two-part tracking control of heterogeneous multi-agents, the following definitions are made:
[0020]
[0021] where w i Indicates that if the states of the ith agent and the leader eventually converge, then w i =1; otherwise w i =-1; when the above conditions are met, it means that the two-part tracking control is completed.
[0022] Furthermore, the step S2 includes the following sub-steps:
[0023] S21. To simplify the nonlinear model of leaders, the following simplifications are made:
[0024]
[0025] in Represents the fractional-order model of the leader; based on the parameter linearization of the neural network, β(r,t) is expressed as:
[0026]
[0027] in The known radial basis function, θ β is an unknown constant vector, and the approximation error of the neural network is ε β ; The radial basis function RBF is defined as:
[0028]
[0029] Where r is the n-dimensional input vector, which refers to the position input of all agents; η β,k is the smoothing parameter; is the center of the basis function; h β is the number of sensing nodes;
[0030] Furthermore, with the help of the radial basis function of the neural network, the i-th agent can calculate the unknown function β i The estimated value of (r,t) is expressed as:
[0031]
[0032] The approximation error of the neural network is further expressed as Represents the difference between the estimated value and the true value; is the constant vector θ for the i-th agent βi An estimated value of
[0033] S22. For the model uncertainty and external disturbance of the followers in the multi-agent, a cascade UDE is designed to estimate them: First, the unknown function g is designed i (r i ,v i ,t) The method is: the control input is divided into the following two parts:
[0034]
[0035] where u i0 (t) is the nominal control input for the multi-agent system to achieve the control target; through the design process of UDE, the estimate of the unknown function is constructed as:
[0036]
[0037] Among them G d (s) is a filter that ensures the physical feasibility of the system. represents the inverse Laplace transform; the filter is selected as a first-order filter, as follows:
[0038]
[0039] where τ d is the time scale parameter of the first-order filter; s represents the complex frequency domain; the ultimate goal is to use the filter G d The bandwidth of (s) covers the unknown part g i (r i ,v i ,t); then, the estimate of the unknown function is written as:
[0040]
[0041] Convert it to the frequency domain and express it as:
[0042]
[0043] in Indicated The frequency domain form; further transformed in the time domain, the unknown function g is obtained i The estimated value of is:
[0044]
[0045] where vi (0) is the initial velocity of the i-th agent;
[0046] S23. Design of uncertain function f i (r i ,t) The specific method is: according to the principle of backstepping control, the state v i (t) as a virtual control input v i (u i ,t), and then the input is controlled by the actual controller u i Decide, therefore v i (t) is expressed as:
[0047]
[0048] where v i0 (t) is the guaranteed state r i (t) The nominal input of the virtual controller converges to its expected value; thus, the uncertainty function f i (r i ,t) is estimated as:
[0049]
[0050] Then, we get the uncertain function f i (r i ,t) is estimated to be:
[0051]
[0052] where r i (0) is the initial position of the ith agent;
[0053] S24, due to the virtual control input v i0 (t) The information is provided by u i Therefore, the two uncertainties appearing in the differential function are cascaded, which are called cascaded UDEs. Cascaded UDEs are used to estimate the uncertainty in the follower model.
[0054] Furthermore, the step S3 includes the following sub-steps:
[0055] S31, the second tracking error is defined as:
[0056]
[0057] where a ij represents the jth element in the i-th row of the adjacency matrix A in the multi-agent network topology, sgn represents the symbolic function, and b i Represents the weight matrix connected to the agent;
[0058] S32. To achieve the convergence of the two-part tracking error in S31, the following coordinate transformation is designed:
[0059]
[0060] where x i1 is the position error of the two-part tracking, x i2 is the difference between the virtual control input and the real speed;
[0061] S33, virtual reference control input α of position loop i The structure is:
[0062]
[0063] k r is the control gain, α i0 (t) represents the nominal virtual reference control input; according to the properties of the neural network estimation The derivative of is:
[0064]
[0065] S34, further, the complete two-part tracking control input u i for:
[0066]
[0067] The controller is divided into three parts according to its function: the first part is the benchmark controller, which implements two-part tracking control without model uncertainty and disturbance, where k r and k v They are all control gains, satisfy in Express The estimate, Indicates the speed error e vi The second part is to compensate for the uncertainty in the follower model, where yes ζ i satisfy The third part deals with the uncertainty of the leader. The first term is the estimation of the second-order derivative of the leader’s position state. The second term considers the influence of the neural network approximation error. The parameter y i satisfy is the adaptive parameter; in summary, the design of the control protocol is completed.
[0068] Furthermore, the step S4 includes the following sub-steps:
[0069] S41. In order to complete the adaptive estimation of the neural network, the following adaptive law is designed:
[0070]
[0071] where k θ and k δ is an adjustable adaptive gain, They are adaptive parameters and The derivative of
[0072] S42, with the help of the control protocol designed by S41, can complete the adaptive parameters and The updated values of the two are then substituted into the control protocol in S34. The parameters in the control protocol satisfy k r >0,k v >0,k θ >0,k δ >0.
[0073] The present invention provides a method based on neural network cascade UDE (NN-CUDE), combined with adaptive parameter estimation, and proposes a method for heterogeneous multi-agent two-part tracking control under a symbolic network, which improves the robustness of system control and reduces the resource consumption of algorithm calculation, and expands the application scenarios of the algorithm. Compared with the prior art, the advantages of the present invention are reflected in the following aspects:
[0074] (1) For mixed-order (fractional and integer order) agents with mismatched uncertainties and disturbances, a backstepping cascade robust controller is designed to solve the mismatch problem between the kinematics of the leader and follower, the position dynamic loop of the follower, and the control input loop.
[0075] (2) The NN-CUDE-based method combines neural networks and UDE to ensure robustness in cluster control. The neural network is used to estimate the fractional-order model, and the UDE algorithm is used to compensate for the unknown part of the follower, reducing the consumption of computing resources. Compared with the traditional neural network-based method, this method reduces the computational complexity and effectively reduces the occurrence of chattering. BRIEF DESCRIPTION OF THE DRAWINGS
[0076] Figure 1 It is a flow chart of the heterogeneous multi-agent two-part tracking control method of the present invention;
[0077] Figure 2 This is a heterogeneous multi-agent communication topology diagram in this embodiment;
[0078] Figure 3 is the position status response of the leader and the follower in this embodiment;
[0079] Figure 4 is the estimation error of the neural network in this embodiment;
[0080] Figure 5 is the estimation error of the uncertainty and disturbance estimator UDE in this embodiment;
[0081] Figure 6 is the time response of the control input in this embodiment. DETAILED DESCRIPTION
[0082] The present invention proposes a method based on a neural network coupled cascade uncertain disturbance estimator (NN-CUDE) to solve such problems. In the presence of unmatched uncertainty and interference in the system, the robustness of heterogeneous multi-agent two-part tracking control under a symbolic network is guaranteed. The invention has the complementary advantages of neural network-based (applicable to uncontrollable fractional-order models) and UDE-based (high computational efficiency). The technical solution of the present invention is further explained below in conjunction with the accompanying drawings and specific embodiments.
[0083] like Figure 1 As shown, a multi-agent two-part tracking control method of a neural network coupled disturbance estimator of the present invention comprises the following steps:
[0084] S1. Establish a fractional-order model of the leader agent and an integer-order model of the follower agent in a multi-agent system to form a heterogeneous multi-agent system. In military and civilian application scenarios, the task of swarm drones tracking ground mobile target vehicles often occurs. The leader is a ground mobile target vehicle and the follower is a swarm drone. The following sub-steps are included:
[0085] S11. There is a leader and n followers in a multi-agent system. The information interaction model between them can be connected through a symbolic network topology graph G. The fractional order model of the leader in the multi-agent system is established as follows:
[0086]
[0087] in, represents the fractional order model, the value range of the fractional order is 0<q≤2, f0(r0(t),t) is an unknown smooth function that other agents cannot obtain, r0(t) is the position variable of the leader agent, and t is time;
[0088] S12. The integer-order model of n follower agents is as follows:
[0089]
[0090] in denote the position, velocity and control input of the ith agent respectively, represents an n-dimensional real vector, They represent the first-order derivatives of position and velocity with respect to time; f i (r i ,t) and g i (r i ,v i , t) represent the combined input of model uncertainty and external disturbance in the position loop and velocity loop respectively; the follower cannot obtain the model information of the leader, but can obtain the leader's position information r0(t);
[0091] S13. To realize the two-part tracking control of heterogeneous multi-agents, the following definitions are made:
[0092]
[0093] where w i Indicates that if the states of the ith agent and the leader eventually converge, then w i =1; otherwise w i =-1; when the above conditions are met, it means that the two-part tracking control is completed.
[0094] S2. For multi-agent systems, a fusion framework of neural network coupled cascade uncertain disturbance estimator is designed to estimate external disturbance and model uncertainty in two-part tracking control respectively; the uncertainty in the leader agent model is estimated by neural network, and the external disturbance and model uncertainty in the follower model are estimated by UDE; the specific sub-steps include:
[0095] S21. To simplify the nonlinear model of leaders, the following simplifications are made:
[0096]
[0097] in Represents the fractional-order model of the leader; based on the parameter linearization of the neural network, β(r,t) is expressed as:
[0098]
[0099] in The known radial basis function, θ β is an unknown constant vector, and the approximation error of the neural network is ε β ; The radial basis function RBF is defined as:
[0100]
[0101] Where r is an n-dimensional input vector, which specifically refers to the position input of all agents in the present invention; η β, k is the smoothing parameter, which determines the width of the center point of the basis function; (μ βr ,k,μ βt ,k) is the center of the basis function; h β is the number of sensing nodes, which is the number of nodes in the hidden layer of the neural network;
[0102] Furthermore, with the help of the radial basis function of the neural network, the estimated value of the unknown function βi(r,t) by the i-th agent is expressed as:
[0103]
[0104] The approximation error of the neural network is further expressed as Represents the difference between the estimated value and the true value; is the constant vector θ for the i-th agent β The estimated value of i, which can be obtained by subsequently designing an adaptive law to calculate and update parameters (see formula 44);
[0105] S22. For the model uncertainty and external disturbance of the followers in the multi-agent, a cascade UDE is designed to estimate them: First, the estimated value of the unknown function gi(ri,vi,t) is designed The method is: the control input is divided into the following two parts:
[0106]
[0107] where u i0 (t) is the nominal control input for the multi-agent system to achieve the control target; through the design process of UDE, the estimate of the unknown function is constructed as:
[0108]
[0109] Where Gd(s) is a filter that ensures the physical feasibility of the system. represents the inverse Laplace transform; the filter is selected as a first-order filter, as follows:
[0110]
[0111] Where τd is the time scale parameter of the first-order filter; s represents the complex frequency domain; the ultimate goal is to use the filter G d The bandwidth of (s) covers the unknown part g i (r i ,v i ,t), and thus, the unknown function g i (r i ,v i ,t) is estimated as:
[0112]
[0113] Convert it to the frequency domain and express it as:
[0114]
[0115] in Indicated The frequency domain form is further converted to the time domain, and the estimated value of the unknown function is:
[0116]
[0117] where v i (0) is the initial velocity of the i-th agent;
[0118] S23. From S22, we can know that the basic principle of UDE is to use the filtered information of the state variables and control inputs in the closed loop to construct an estimate that is asymptotically close to the actual value of the unknown part in the closed loop. Using similar ideas, we can design the uncertain function f i (r i ,t) The specific method is: according to the principle of backstepping control, the state v i (t) as a virtual control input v i (u i ,t), and then the input is controlled by the actual controller u i Decide, therefore v i (t) is expressed as:
[0119]
[0120] where v i0 (t) is the guaranteed state r i (t) The nominal input of the virtual controller converges to its expected value; thus, the uncertainty function f i (r i ,t) is estimated as:
[0121]
[0122] Then, we get the uncertain function f i (r i ,t) is estimated to be:
[0123]
[0124] where r i (0) is the initial position of the ith agent;
[0125] S24, due to the virtual control input v i0 (t) The information is provided by ui Therefore, the two uncertainties appearing in the differential function (Formula 35 and Formula 38) are cascaded, which are called cascaded UDEs. The cascaded UDEs are used to estimate the uncertainty in the follower model.
[0126] S3, defining the two-part tracking error, compensating the estimated value obtained in S2 into the controller, and then designing a robust two-part tracking control protocol; including the following sub-steps:
[0127] S31, the second tracking error is defined as:
[0128]
[0129] where a ij represents the jth element in the i-th row of the adjacency matrix A in the multi-agent network topology, sgn represents the symbolic function, and b i Represents the weight matrix connected to the agent;
[0130] S32. To achieve the convergence of the two-part tracking error in S31, the following coordinate transformation is designed:
[0131]
[0132] where x i1 is the position error of the two-part tracking, x i2 is the difference between the virtual control input and the real speed;
[0133] S33, virtual reference control input α of position loop i (t) is constructed as:
[0134]
[0135] k r is the control gain, α i0 (t) represents the nominal virtual reference control input, that is, the control input under the condition that the system does not have uncertainty; according to the properties of the neural network estimation The derivative of is:
[0136]
[0137] S34, further, the complete two-part tracking control input u i for:
[0138]
[0139] The controller is divided into three parts according to its function: the first part is the benchmark controller, which implements two-part tracking control under the condition of no model uncertainty and no disturbance, where k r and kv They are all control gains, satisfy in Express Estimate of the speed error e vi The second part is to compensate for the uncertainty in the follower model, where yes ζ i satisfy The third part deals with the uncertainty of the leader. The first term is the estimation of the second-order derivative of the leader’s position state. The second term considers the influence of the neural network approximation error. The parameter y i satisfy is the adaptive parameter; in summary, the design of the control protocol is completed.
[0140] S4, designing a parameter adaptive law for the control protocol established in S3, and completing the design of the neural network adaptive parameters; including the following sub-steps:
[0141] S41. In order to complete the adaptive estimation of the neural network, the following adaptive law is designed:
[0142]
[0143] where k θ and k δ is an adjustable adaptive gain, They are adaptive parameters and The derivative of
[0144] S42, with the help of the control protocol designed by S41, can complete the adaptive parameters and The updated values of the two are then substituted into the control protocol in S34. The parameters in the control protocol satisfy k r >0,k v >0,k θ >0,k δ >0.
[0145] S5. Apply the control protocol to the dynamic model of the follower agent, so that the multi-agent under the symbolic network completes the two-part division, one group of agents tracks the trajectory of the leader, and the other group moves away from the leader's trajectory, and the trajectory is symmetrical with the leader's trajectory about the x-axis, completing the two-part division of the followers and the two-part tracking of the leader; the specific method is: apply the control protocol of S4 to the kinematic model of the follower agent, so that the multi-agent under the symbolic network completes the two-part division, one group of agents tracks the trajectory of the leader, and the other group moves away from the leader's trajectory, and the motion trajectory is symmetrical with the leader's trajectory about the x-axis, completing the two-part division of the followers and the two-part tracking of the leader.
[0146] The heterogeneous multi-agent communication topology diagram in this embodiment is as follows Figure 2 As shown in the figure, there are 5 follower agents and 1 leader agent, where the node numbered 0 represents the leader. According to the principle of bipartitioning under symbolic networks, the above multi-agent system will be divided into two groups V1, V2: {v0, v1, v2}∈V1 and {v3, v4, v5}∈V2. The integer-order model of followers with model uncertainty is:
[0147]
[0148]
[0149] The fractional order model for leaders is:
[0150]
[0151] The number of sensor nodes is h βi =17×17, the width of the basis function is set to η βi,k =6, center (μ βir,k ,μ βit,k ) are uniformly distributed in [-25,25]×[0,50]. The control gain size is k r =0.5,k v =0.5, let the filter parameter τ d =0.01, q=0.4, define the parameter k of the estimator to update the control law θ =2,k δ =0.001.
[0152] Figure 3 Demonstrates the positional state responses of leaders and followers in heterogeneous multi-agent systems. Figure 4 is the estimation error of the neural network, Figure 5 is the estimation error of the uncertainty and disturbance estimator UDE, Figure 6Represents the time response of the control input. From the simulation results, it can be seen that the two-part division of the follower and the two-part tracking control of the leader are realized in the final embodiment. The estimation error of the neural network and the estimation error of the UDE gradually converge to zero. The simulation results show that the proposed control scheme can realize robust two-part tracking control of heterogeneous multi-agents.
[0153] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.
Claims
1. A multi-agent two-part tracking control method with a neural network coupled disturbance estimator, characterized in that: The following steps are involved: S1. Establish a fractional-order model of the leader agent and an integer-order model of the follower agent in the multi-agent system to form a heterogeneous multi-agent system. It includes the following sub-steps: S11. There is a leader and n followers in a multi-agent system, which are connected through a symbolic network G. The fractional-order model of the leader in the multi-agent system is established as follows: in, represents the fractional order model, the value range of the fractional order is 0<q≤2, f0(r0(t),t) is an unknown smooth function that other agents cannot obtain, r0(t) is the position variable of the leader agent, and t is time; S12. The integer-order model of n follower agents is as follows: in denote the position, velocity and control input of the ith agent respectively, represents an n-dimensional real vector, They represent the first-order derivatives of position and velocity with respect to time; f i (r i ,t) and g i (r i ,v i , t) represent the combined input of model uncertainty and external disturbance in the position loop and velocity loop respectively; the follower cannot obtain the model information of the leader, but can obtain the leader's position information r0(t); S13. To realize the two-part tracking control of heterogeneous multi-agents, the following definitions are made: where w i Indicates that if the states of the ith agent and the leader eventually converge, then w i =1; otherwise w i =-1; When the above conditions are met, it means that the two-part tracking control is completed; S2. For multi-agent systems, a fusion framework of neural network coupled cascade uncertain disturbance estimator is designed to estimate external disturbance and model uncertainty in two-part tracking control respectively; it includes the following sub-steps: S21. To simplify the nonlinear model of leaders, the following simplifications are made: in Represents the fractional-order model of the leader; based on the parameter linearization of the neural network, β(r,t) is expressed as: in The known radial basis function, θ β is an unknown constant vector, and the approximation error of the neural network is ε β ; The radial basis function RBF is defined as: Where r is the n-dimensional input vector, which refers to the position input of all agents; η β,k is the smoothing parameter; (μ βr,k ,μ βt,k ) is the center of the basis function; h β is the number of sensing nodes; Furthermore, with the help of the radial basis function of the neural network, the i-th agent can calculate the unknown function β i The estimated value of (r,t) is expressed as: The approximation error of the neural network is further expressed as Represents the difference between the estimated value and the true value; is the constant vector θ for the i-th agent βi An estimated value of S22. For the model uncertainty and external disturbance of the followers in the multi-agent, a cascade UDE is designed to estimate them: First, the unknown function g is designed i Estimated value of The controller is divided into two parts as follows: where u i0 is the nominal control input for the multi-agent system to achieve the control target; through the design process of UDE, the estimate of the unknown function is constructed as: Among them G d (s) is a filter that ensures the physical feasibility of the system. represents the inverse Laplace transform; the filter is selected as a first-order filter, as follows: where τ d is the time scale parameter of the first-order filter; the ultimate goal is to use the filter G d The bandwidth of (s) covers the unknown part g i The spectrum of the unknown smooth function The estimate is written as: Convert it to the frequency domain and express it as: Further transformation in the time domain gives the unknown function g i The estimated value of is: where v i (0) is the initial velocity of the agent; S23. Design of uncertain function f i (r i ,t) The specific method is: according to the principle of backstepping control, the state v i (t) as a virtual control input v i (u i ,t), and then the input is controlled by the actual controller u i Decide, therefore v i (t) is expressed as: where v i0 (t) is the guaranteed state r i (t) The nominal input of the virtual controller converges to its expected value; thus, the uncertainty function f i (r i ,t) is estimated as: Then, we get the uncertain function f i (r i ,t) is estimated to be: where r i (0) is the initial position of the ith agent; S24, due to the virtual control input v i0 (t) The information is provided by u i Therefore, the two uncertainties appearing in the differential function are cascaded, which are called cascaded UDEs. The cascaded UDEs are used to estimate the uncertainty in the follower model. S3, define the two-part tracking error, compensate the estimated value obtained in S2 into the controller, and then design a robust two-part tracking control protocol; S4, designing a parameter adaptive law for the control protocol established in S3, and completing the design of the adaptive parameters of the neural network; S5. Apply the control protocol to the dynamic model of the follower agent, so that the multi-agent under the symbolic network completes the two-part division, one group of agents tracks the trajectory of the leader, and the other group moves away from the leader's trajectory. The trajectory is symmetric with the leader's trajectory about the x-axis, completing the two-part division of the followers and the two-part tracking of the leader.
2. The multi-agent two-part tracking control method of a neural network coupled disturbance estimator according to claim 1 is characterized in that: The step S3 comprises the following sub-steps: S31, the second tracking error is defined as: where a ij represents the jth element in the i-th row of the adjacency matrix A in the multi-agent network topology, sgn represents the symbolic function, and b i Represents the weight matrix connected to the agent; S32. To achieve the convergence of the two-part tracking error in S31, the following coordinate transformation is designed: where x i1 is the position error of the two-part tracking, x i2 is the difference between the virtual control input and the real speed; S33, virtual reference control input α of position loop i The structure is: k r is the control gain, α i0 (t) represents the nominal virtual reference control input; according to the properties of the neural network estimation The derivative of is: S34, further, the complete two-part tracking control input u i for: The controller is divided into three parts according to its function: the first part is the benchmark controller, which implements two-part tracking control under the condition of no model uncertainty and no disturbance, where k r and k v They are all control gains, satisfy in Express The estimate, Indicates the speed error e vi The second part is to compensate for the uncertainty in the follower model, where yes ζ i satisfy The third part deals with the uncertainty of the leader. The first term is the estimation of the second-order derivative of the leader’s position state. The second term considers the influence of the neural network approximation error. The parameter y i satisfy is the adaptive parameter; in summary, the design of the control protocol is completed.
3. The multi-agent two-part tracking control method of a neural network coupled disturbance estimator according to claim 2 is characterized in that: The step S4 comprises the following sub-steps: S41. In order to complete the adaptive estimation of the neural network, the following adaptive law is designed: where k θ and k δ is an adjustable adaptive gain, They are adaptive parameters and The derivative of S42, with the help of the control protocol designed by S41, can complete the adaptive parameters and The updated values of the two are then substituted into the control protocol in S34. The parameters in the control protocol satisfy k r >0,k v >0,k θ >0,k δ >0.
