Hip exoskeleton adaptive control method based on angle tracking

Abstract

The invention relates to a hip exoskeleton adaptive control method based on angle tracking. The method comprises the following steps: S1, inputting a hip joint expected angle through an expected model; s2, taking the expected hip joint angle as the input of a reference model to obtain a reference hip joint angle track; s3, calculating a track error between the reference hip joint angle and the actual hip exoskeleton angle; s4, inputting the track error obtained in the S3 into a fitness function, and adaptively adjusting and controlling a gain parameter; s5, optimizing an initial gain parameter of the controller by using a differential evolution algorithm and a particle swarm optimization algorithm; s6, exoskeleton control instruction parameters are calculated in combination with the optimized gain parameters and steady-state errors; and S7, inputting the control instruction into an exoskeleton kinetic equation for angle tracking, and feeding back the control instruction in real time by an encoder to form closed-loop control. Compared with the prior art, accurate angle tracking and error suppression are realized; through multi-algorithm collaborative optimization and parameter adaptability, scene adaptability is enhanced; the stability is high.

Description

Technical Field

[0001] The present invention relates to the technical field of exoskeleton control, and in particular to a hip exoskeleton adaptive control method based on angle tracking. Background Art

[0002] With the aging of the population and the acceleration of industrialization, the load on the lower limbs during daily activities and high-intensity work is becoming increasingly prominent. The hip-assisted exoskeleton, a wearable robot, can provide assistive torque during walking, heavy lifting, and rehabilitation training, thereby reducing muscle strain and improving mobility and work efficiency.

[0003] Existing hip-assisted exoskeleton technologies primarily utilize motors or hydraulics, providing assistance through dynamic modeling and control algorithms. However, existing technologies still suffer from low power control accuracy, limited response speed, and limited adaptability of the exoskeleton system.

[0004] Patent CN201910946543.5 discloses a lower-limb exoskeleton control method, which includes collecting kinematic data of a subject's lower limbs; establishing a dynamic model of the lower-limb exoskeleton; designing a nonlinear integral sliding surface; and designing a fuzzy sliding mode controller to obtain a fuzzy sliding mode control law. The Euler-Lagrange method is used to establish the dynamic model of the lower-limb exoskeleton. To eliminate the chattering phenomenon and the Windup effect caused by the integral term, a nonlinear potential energy function is introduced on the basis of the sliding mode variable structure controller to replace the traditional integral sliding surface. To overcome interference caused by modeling errors, signal noise, and external disturbances during the lower-limb exoskeleton modeling process, the approximation properties of fuzzy systems are utilized to design a fuzzy sliding mode controller to achieve satisfactory lower-limb exoskeleton control performance. However, the fuzzy sliding mode controller relies on a large number of fuzzy rules and membership functions, resulting in a high computational load and potential real-time delays. For exoskeleton systems that require fast response, this delay can affect control effectiveness and safety.

[0005] Therefore, developing a hip-assisted exoskeleton with efficient adaptive capabilities that can adjust the assistance strategy in real time according to the wearer's gait to improve energy efficiency, optimize human interaction performance, and adapt to complex environments is a key technical requirement in this field. Summary of the Invention

[0006] The purpose of the present invention is to overcome the defects of the above-mentioned existing technologies and provide a hip exoskeleton adaptive control method based on angle tracking to achieve precise angle tracking and error suppression; multi-algorithm collaborative optimization and parameter adaptability, reduce manual dependence, enhance scene adaptability, and improve stability.

[0007] The present invention provides a hip exoskeleton adaptive control method based on angle tracking, comprising the following steps:

[0008] S1: Input the desired hip joint angle through the desired model;

[0009] S2: Taking the desired hip joint angle as the input of the reference model, the reference hip joint angle trajectory is obtained;

[0010] S3: Calculate the trajectory error between the reference hip joint angle and the actual angle of the hip exoskeleton;

[0011] S4: Input the trajectory error obtained in S3 into the fitness function and adaptively adjust the control gain parameters;

[0012] S5: Jointly optimize the initial gain parameters of the controller using differential evolution algorithm and particle swarm optimization algorithm;

[0013] S6: Calculate the exoskeleton control command parameters based on the optimized gain parameters and steady-state error;

[0014] S7: Input the control command into the exoskeleton dynamic equation for angle tracking, and the encoder outputs the current angle in real time to form a closed-loop control.

[0015] By jointly optimizing the initial gain through differential evolution and particle swarm optimization, the parameter convergence speed is significantly improved; combining adaptive gain adjustment with closed-loop tracking, the trajectory tracking accuracy and the ability to adapt to different gaits are improved.

[0016] Furthermore, in S2, the reference model is represented by the following formula:

[0017]

[0018] Where H1 is a positive constant. d (s) is the expected angular trajectory; θ(s) is the actual angular trajectory; G(s) is the frequency domain mathematical model of the hip joint.

[0019] The frequency domain model is used to accurately describe the dynamic characteristics of the exoskeleton, providing a theoretical basis for error calculation.

[0020] Furthermore, in S3, the trajectory error between the reference hip joint angle and the actual angle of the hip exoskeleton is calculated as follows:

[0021] e m =θ-θ m ;

[0022] Where θ is the actual angle trajectory of the hip exoskeleton, θ m Represents the angular trajectory of the reference model output.

[0023] Clarify the error definition to ensure the input accuracy of adaptive adjustment.

[0024] Furthermore, in S4, the controller gain is adaptively adjusted using the gradient descent method according to the error signal so that the system output is as close as possible to the reference model output. The fitness function is:

[0025]

[0026] Where β = [AB] T is the gain matrix of the controller, which is adaptively adjusted to the change of β in the negative gradient direction of J. The adjustment formula is:

[0027]

[0028] The negative sign means that the change of β makes J minimum, is the sensitivity derivative of the tracking error, σ represents the speed of adaptation, so so According to the reference model, the sensitivity derivative of the tracking error is Therefore, the equation can be redefined as the following equation:

[0029]

[0030] Since HG(s)θ d It is not possible to directly obtain the exoskeleton model in the frequency domain To sort out the available therefore:

[0031]

[0032] in Applying the adjusted control rate to calculate the control input, the above equation represents the adjustment of the controller parameter β over time. Integrating it with respect to time t yields:

[0033]

[0034] Where β1 = [A1 B1] T are the initial values of the controller parameters.

[0035] The gain is dynamically optimized by gradient descent to minimize the tracking error.

[0036] Furthermore, in S5, the steps of optimizing the initial gain of the controller using the differential evolution algorithm include:

[0037] S51: Set the search range of gain parameters A and B [A min , A max ],[B min, B max ];

[0038] S52: Initialize the population: Generate N candidate solutions, each individual consists of gain parameters A and B, and is randomly initialized;

[0039] S53: For each individual in the population, three different individuals are randomly selected to generate mutation vectors; the mutation operation adjusts individual parameters through weighted difference calculation to improve the diversity and optimization ability of the population; for each individual x i , choose three different individuals x a , x b , x c (a≠b≠c≠i). Generate mutation vector: v i =x a +F(x b -x o ), where F is the factor of variation.

[0040] S54: Crossover is performed on the mutated individuals to generate new test individuals. The crossover process determines whether to use the parameter value of the mutated individual based on the set crossover probability, ensuring that at least one parameter of the new individual is different from that of the original individual, thereby increasing the exploration capability of the search space.

[0041] S55: Calculate the fitness function values of the current individual and the test individual, and select based on the fitness values, retaining the better individuals to enter the next generation population to ensure continuous optimization of population quality;

[0042] S56: Iterative optimization: Repeat mutation, crossover, and selection until the maximum number of iterations is reached or the objective function converges; finally, the individual with the smallest error is selected from the optimized population as the final optimized gain parameter (A′, B′) = argminf(A, B).

[0043] Enhance the diversity and global convergence of parameter searches.

[0044] Furthermore, in S52, each individual is x i =(A i , B i ), randomly initialized:

[0045] A i =A min +r(A max -A min );

[0046] B i =B min +r(B max -B min );

[0047] Where r is a random number between [0, 1];

[0048] In S54, the test individual u is generated i :

[0049]

[0050] Among them C r The crossover probability is usually between 0.1 and 0.9, j r is a randomly selected dimension that ensures that at least one parameter is changed.

[0051] In S55, calculate f(u i ) and f(x i ), select better individuals to enter the next generation;

[0052]

[0053] Furthermore, in S5, the step of optimizing the initial gain of the controller by particle swarm optimization includes:

[0054] S57: Initialization: Set the appropriate number of particles, each particle represents a set of initialization model reference adaptive control gain parameters (A, B); randomly initialize the position (gain parameter value) and speed (parameter change rate) of each particle; set the global optimal position g B and the individual optimal position p B The initial value of

[0055] S58: Calculate fitness: calculate the absolute error of the integral time of each particle and minimize the absolute error;

[0056] S59: Iterative optimization: Set the appropriate number of iterations so that each iteration generates a new generation of particles, updates the position and velocity of the particles; calculates the individual optimal position p of each particle B ; Update the global optimal position g B , if p B <g B Then keep g B ; When the maximum number of iterations is reached or the error converges, the final g is selected B as output.

[0057] Combined with time-weighted error evaluation, the convergence of local optimal solutions is accelerated.

[0058] Furthermore, in S58, the integral time absolute error is used as the fitness function to evaluate the controller performance to accelerate the convergence speed. The fitness function formula is as follows:

[0059]

[0060] Where t represents the elapsed time, e o (t) is the time domain steady-state error.

[0061] Furthermore, in S59, the particle of the next iteration is the sum of the position and velocity of the particle of the previous generation, which is expressed by the following formula:

[0062] x i,j =x i-1,j +v i,j ;

[0063] Where j and i represent the number of particles and the number of iterations respectively; x i-1，j Indicates the position of the particle in the previous iteration; v i，j It represents the speed and direction of the current particle to the next iteration, which can be expressed as follows:

[0064] v i，j =w i v i-1，j +C1ζ1(p B，i-1 -x i，j )-C2ζ2(g B -x i，j );

[0065] Among them, ζ1 and ζ2 are randomly generated between 0 and 1, C1 and C2 are positive coefficients of self-cognition component and social component respectively; w i is the inertia weight, the value is readjusted in each iteration, let:

[0066] w i =w d w i-1 ;

[0067] where w d is the damping value, which is evaluated by the objective function; each particle with the lowest objective function in each iteration is selected as p B，i ; After evaluation, the lowest p B，i Defined as the global optimal g B .

[0068] Furthermore, in S6, when the steady-state error and gain parameters are introduced to obtain the exoskeleton control command parameters, the control law is τ = AI, where where A∈R n×n , B∈R n×n is the symmetric matrix of controller gains, τ is the controller output, and e is the steady-state error; it is obtained from the following formula:

[0069] e=θ d -θ;

[0070] where θ d ∈R n×1and θ∈R n×1 represents the desired angular trajectory and the actual angular trajectory;

[0071] In the frequency domain, the control law becomes the following:

[0072]

[0073] where C(s) is the controller in the frequency domain. When the input of the closed-loop control system is set to a unit step response, the objective function is determined; therefore, the steady-state error of the optimization problem is expressed as follows:

[0074]

[0075] Furthermore, the input trajectory error in S6 is used to adaptively adjust the control gain, and the stability of the system is analyzed using Lyapunov stability theory. In the steady-state error equation, we assume that θ d is a constant, so:

[0076]

[0077] Considering the general dynamics in the dynamics equation, we can get:

[0078]

[0079] The equation I = e + Be can be transformed into the following equation:

[0080]

[0081] Where D is obtained from the following formula:

[0082]

[0083] Where β is a positive constant matrix, K 4×4 is the identity matrix; Equation (29) is described in the state space as follows:

[0084]

[0085] in

[0086]

[0087] If there exists a positive definite matrix P∈R n×n , which can satisfy E T P + PE = -Q; where Q∈R n×n is a positive definite symmetric matrix; the candidate positive definite Lyapunov function is Therefore, integrating these two equations yields:

[0088]

[0089] therefore, The determined controller output is bounded by |τ|≤σ, where σ is a positive constant and H and P are assumed to be positive definite matrices. I depends on the steady-state error e and its derivative From t→∞, e→0 and therefore therefore:

[0090]

[0091] Where γ is a positive constant. The system is bounded within a certain range and is proven to be asymptotically stable.

[0092] Furthermore, in S6, the dynamic equations are given by the dynamic equations described in S2, and the encoder is a brushless DC motor encoder. It provides a high-precision angle feedback signal to ensure the reliability of closed-loop control.

[0093] Compared with the prior art, the present invention has the following advantages:

[0094] (1) Accurate angle tracking and error suppression: The angle tracking strategy is used to achieve precise tracking of the ideal hip joint angle, reduce trajectory errors, and improve torque output stability. The closed-loop tracking mechanism inputs the desired angle into the reference model and forms a closed-loop control with the actual angle (encoder feedback). The trajectory error is calculated in real time and dynamically corrected through the control law. Adaptive gain adjustment uses the fitness function and gradient descent method to dynamically adjust the gain matrix, combines the steady-state error to generate control instructions, and realizes error proportional-differential compound control. Compared with the traditional fixed gain algorithm, adaptive adjustment eliminates static error accumulation, improves trajectory tracking accuracy, and ensures continuous and stable torque output during the gait cycle.

[0095] (2) Multi-algorithm collaborative optimization and parameter adaptability: Initial gains are jointly optimized through differential evolution and particle swarm optimization to reduce dependence on manual parameter adjustment; adaptive parameter adjustment enhances system intelligence and scenario adaptability. Joint optimization algorithm: The differential evolution algorithm searches the parameter range globally through mutation vectors, and the particle swarm optimization integrates the absolute error of time. The two jointly optimize the initial gains A and B to avoid local optimality. Automatic parameter adjustment: Gains are updated online based on the gradient descent method, combined with the optimized initial parameters to adapt to different gaits and load changes. The time consumption of initial parameter optimization is reduced, and the adaptive mechanism enables the exoskeleton to have small control error fluctuations in complex scenarios (such as uphill and downhill) without the need for manual intervention.

[0096] (3) Enhanced system stability and robustness: Lyapunov stability verification and model matching are used to ensure system robustness under different individual gaits. Lyapunov functions are used to constrain the risk of control command divergence and ensure global asymptotic stability. The exoskeleton dynamics model accurately describes inertia, Coriolis force, and gravity terms, and is combined with a reference model to achieve frequency domain characteristic matching. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0098] Figure 2 is a block diagram of the control system of the present invention;

[0099] Figure 3 Schematic diagram of particle swarm optimization algorithm;

[0100] Figure 4 Schematic diagram of optimization for differential evolution algorithm;

[0101] Figure 5 It is the control object of the present invention. DETAILED DESCRIPTION

[0102] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Component models, material names, connection structures, control methods, algorithms, and other features not explicitly described in this technical solution are considered common technical features disclosed in the prior art.

[0103] Example 1

[0104] This embodiment provides a hip exoskeleton adaptive control method based on angle tracking. Figure 1-5 As shown, the following steps are included:

[0105] S1: Input the desired hip joint angle through the desired model;

[0106] S2: If Figure 2 As shown in Figure 2, the desired hip joint angle is used as the input of the reference model to obtain the reference hip joint angle trajectory; the reference model is expressed as follows:

[0107]

[0108] Where H1 is a positive constant. d (s) is the desired angular trajectory; θ(s) is the actual angular trajectory; G(s) is the frequency domain mathematical model of the hip joint. G(s) is given by the following formula:

[0109]

[0110] Where G(s) represents the frequency domain mathematical model of the hip joint. b and a i, i = 1, 2, 3, 4 are unknown parameters. Since the hip exoskeleton is symmetrical, we can simplify the control strategy to a single leg. The corresponding unknown parameters can be obtained through system identification in MATLAB. u is obtained by modeling the dynamics of the hip exoskeleton and can be expressed as follows:

[0111]

[0112] Where M(θ) represents the inertia scalar, θ represents the hip joint angle, denotes the centripetal and Coriolis vectors, G(θ) denotes gravity, and τ is the control output.

[0113] The frequency domain model is used to accurately describe the dynamic characteristics of the exoskeleton, providing a theoretical basis for error calculation.

[0114] S3: If Figure 2 As shown in Figure 2, the trajectory error between the reference hip joint angle and the actual angle of the hip exoskeleton is calculated; the trajectory error is calculated by the following formula:

[0115] e m =θ-θ m ;

[0116] Where θ is the actual angle trajectory of the hip exoskeleton, θ m Represents the angular trajectory of the reference model output.

[0117] Clarify the error definition to ensure the input accuracy of adaptive adjustment.

[0118] S4: Input the trajectory error obtained in S3 into the fitness function and adaptively adjust the control gain parameters. According to the error signal, the controller gain is adaptively adjusted using the gradient descent method to make the system output as close as possible to the reference model output. The fitness function is:

[0119]

[0120] Where β = [AB] T is the gain matrix of the controller, which is adaptively adjusted to the change of β in the negative gradient direction of J. The adjustment formula is:

[0121]

[0122] The negative sign means that the change of β makes J minimum, is the sensitivity derivative of the tracking error, σ represents the speed of adaptation, so so According to the reference model, the sensitivity derivative of the tracking error is Therefore, the equation can be redefined as the following equation:

[0123]

[0124] Since HG(s)θ d It is not possible to directly obtain the exoskeleton model in the frequency domain To sort out the available therefore:

[0125]

[0126] in Applying the adjusted control rate to calculate the control input, the above equation represents the adjustment of the controller parameter β over time. Integrating it with respect to time t yields:

[0127]

[0128] Where β1 = [A1 B1] T are the initial values of the controller parameters.

[0129] The gain is dynamically optimized by gradient descent to minimize the tracking error.

[0130] S5: Jointly optimize the initial gain parameters of the controller using differential evolution algorithm and particle swarm optimization algorithm;

[0131] like Figure 3 As shown in Figure 2, the steps for optimizing the initial controller gain using the differential evolution algorithm include:

[0132] S51: Set the search range of gain parameters A and B [A min , A max ],[B min , B max ];

[0133] S52: Initialize the population: Generate N candidate solutions, each individual consists of gain parameters A and B, and is randomly initialized; each individual is x i =(A i , B i ), randomly initialized:

[0134] A i =A min +r(A max -A min );

[0135] B i =B min +r(B max -B min );

[0136] Where r is a random number between [0, 1];

[0137] S53: For each individual in the population, three different individuals are randomly selected to generate mutation vectors; the mutation operation adjusts individual parameters through weighted difference calculation to improve the diversity and optimization ability of the population; for each individual x i , choose three different individuals x a , x b , x o (a≠b≠c≠i). Generate mutation vector: v i =x a +P(x b -x c ), where F is the factor of variation.

[0138] S54: Cross the mutated individuals to generate new test individuals u i :

[0139]

[0140] Among them C r The crossover probability is usually between 0.1 and 0.9, j r is a randomly selected dimension that ensures that at least one parameter is changed.

[0141] The crossover process decides whether to use the parameter value of the mutated individual based on the set crossover probability, ensuring that at least one parameter of the new individual is different from the original individual, thereby increasing the exploration ability of the search space;

[0142] S55: Calculate the fitness function values of the current individual and the test individual, and select according to the fitness value, retaining the better individuals to enter the next generation population to ensure the continuous optimization of the population quality; In S55, calculate f(u i ) and f(x i ), select better individuals to enter the next generation;

[0143]

[0144] S56: Iterative optimization: Repeat mutation, crossover, and selection until the maximum number of iterations is reached or the objective function converges; finally, the individual with the smallest error is selected from the optimized population as the final optimized gain parameter (A′, B′) = argminf(A, B).

[0145] Enhance the diversity and global convergence of parameter searches.

[0146] like Figure 4 As shown in Figure 2, the steps for particle swarm optimization to optimize the initial gain of the controller include:

[0147] S57: Initialization: Set the appropriate number of particles, each particle represents a set of initialization model reference adaptive control gain parameters (A, B); randomly initialize the position (gain parameter value) and speed (parameter change rate) of each particle; set the global optimal position g B and the individual optimal position p B The initial value of

[0148] S58: Calculate fitness: Calculate the absolute error of the integral time of each particle and minimize the absolute error; use the absolute error of the integral time as the fitness function to evaluate the controller performance to speed up the convergence. The fitness function formula is as follows:

[0149]

[0150] Where t represents the elapsed time, e o (t) is the time domain steady-state error.

[0151] S59: Iterative optimization: Set the appropriate number of iterations so that each iteration generates a new generation of particles, updates the position and velocity of the particles; calculates the individual optimal position p of each particle B ; Update the global optimal position g B , if p B <g B Then keep g B ; When the maximum number of iterations is reached or the error converges, the final g is selected B as output.

[0152] The next generation of particles is the sum of the position and velocity of the previous generation of particles, which is expressed as follows:

[0153] x i,j =x i-1,j +v i,j ;

[0154] Where j and i represent the number of particles and the number of iterations respectively; x i-1，j Indicates the position of the particle in the previous iteration; v i，j It represents the speed and direction of the current particle to the next iteration, which can be expressed as follows:

[0155] v i，j =w i v i-1,j +C1ζ1(p B，i-1 -x i，j )-C2ζ2(g B -x i，j );

[0156] Among them, ζ1 and ζ2 are randomly generated between 0 and 1, C1 and C2 are positive coefficients of self-cognition component and social component respectively; w iis the inertia weight, the value is readjusted in each iteration, let:

[0157] w i =w d w i-1 ;

[0158] where w d is the damping value, which is evaluated by the objective function; each particle with the lowest objective function in each iteration is selected as p B，i ; After evaluation, the lowest p B，i Defined as the global optimal g D .

[0159] Combined with time-weighted error evaluation, the convergence of local optimal solutions is accelerated.

[0160] S6: Calculate the exoskeleton control command parameters by combining the optimized gain parameters and steady-state error. In the process of obtaining the exoskeleton control command parameters by substituting the steady-state error and gain parameters, the control law is τ = Ai, where where A∈R n×n , B∈R n×n is the symmetric matrix of controller gains, τ is the controller output, and e is the steady-state error; it is obtained from the following formula:

[0161] e=θ d -θ;

[0162] where Q d ∈R n×1 and θ∈R n×1 represents the desired angular trajectory and the actual angular trajectory;

[0163] In the frequency domain, the control law becomes the following:

[0164]

[0165] where C(s) is the controller in the frequency domain. When the input of the closed-loop control system is set to a unit step response, the objective function is determined; therefore, the steady-state error of the optimization problem is expressed as follows:

[0166]

[0167] The input trajectory error is used to adaptively adjust the control gain, and the stability of the system is analyzed using Lyapunov stability theory. In the steady-state error equation, we assume that θ d is a constant, so:

[0168]

[0169] Considering the general dynamics in the dynamics equation, we can get:

[0170]

[0171] equation It can be transformed into the following equation:

[0172]

[0173] Where D is obtained from the following formula:

[0174]

[0175] Where β is a positive constant matrix, K 4×4 is the identity matrix; Equation (29) is described in the state space as follows:

[0176]

[0177]

[0178] in

[0179] If there exists a positive definite matrix P∈R n×n , which can satisfy E T P + PE = -Q; where Q∈R n×n is a positive definite symmetric matrix; the candidate positive definite Lyapunov function is Therefore, integrating these two equations yields:

[0180]

[0181] therefore, The determined controller output is bounded by |τ|≤σ, where σ is a positive constant and H and P are assumed to be positive definite matrices. I depends on the steady-state error e and its derivative By t>∞,e>0 and therefore therefore:

[0182]

[0183] Where γ is a positive constant. The system is bounded within a certain range and is proven to be asymptotically stable.

[0184] The dynamic equations are given by the dynamic equations described in S2, and the encoder is a brushless DC motor encoder. It provides high-precision angle feedback signals to ensure the reliability of closed-loop control.

[0185] S7: Input the control command into the exoskeleton dynamic equation for angle tracking, and the encoder outputs the current angle in real time to form a closed-loop control.

[0186] By jointly optimizing the initial gain through differential evolution and particle swarm optimization, the parameter convergence speed is significantly improved; combining adaptive gain adjustment with closed-loop tracking, the trajectory tracking accuracy and the ability to adapt to different gaits are improved.

[0187] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. Such program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0188] In the context of the present invention, machine-readable medium can be a tangible medium that can contain or store a program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0189] Components not described in detail in this embodiment are all existing components that can be purchased through public channels.

[0190] The above description of the embodiments is intended to facilitate understanding and use of the invention by those skilled in the art. It will be apparent that those skilled in the art can readily make various modifications to these embodiments and apply the general principles described herein to other embodiments without requiring inventive effort. Therefore, the present invention is not limited to the above-described embodiments. Improvements and modifications made by those skilled in the art based on the disclosure of the present invention, without departing from the scope of the present invention, should be within the scope of protection of the present invention.

Claims

1. A hip exoskeleton adaptive control method based on angle tracking, characterized in that: The following steps are involved: S1: Input the desired hip joint angle through the desired model; S2: Taking the desired hip joint angle as the input of the reference model, the reference hip joint angle trajectory is obtained; S3: Calculate the trajectory error between the reference hip joint angle and the actual angle of the hip exoskeleton; S4: Input the trajectory error obtained in S3 into the fitness function and adaptively adjust the control gain parameters; S5: Jointly optimize the initial gain parameters of the controller using differential evolution algorithm and particle swarm optimization algorithm; S6: Calculate the exoskeleton control command parameters based on the optimized gain parameters and steady-state error; S7: Input the control command into the exoskeleton dynamic equation for angle tracking, and the encoder outputs the current angle in real time to form a closed-loop control.

2. The hip exoskeleton adaptive control method based on angle tracking according to claim 1, characterized in that: In S2, the reference model is represented by the following formula: ； in is a positive constant; is the desired angular trajectory; is the actual angular trajectory; It is the frequency domain mathematical model of the hip joint.

3. The hip exoskeleton adaptive control method based on angle tracking according to claim 1, characterized in that: In S3, the trajectory error between the reference hip joint angle and the actual angle of the hip exoskeleton is calculated as follows: ； in is the actual angle trajectory of the hip exoskeleton, Represents the angular trajectory of the reference model output.

4. The method for adaptive control of a hip exoskeleton based on angle tracking according to claim 1, characterized in that: In S4, the fitness function is: ； in, is the controller gain matrix, which is adaptively adjusted to exist The change in the negative gradient direction, the adjustment formula is: 。 5. The method for adaptive control of a hip exoskeleton based on angle tracking according to claim 1, characterized in that: In S5, the steps of optimizing the initial controller gain using the differential evolution algorithm include: S51: Set gain parameters and Search scope ， ; S52: Initialize the population: Generate N candidate solutions, each individual is determined by the gain parameter and Composition, and random initialization; S53: For each individual in the population, three different individuals are randomly selected to generate a mutation vector; S54: Crossover is performed on the mutated individuals to generate new test individuals. The crossover process determines whether to use the parameter value of the mutated individual based on the set crossover probability, ensuring that at least one parameter of the new individual is different from that of the original individual, thereby increasing the exploration capability of the search space. S55: Calculate the fitness function values of the current individual and the test individual, and select based on the fitness values, retaining the better individuals to enter the next generation population to ensure continuous optimization of population quality; S56: Iterative optimization: Repeat mutation, crossover, and selection until the maximum number of iterations is reached or the objective function converges; finally, the individual with the smallest error is selected from the optimized population as the final optimized gain parameter.

6. The method for adaptive control of a hip exoskeleton based on angle tracking according to claim 5, characterized in that: In S52, each individual is =( ), randomly initialized: ； ； in yes A random number between In S54, a test individual is generated : ； in The crossover probability is usually between 0.1 and 0.

9. is a randomly selected dimension, ensuring that at least one parameter is changed; In S55, calculate and , select better individuals to enter the next generation; 。 7. The method for adaptive control of a hip exoskeleton based on angle tracking according to claim 1, characterized in that: In S5, the steps of particle swarm optimization for optimizing the initial gain of the controller include: S57: Initialization: Set the appropriate number of particles, each particle represents a set of initialization model reference adaptive control gain parameters ( ); Randomly initialize the position and velocity of each particle; Set the global optimal position and individual optimal position The initial value of S58: Calculate fitness: calculate the absolute error of the integral time of each particle and minimize the absolute error; S59: Iterative optimization: Set the number of iterations so that each iteration generates a new generation of particles, updates the position and velocity of the particles; calculates the individual optimal position of each particle ; Update the global optimal position ,if Keep ; When the maximum number of iterations is reached or the error converges, the final as output.

8. The method for adaptive control of a hip exoskeleton based on angle tracking according to claim 7, characterized in that: In S58, the integral time absolute error is used as the fitness function to evaluate the controller performance to speed up the convergence. The fitness function formula is as follows: in Indicates the time that has passed. is the time domain steady-state error.

9. The method for adaptive control of a hip exoskeleton based on angle tracking according to claim 7, characterized in that: In S59, the next generation of iterative particles is the sum of the position and velocity of the previous generation of particles, which is expressed by the following formula: ； Where j and i represent the number of particles and the number of iterations respectively; Indicates the position of the particle in the previous iteration; It represents the speed and direction of the current particle to the next iteration, which can be expressed as follows: ； in and Randomly generated between 0 and 1, and are the positive coefficients of the self-perception component and the social component, respectively; is the inertia weight, the value is readjusted in each iteration, let: ； in is the damping value, which is evaluated using the objective function; Each particle with the lowest objective function in each iteration is selected as ; After evaluation, the lowest Defined as the global optimal .

10. The method for adaptive control of a hip exoskeleton based on angle tracking according to claim 1, characterized in that: In S6, when the steady-state error and gain parameters are introduced to obtain the exoskeleton control command parameters, the control law is: ,in ;in , is the symmetric matrix of controller gains, is the controller output, is the steady-state error; The steady-state error of the optimization problem is expressed as follows: 。

Images