Multi-single-joint mechanical arm optimization inclusion control method based on output mask function

Abstract

The invention discloses a multi-single-joint mechanical arm optimization inclusion control method based on an output mask function, and the method comprises the steps: firstly carrying out the mathematical modeling of a follower of a multi-single-joint mechanical arm, and then constructing an unknown control coefficient and a state equation of the follower under disturbance according to the physical characteristics of the multi-single-joint mechanical arm, according to the state equation and the transfer function of the converted follower, constructing a nonlinear error conversion function with an auxiliary variable, introducing an output mask function, constructing a Hamiltonian-Jacobian-Bellman equation, and solving a preset time approximate optimal virtual control signal, a preset time approximate optimal controller and corresponding adaptive parameters; a leader estimator and an execution-evaluation network estimate a weight update law for predetermined time control of the multi-single-joint robot arm. According to the scheme, the problem of optimization including control of the multi-single-joint mechanical arm in cooperative control under the mask output state is solved, and meanwhile preset time control of a follower of the multi-single-joint mechanical arm under privacy protection is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial process control, and specifically relates to an optimized inclusion control method for multiple single-joint manipulators based on an output masking function. Background Technique

[0002] In recent years, the cooperative control of single-joint manipulator systems has been a research hotspot in the field of robotics. Compared with a single manipulator system, multiple single-joint manipulator systems have greater flexibility, higher reliability, and can complete more complex tasks.

[0003] When there are multiple manipulators acting as leaders to guide the group movement and achieve the control problem of making the controlled group enter the range formed by the leader group or the leader surrounding the controlled group, it is called the inclusion control problem. To achieve multiple robots entering the target area composed of multiple leaders. Furthermore, in the cooperative control of multiple single-joint manipulators, taking the joint angular displacement of the manipulator as the output quantity, making the output quantities of multiple leaders form a static convex hull, and the output quantities of other single-joint manipulators move into the static convex hull composed of leaders according to the information transmitted by the leaders and followers. However, in the execution of military tasks, the communication between followers usually plays a crucial role. In this case, in order to ensure the reliability of the entire strategy, sensitive information related to the execution of comprehensive decisions cannot be leaked. Therefore, the present invention uses privacy protection and adaptive inversion techniques based on the output masking function, making the research on the inclusion control of multiple single-joint manipulator followers have important theoretical significance and practical value. Therefore, how to achieve pre-specified time inclusion control for multiple-joint manipulator followers under privacy protection has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of the present invention is to provide an optimized inclusion control method for multiple single-joint manipulators based on an output masking function, which can achieve pre-specified time control functions for multiple single-joint manipulators under privacy protection.

[0005] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0006] An optimized inclusion control method for multiple single-joint manipulators based on an output masking function includes the following steps:

[0007] S1. Perform mathematical modeling on multiple single-joint manipulator followers, and construct the state equation of the followers under unknown control coefficients and disturbances according to the physical characteristics of multiple single-joint manipulators;

[0008] S2. According to the state equation and transfer function of the converted followers, construct a non-linear error conversion function with auxiliary variables;

[0009] S3. Introduce the output masking function and define the inclusion error and error variable;

[0010] S4. Based on the first error surface ξ i1 , the ideal optimal virtual control signal and the optimal theory method, establish the optimal cost function of the first subsystem, and approximately solve it based on the adaptive dynamic programming method of the neural network architecture to obtain a new ideal optimal virtual control signal;

[0011] S5. Execute the evaluation neural network to approximately process and obtain the approximate optimal virtual control signal at a predetermined time, design the adaptive parameters corresponding to the approximate optimal virtual control signal at a predetermined time, and the update laws of the leader estimator and the execution-evaluation network estimation weights;

[0012] S6. Based on the second error surface ξ i2 , the ideal optimal virtual controller and the optimal theory method, establish the optimal cost function of the second subsystem, and approximately solve it based on the adaptive dynamic programming method of the neural network architecture to obtain a new ideal optimal controller;

[0013] S7. Execute the evaluation neural network to approximately process and obtain the approximate optimal controller at a predetermined time, and design the adaptive parameters corresponding to the approximate optimal controller at a predetermined time and the update law of the execution-evaluation network estimation weights.

[0014] Further, in the present invention, the method further includes constructing a complete Lyapunov function: And taking the time derivative of it for stability analysis, it can be obtained that where is the minimum eigenvalue of the matrix , I represents the number of followers, and both represent parameter estimation errors, m = 1, 2.

[0015] Further, in the present invention, in the step S2, the expression of the transfer function is:

[0016]

[0017] where T s represents the delay time.

[0018] Further, in the step S2, the expression of the non-linear error conversion function is:

[0019]

[0020] where The two C n functions are expressed as:

[0021]

[0022] where \(m\in Z\) + , \(2m > n\),

[0023] s i1 is an auxiliary variable,

[0024] Furthermore, in the step S3, the expression of the output masking function is:

[0025]

[0026] where \(\omega\) i =\(\{\upsilon\) i , \(\lambda\) i}\), thus, the follower state under the output mask can be expressed as:

[0027]

[0028] where \(\lambda\) im > 0 and \(\upsilon\) im \(\neq\) 0 are both defined parameters.

[0029] Furthermore, in the step S3, the expression of the error included is:

[0030]

[0031] where \(a\) ij = 1 means that the \(i\)-th follower can receive information from the adjacent \(j\)-th follower, otherwise \(a\) ij = 0; \(a\) ir = 1 indicates that the \(i\)-th follower can receive the signal of the leader , otherwise, \(a\) ir = 0;

[0032] represents the protection signal received by the leader, represents the protection signal received from the adjacent follower \(j\);

[0033] The expression of the error variable is:

[0034]

[0035] where \(\xi\) il (\(l = 1, 2\)) represents the error surface, \(r\) i1 represents the filtering error, and the virtual control law \(\alpha\) i1 is used as the input of the filter to obtain the output signal \(\iota\)i Represents a time scalar.

[0036] Furthermore, in the step S4, the specific process of obtaining the new ideal optimal virtual control signal is as follows:

[0037] According to the first error surface ξ i1 , the ideal optimal virtual control signal and the optimal theory method to design the optimal cost function of the first subsystem Derive the optimal cost function Construct the Hamilton-Jacobi-Bellman equation of the first subsystem by taking the derivative. According to Bellman's optimal principle, take the partial derivatives of the Hamilton-Jacobi-Bellman equation of the first subsystem and the ideal optimal virtual control signal to obtain the ideal optimal virtual control signal, and approximately solve the ideal optimal virtual control signal based on the neural network architecture-based adaptive dynamic programming method to obtain the new ideal optimal virtual control signal

[0038] Furthermore, in the step S5, the predetermined time approximate optimal virtual control signal α i1 is:

[0039]

[0040] Where θ i1 , ρ i are positive design parameters; i = 1, 2, 3, …… I, where I represents the number of followers, represents the basis function for the neural network to approximate the first unknown function, is the basis function for the neural network to approximate the first unknown cost function. Represents a positive constant. After taking the derivative of the variable ξ i1 , σ is obtained i1 ; is 's estimated value, is the ideal weight vector of the neural network, is the approximation error, represents 's estimated value, is the optimal weight vector of the neural network corresponding to the execution optimal weight vector; represents the recognition of the leader by the follower i ; The predetermined time function is expressed as:

[0041]

[0042] Where Ti > 0 and are the predetermined convergence time and the defined scale respectively, and n = 2 represents the order of the system;

[0043] The leader estimator, the adaptive law, and the weight update law of the execution-evaluation network are:

[0044]

[0045] where and are positive design parameters satisfying γ ic1 > γ ia1 / 2, and represent respectively and estimated values. is the optimal weight vector of the neural network corresponding to the optimal weight vector of the evaluation.

[0046] Furthermore, in the step S6, the specific method for obtaining the new ideal optimal controller is:

[0047] According to the second error surface ξ i2 , the ideal optimal virtual controller and the optimal theory method to design the optimal cost function of the second subsystem Derive the optimal cost function to construct the Hamilton-Jacobi-Bellman equation of the second subsystem. According to Bellman's optimal principle, take the partial derivatives of the Hamilton-Jacobi-Bellman equation of the second subsystem and the ideal optimal virtual controller to obtain the ideal optimal virtual controller, and approximately solve the ideal optimal virtual controller based on the adaptive dynamic programming method of the neural network architecture to obtain the new ideal optimal controller

[0048] Furthermore, in the step S7, the predetermined time approximate optimal controller u i is:

[0049]

[0050] where are positive design parameters, represent respectively estimated values, is the optimal weight vector of the neural network corresponding to the optimal weight vector of the execution, represents the basis function for the neural network to approximate the second unknown function, is the approximation error, is the basis function for the neural network to approximate the second unknown cost function, and the neural network input and represent positive constants, and ω i2 is the upper bound value of g i2 , where g i2 = 1 / P,

[0051] The predetermined time approximate optimal controller u i The corresponding adaptive parameters and the update law of the execution-evaluation network estimation weights are:

[0052]

[0053] where l i2 , γ ic2 , γ ia2 are positive design parameters that satisfy γ ic2 > γ ia2 / 2, and respectively represent and 's estimated values, is the optimal weight vector of the neural network The corresponding execution and evaluation optimal weight vectors, is a positive constant.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] The present invention takes into account that the followers in multi-robot cooperation need to ensure confidential state information, and introduces an output masking function to ensure that other followers cannot identify the real state message, which may increase the complexity of the inclusion control process of MAS. In addition, by adding a transfer function to the nonlinear mapping function to obtain an auxiliary error, the delay constraint requirement is achieved, avoiding the feasibility criterion and realizing unified control. By constructing an intermediate variable Q i1 to solve the problem of unknown control gain acting on the optimization-based adaptive backstepping control. In addition, the disadvantage of the semi-global boundedness of the error surface caused by the dynamic surface control is effectively avoided by means of the Lyapunov energy function. This solution solves the optimization inclusion problem of multiple single-joint manipulators in cooperative tracking under the output masking state, and realizes the predetermined time control of multiple single-joint manipulators under privacy protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 is a schematic flow diagram of the multi single-joint manipulator of the present invention.

[0057] Figure 2This is the communication topology diagram of the multi-single joint robotic arm of the present invention.

[0058] Figure 3 This is the simulation result diagram of the joint angle position of the multi-single joint robotic arm of the present invention.

[0059] Figure 4 This is the error simulation result diagram of the multi-single joint robotic arm of the present invention.

[0060] Figure 5 This is the simulation result diagram of the leader estimator of the multi-single joint robotic arm of the present invention. Detailed implementation manners

[0061] The present invention will be further described below in conjunction with the accompanying drawings and embodiments. The implementation manners of the present invention include but are not limited to the following embodiments.

[0062] Embodiment

[0063] As Figure 1 、 Figure 2 shown, the optimization of the multi-single joint robotic arm based on the output masking function includes a control method, which includes multiple steps. The technical solutions of each step will be described in detail below:

[0064] Step 1: First, perform mathematical modeling on the multi-single joint robotic arm follower, and then construct the state equation of the follower under unknown control coefficients and disturbances according to the physical characteristics of the multi-single joint robotic arm.

[0065] Specifically, it includes the following steps:

[0066] Step 1.1: First, establish the mathematical model of the multi-single joint robotic arm follower as:

[0067]

[0068] Where o i represents the angular position. A i represents the motor armature current. represents the unknown external disturbance. In the experiment, i = 1, 2, 3, 4 is selected.

[0069] Step 1.2: Define the variables x i1 = o i , g i2 = 1 / P and u i = A. The motion model of the multi-single joint robotic arm follower can be expressed as the following state space equation:

[0070]

[0071] Among them, P, Q, and Ψ can be set according to the single-joint robotic arm system. In the experiment, the robotic arm parameters selected are P = Q = 1 and Ψ = 10.

[0072] Step 2: According to the state equation and transfer function of the converted follower, construct a non-linear error conversion function with auxiliary variables. The specific steps are as follows:

[0073] Step 2.1: To achieve the delayed output constraint, introduce the transfer function where T s represents the delay time.

[0074] Step 2.2: Define the auxiliary variable The non-linear error conversion function is expressed as:

[0075]

[0076] where The two C n functions are expressed as:

[0077]

[0078] where m ∈ z + , 2m > n,

[0079] Step 3: Introduce the output masking function and define the inclusion error and error variables. The specific steps are as follows:

[0080] Step 3.1: Define the output masking function where ω i = {υ i , λ i}.

[0081] Therefore, the protected follower state can be expressed as where λ im > 0 and υ im ≠ 0 are both defined parameters.

[0082] Step 3.2: Due to considering the privacy protection issue, the conventional inclusion error cannot be applied to the subsequent controller design. Therefore, according to the output masking state, the inclusion error of the follower is defined as:

[0083]

[0084] where a ij = 1 means that the i-th follower can receive the information of the adjacent j-th follower, otherwise a ij = 0. a ir = 1 indicates that the i-th follower can accept the leader signal, otherwise, a ir = 0. represents the protection signal received by the leader, represents the protection signal received from the adjacent follower j.

[0085] Step 3.3. Define the error coordinates to obtain the error variables:

[0086]

[0087] where ξ il (l = 1, 2) represents the error surface, r i1 represents the filtering error, and the virtual control law α i1 is used as the input of the filter to obtain the output signal of the filter ι i represents the time scalar.

[0088] Step 4. According to the error surface ξ i1 , the ideal optimal virtual control signal and the optimal theory method, construct the Hamilton-Jacobi-Bellman equation of the first subsystem, and solve the predetermined time approximate optimal virtual control signal α i1 and the corresponding adaptive parameter leader estimator and the execution-evaluation network estimation weight update law

[0089] Step 4.1. Based on the error surface ξ i1 , the ideal optimal virtual control signal and the optimal theory method, design the optimal cost function:

[0090]

[0091] where υ i1 > 0 represents the discount factor, Ω x represents the input set of admissible control, represents the ideal optimal virtual control signal that minimizes , ξ i1 (τ) is abbreviated as ξ i1 , is abbreviated as

[0092] Step 4.2. Derive the following Hamilton-Jacobi-Bellman equation:

[0093]

[0094] where σ i1 σi2 is obtained by taking the derivative of the variable φ i and is

[0095]

[0096] Step 4.3. According to Bellman's principle of optimality, by solving the ideal optimal virtual control signal can be obtained as

[0097]

[0098] where the intermediate variable is defined as

[0099] The ideal optimal virtual control signal in the above formula is approximately solved by an adaptive dynamic programming method based on a radial basis neural network architecture to obtain a new ideal optimal controller for controlling the multi-single-joint manipulator follower.

[0100]

[0101] where is an auxiliary function,[[]] represents the basis function for the neural network to approximate the first unknown function,[[]] represents the weight vector of the neural network,[[]] is the approximation error,[[]] represents a positive constant,[[]] is a positive design parameter. is a positive design parameter. represents the estimated value of,[[]] is the optimal weight vector of the neural network corresponding to the execution optimal weight vector. represents the identification of the follower i with respect to the leader . The predetermined time function is expressed as

[0102]

[0103] where T i > 0 and are the predetermined convergence time and the defined scale respectively. n = 2 represents the order of the system.

[0104] Step 4.4. The optimal gradient function can be decomposed into the following form:[[]]

[0105]

[0106] Step 4.5: Use a radial basis neural network to approximate the unknown cost function for approximation processing, , where is the ideal weight vector, is the basis function for the neural network to approximate the first unknown cost function, is the approximation error, satisfying the relationship is an arbitrarily small positive constant. Adopt a reinforcement learning strategy with an execution-evaluation structure, and use the execution-evaluation neural network to approximately process to obtain an approximate optimal gradient function and a predetermined-time approximate optimal virtual control signal α i1 :

[0107]

[0108] where respectively represent the estimated values of is the optimal weight vector of the neural network the corresponding execution optimal weight vector.

[0109] Step 4.6: Design a leader estimator, the adaptive parameter and the execution-evaluation network estimated weight update law are:

[0110]

[0111] where and are positive design parameters satisfying γ ic1 > γ ia1 / 2.

[0112] Step 5: Similar to the optimization process in Step 3, according to the error surface ξ i2 , the ideal optimal controller and the optimal theory method, construct the Hamilton-Jacobi-Bellman equation of the second subsystem, and solve the predetermined-time approximate optimal controller u i and the corresponding adaptive parameter and the execution-evaluation network estimated weight update law to be used for controlling the multi-single-joint robotic arm follower. Specifically, it includes the following steps:

[0113] Step 5.1: Design the predetermined-time approximate optimal controller for the second subsystem:

[0114]

[0115] where is a positive design parameter, represents the normal amount, and respectively represent and the estimated values of ω i2 is the upper bound value of g i2 ; represents the basis function for the neural network to approximate the second unknown function, is the approximation error, is the basis function for the neural network to approximate the second unknown cost function, and the neural network input and are the optimal weight vectors of the neural network corresponding to the optimal weight vectors for execution and evaluation;

[0116] Step 5.2, design the adaptive law and the execution-evaluation network estimated weight update law as:

[0117]

[0118] where l i2 , γ ic2 , γ ia2 are positive design parameters that satisfy γ ic2 > γ ia2 / 2.

[0119] Step 6, construct the complete Lyapunov function: where and both represent the parameter estimation error, and taking the time derivative of it for stability analysis gives Prove that the multi-single joint manipulator system achieves closed-loop stability, where is a defined scalar, is a positive constant, is the matrix the minimum eigenvalue of, m = 1, 2.

[0120] According to the Lyapunov stability theorem, the optimal control of the multi-single joint manipulator can be achieved by adjusting the parameters. The closed-loop system is a traditional concept in the control field, that is, a feedback control system. The present invention proves that the multi-single joint manipulator achieves closed-loop stability, solves the optimization inclusion problem of the multi-single joint manipulator in cooperative control under the output mask state, and at the same time realizes the preset time control of the multi-single joint manipulator under privacy protection.

[0121] Simulations were carried out using Matlab, and its stability and real-time performance were analyzed. The control objective of the simulation experiment was to move the joint angle positions of the four single-joint manipulators into the static convex hull formed by the leader according to the information transmitted by the leader and the follower, and delayed output constraint control was achieved.

[0122] The initial conditions of the simulation were:

[0123] x 11 (0) = -0.4, x 12 (0) = x 22 (0) = 4, x 21 (0) = 0.5, x 31 (0) = -0.6, x 32 (0) = 5,

[0124] x 41 (0) = -0.6, x 42 (0) = -4,

[0125] Design parameters: Gaussian function: Reference trajectory: Constraint function:

[0126] The design objective of this application is to design an optimization inclusion control method for multi-single-joint manipulators based on an output masking function. The present invention solves the optimization inclusion problem of multi-single-joint manipulators in cooperative tracking under the output masking state, and realizes the predetermined-time control of multi-single-joint manipulators under privacy protection. The joint angle positions of the multi-single-joint manipulators are as Figure 3 shown. Figure 4 is the error simulation result of the multi-single-joint manipulator. Figure 5 is the simulation result of the leader estimator of the multi-single-joint manipulator.

[0127] The above embodiments are only one of the preferred embodiments of the present invention and should not be used to limit the protection scope of the present invention. Any meaningless modifications or polishings made on the main design idea and spirit of the present invention, as long as the technical problems solved are still the same as those of the present invention, should be included in the protection scope of the present invention.

Claims

1. A multi-joint robot arm optimization control method based on output mask function, characterized in that: The steps include: S1, mathematical modeling of the multi-joint robot follower, constructing the unknown control coefficient and the state equation of the follower under disturbance according to the physical characteristics of the multi-joint robot; S2, constructing a nonlinear error transfer function with auxiliary variables based on the state equation and transfer function of the converted follower; S3, introduce the output mask function, define the error and error variables; S4, according to the first error surface ξ i1 , ideal optimal virtual control signal The optimal cost function of the first subsystem is established by the optimal theoretical method, and the adaptive dynamic programming method based on the neural network architecture is used to approximate the solution to obtain a new ideal optimal virtual control signal; S5, executing the evaluation neural network to approximate the optimal virtual control signal for the predetermined time, designing the adaptive parameters corresponding to the optimal virtual control signal for the predetermined time, the leader estimator and the execution-evaluation network estimation weight update law; S6, according to the second error surface ξ i2 , Ideal optimal virtual controller The optimal cost function of the second subsystem is established by the optimal theoretical method, and the adaptive dynamic programming method based on the neural network architecture is used to approximate the solution to obtain a new ideal optimal controller; S7, execute the evaluation neural network to approximate the predetermined time approximate optimal controller, design the adaptive parameters corresponding to the predetermined time approximate optimal controller and the execution-evaluation network estimation weight update law.

2. The multi-joint robot arm optimization control method based on output mask function according to claim 1 is characterized in that: It also includes the construction of complete Lyapunov functions: And the stability analysis of its time derivative is obtained in is a matrix The minimum eigenvalue of , I represents the number of followers, and Both represent parameter estimation errors, m=1,2.

3. The multi-joint robot arm optimization control method based on output mask function according to claim 2 is characterized in that: In step S2, the expression of the transfer function is: Where T s Indicates the delay time.

4. The multi-joint robot arm optimization control method based on output mask function according to claim 3 is characterized in that: In step S2, the expression of the nonlinear error conversion function is: in Two Cs n The function is expressed as: in s i1 is an auxiliary variable, 5. The multi-joint robot arm optimization control method based on output mask function according to claim 4 is characterized in that: In step S3, the expression of the output mask function is: where ω i ={ i ,λ i }, therefore, the follower state under the output mask can be expressed as: where λ im >0 and υ im ≠0 are all definition parameters.

6. The multi-joint robot arm optimization control method based on output mask function according to claim 5 is characterized in that: In step S3, the expression containing the error is: where a ij =1 means that the ith follower can receive information from the adjacent j followers, otherwise a ij =0; a ir =1 means the i-th follower can accept the leader signal, otherwise, a ir =0; Represents the protection signal received by the leader, represents the protection signal received from the adjacent follower j; The expression of the error variable is: where ξ il (l=1,2) represents the error surface, r i1 represents the filtering error, the virtual control law α i1 As a filter The output signal of the filter is obtained ι i Represents a time scalar.

7. The multi-joint robot arm optimization control method based on output mask function according to claim 6 is characterized in that: In step S4, the specific process of obtaining a new ideal optimal virtual control signal is as follows: According to the first error surface ξ i1 , ideal optimal virtual control signal and optimal theoretical method to design the optimal cost function of the first subsystem For the optimal cost function The Hamilton-Jacobi-Bellman equation of the first subsystem is constructed by derivation. According to Bellman's optimal principle, the Hamilton-Jacobi-Bellman equation of the first subsystem and the ideal optimal virtual control signal are derived to obtain the ideal optimal virtual control signal. The ideal optimal virtual control signal is approximated by the adaptive dynamic programming method based on the neural network architecture to obtain a new ideal optimal virtual control signal.

8. The multi-joint robot arm optimization control method based on output mask function according to claim 7 is characterized in that: In step S5, the predetermined time approximates the optimal virtual control signal α i1 for: in ρ i is a positive design parameter; i=1,2,3,……I, I represents the number of followers, represents the basis function of the neural network approximating the first unknown function, is the basis function for the neural network to approximate the first unknown cost function. represents a positive constant, for the variable ξ i1 After taking the derivative, we get σ i1 ; for The estimated value of is the ideal weight vector of the neural network, is the approximation error, express The estimated value of is the optimal weight vector of the neural network The corresponding execution optimal weight vector; Represents follower i to the leader Identification; the predetermined time function is expressed as: Where T i >0 and are the predetermined convergence time and the defined scale, respectively, and n = 2 represents the order of the system; The leader estimator, the adaptive law, and the execution-evaluation network estimate weight update law are: in and γ ic1 ,γ ia1 , For positive design parameters, γ ic1 >γ ia1 / 2, and Respectively and The estimated value of . is the optimal weight vector of the neural network The corresponding evaluation optimal weight vector.

9. The multi-joint robot arm optimization control method based on output mask function according to claim 8 is characterized in that: In step S6, the specific method of obtaining a new ideal optimal controller is: According to the second error surface ξ i2 , Ideal optimal virtual controller and the optimal cost function of the second subsystem using the optimal theory method For the optimal cost function The Hamilton-Jacobi-Bellman equation of the second subsystem is constructed by derivation. According to Bellman's optimal principle, the Hamilton-Jacobi-Bellman equation of the second subsystem and the ideal optimal virtual controller are derived to obtain the ideal optimal virtual controller. The ideal optimal virtual controller is approximated by the adaptive dynamic programming method based on the neural network architecture to obtain a new ideal optimal controller.

10. The multi-joint robot arm optimization control method based on output mask function according to claim 9 is characterized in that: In step S7, the predetermined time approximates the optimal controller u i for: in κ i2 is a positive design parameter, Respectively The estimated value of is the optimal weight vector of the neural network The corresponding execution optimal weight vector, represents the basis function of the neural network approximating the second unknown function, is the approximation error, is the basis function for the neural network to approximate the second unknown cost function, and the neural network input Indicates normal amount, g i2 The upper bound value, g i2 =1 / P, Scheduled time approximation optimal controller u i The corresponding adaptive parameters and the update law of the execution-evaluation network estimated weights are: in γ ic2 ,γ ia2 For positive design parameters, γ ic2 >γ ia2 / 2, and Respectively and The estimated value of is the optimal weight vector of the neural network The corresponding execution and evaluation optimal weight vectors, For normal amount.

Images