CN115542723A - Parameter setting method for ankle exoskeleton PD controller for man-machine in-loop debugging - Google Patents

Abstract

The invention discloses a man-machine-in-the-loop debugging ankle exoskeleton PD controller parameter setting method, which can be used to adjust the PD controller parameters of the ankle exoskeleton under different conditions of wearers and walking conditions so that the control effect can be improved. To achieve the optimum, the setting process is adjusted online through human-machine-in-the-loop, without the intervention of specific test environments or equipment. The process of the method is as follows: the wearer collects the combination of the proportional coefficient K _p and the differential coefficient K _d of the PD controller during the actual operation of the ankle exoskeleton in the walking condition, and the corresponding auxiliary force control error, and then imports the Gaussian regression algorithm Surface fitting is carried out in , and the K _p and K _d parameters of the lowest point of the surface (that is, the predicted control error is the smallest) are obtained, and then imported into the exoskeleton PD controller to update the parameters, and repeat the parameter iteration process of man-machine in the loop, When the final K _p and K _d parameters converge, it means that the PD controller parameters are adjusted to the optimum for the current wearer and walking conditions.

Description

A human-machine-in-the-loop debugging method for parameter tuning of ankle exoskeleton PD controller

technical field

The invention relates to the technical field of exoskeleton control, in particular to a method for parameter setting of an ankle exoskeleton PD controller for man-machine-in-the-loop debugging.

Background technique

The exoskeleton system is a typical human-machine system. Its working process is the mutual cooperation between the human body and the robot. It takes various subjective intentions of the human body as the planning core, and combines the perception of sensor technology and the strength advantages of robots to implement and expand. The ankle joint exoskeleton applies an auxiliary torque to the ankle joint through the exoskeleton drive system, which partially replaces the force of the human ankle joint itself during walking, thereby reducing the metabolic consumption of the human body. Because the walking motion of the ankle joint has specific biomechanical properties, the auxiliary torque applied by the exoskeleton must have a suitable style to fit it, so as to achieve the effect of replacing the force of the ankle joint. Therefore, the ankle joint exoskeleton must have appropriate control The device controls the driving force of the exoskeleton according to the appropriate auxiliary torque pattern to achieve the target auxiliary torque required for the current gait cycle phase.

At present, the control of ankle exoskeleton usually adopts PD controller based on force feedback, and its properties are determined by the proportional coefficient K _p and differential coefficient K _d . However, different wearers and different walking conditions will affect the relationship between the exoskeleton and the human body. system, the K _p and K _d parameters of the PD controller are often different when the optimal control effect is achieved. For this reason, it is necessary to adjust the parameters of the exoskeleton PD controller according to different working conditions to optimize the control effect, and the exoskeleton has special features, unlike ordinary equipment, which can be isolated on the test bench for controller parameters setting, the system controlled by the PD controller must be characterized in combination with the target wearer and in the target walking condition. At present, it is urgent to propose a human-machine-in-the-loop exoskeleton PD controller parameter tuning method.

Contents of the invention

The object of the present invention is to solve the above-mentioned defects in the prior art, and provide a method for parameter setting of an ankle joint exoskeleton PD controller for man-machine-in-the-loop debugging. This method collects the parameter combination of the proportional coefficient K _p and the differential coefficient K _d of the PD controller during the actual operation of the ankle exoskeleton in the walking condition, and the corresponding auxiliary force control error, and imports it into the Gaussian regression algorithm. Surface fitting, get the K _p and K _d parameters of the lowest point of the surface (that is, the predicted control error is the smallest), and then import them into the exoskeleton PD controller to update the parameters, repeat the parameter iteration process of man-machine in the loop, and finally K When the _p and K _d parameters converge, it means that the PD controller parameters are adjusted to the optimum for the current wearer and walking conditions.

The purpose of the present invention can be achieved by taking the following technical solutions:

A method for parameter tuning of an ankle joint exoskeleton PD controller for man-machine in-loop debugging, the parameter tuning method comprising the steps of:

Step S1, select the initial PD controller parameters, and establish the initial sample set of exoskeleton PD controller parameter tuning by recording the assist force control error of the exoskeleton under the current PD controller parameters. This step provides the Gaussian regression process of step S2. The initial sample set is used for training;

Step S2, use the existing sample set including the initial sample set as training data, import it into the Gaussian regression algorithm for training, and repeatedly solve the optimal sample point until the convergence condition is satisfied, wherein the optimal sample point is to make each gait cycle PD The point at which the controller error is the smallest meets the convergence conditions, and the optimal proportional coefficient K _p and differential coefficient K _d are determined as the parameter tuning results of the ankle exoskeleton PD controller.

Further, the process of step S1 is as follows:

S11. It is stipulated that the initial definition domains of the proportional coefficient K _p and the differential coefficient K _d of the PD controller are [0,p ₀ ] and [0,d ₀ ] respectively, and in [0,p ₀ ] and [0,d ₀ ] Sampling at the step size of a ₀ and b ₀ respectively, as the initial sample set, to obtain the initial sample point matrix {X=[K _p ,K _d ],K _p ∈{0,a ₀ ,2a ₀ ,…,p ₀ },K _d ∈{0,b ₀ ,2b ₀ ,…,d ₀ }}, the initial sample point matrix is used as the independent variable of the training set of the Gaussian regression algorithm in step S2, and is the factor for obtaining the training set variable, perform mechanical sampling in the initial sample point lattice, and determine N uniformly distributed initial sample points X _n , n=1,2,...,N;

S12. Take an unselected initial sample point from the N initial sample points in the above-mentioned mechanical sampling, and set it as the parameter of the PD controller of the current ankle joint exoskeleton, and stipulate that the wearer can walk in the current working condition to complete the complete M gait cycles of , and record the control error e _m during the process, m=1,2,...,M;

S13. Perform uniform spline interpolation on the recorded control error in each gait cycle to obtain M control error sequences {e _m,i ,m=1,2,...,M|i with 100 elements =1,2,...,100}, where e _m,i represent the i-th element of the m-th gait cycle of the control error;

S14. In order to accurately evaluate the exoskeleton control error using different PD controller parameters, the evaluation index E _m is calculated for the control error sequence of each gait cycle, and E _m represents the evaluation index of the m-th gait cycle, and the calculation method is as follows:

Among them, F _i is the corresponding target ankle auxiliary moment when the control error occurs, and M evaluation indices E _m are calculated, m=1, 2,...,M, and then the average evaluation index is obtained by averaging the M evaluation indices

其中

表示第n个初始样本点的平均评估指数

平均评估指数在步骤S2中作为高斯回归算法的训练集的因变量。S15. Repeat steps S12 to S14 until N initial sample points are selected, and finally obtain the average evaluation index corresponding to the initial sample point X _n

in

Indicates the average evaluation index of the nth initial sample point

The average evaluation index is used as the dependent variable of the training set of the Gaussian regression algorithm in step S2.

Further, the process of step S2 is as follows:

作为因变量，并基于最大似然法，使用牛顿法或共轭梯度法等非线性数值优化算法来寻优高斯回归算法的超参数，使之最匹配现有样本集的真值分布和噪声分布，然后回归得到拟合曲面，拟合曲面上每一点包含自变量和因变量信息，所述自变量为PD控制器的参数，所述因变量为该自变量对应的预测平均评估指数；S21. Import all existing sample points including the initial sample points into the Gaussian regression algorithm for training, wherein the parameter X _n = [K _p , K _d ] of the current PD controller is used as an independent variable, and the average evaluation index

As a dependent variable, and based on the maximum likelihood method, use nonlinear numerical optimization algorithms such as Newton method or conjugate gradient method to optimize the hyperparameters of the Gaussian regression algorithm so that it best matches the true value distribution and noise distribution of the existing sample set , and then regress to obtain a fitting surface, each point on the fitting surface contains independent variable and dependent variable information, the independent variable is a parameter of the PD controller, and the dependent variable is the predicted average evaluation index corresponding to the independent variable;

S22. Retrieve the predicted dependent variable of each point on the lattice one by one on the fitted surface obtained by the trained Gaussian regression algorithm, and obtain the current optimal sample point where the minimum value of the predicted dependent variable is x _opt = [K _popt , K _dopt ], where K _popt and K _dopt are the current optimal parameter values of the proportional coefficient K _p and differential coefficient K _d of the PD controller respectively, and the current optimal sample point is the local optimal sample point, not the global optimal sample point , it is also necessary to perform convergence discrimination through steps S23 to S25;

S23. Set the current optimal parameter values K _popt and K _dopt as the new proportional coefficient K _p and differential coefficient K _d as the parameters of the current ankle exoskeleton PD controller, and stipulate that the wearer can walk in the current working condition Movement, complete M complete gait cycles, and record the control error em during the process, _m =1,2,...,M;

为当前最优样本点x_opt的平均评估指数；S24. Repeat steps S13 to S14 to obtain the average evaluation index

is the average evaluation index of the current optimal sample point x _opt ;

最小者为最优，即平均评估指数

最小的点为全局最优样本点，所述全局最优样本点的比例系数K_p和微分系数K_d取值作为踝关节外骨骼PD控制器的参数整定结果。S25. Record the current optimal sample point x _opt = [k _popt , k _dot ] as a new sample point, and form an existing sample set together with the initial sample point. If any sample point in the existing sample set is the same as If the Euclidean distance of the new sample point is less than a certain threshold, it meets the convergence judgment of the Gaussian regression process, and the average evaluation index between the point and the new sample point is selected

The smallest is the best, that is, the average evaluation index

The smallest point is the global optimal sample point, and the values of the proportional coefficient K _p and the differential coefficient K _d of the global optimal sample point are used as the parameter tuning results of the ankle exoskeleton PD controller.

Further, the step S25 also includes:

最小者为最优，平均评估指数

最小者的比例系数K_p和微分系数K_d取值作为踝关节外骨骼PD控制器的参数整定结果；Step S25.1. If the new sample point is not on the boundary of the definition domain, and the Euclidean distance to any point in the sample set is less than a certain threshold, it means that convergence has been achieved, and the arbitrary point and the new sample point are selected. average evaluation index

The smallest is the best, the average evaluation index

The value of the smallest proportional coefficient K _p and differential coefficient K _d is used as the parameter tuning result of the ankle exoskeleton PD controller;

Step S25.2. If the new sample point is not on the boundary of the defined domain, and the Euclidean distance from any point in the sample set is not less than a certain threshold, then start from step S21 and then execute the process steps in order;

Step S25.3. If the new sample point is on the domain boundary of the proportional coefficient K _p and the differential coefficient K _d , in order to find the global optimal sample point more accurately, extend the domain of definition to the outside of the boundary by a certain distance Δl to form a new After the domains [0,p] and [0,d] are defined, the process steps are executed sequentially from step S21.

Compared with the prior art, the present invention has the following advantages and effects:

(1) The present invention fully considers the problem that the optimal PD controller parameters of the exoskeleton are different under different working conditions, effectively reduces the auxiliary force control error of the actual operation of the exoskeleton under different working conditions, and maximizes the auxiliary power of the exoskeleton Effect.

(2) The present invention fuses and continuously updates the actual human motion information in the control process of the exoskeleton. The whole debugging process is an online adjustment of man-machine in the loop, and repeatedly utilizes the model-free prediction ability of the Gaussian process, from one input to the next Starting from the combination with the output, it gradually approaches the optimal value of the target.

(3) The present invention can be used to adjust the parameters of the PD controller of the ankle joint exoskeleton under different conditions of wearers and walking conditions to optimize the control effect, and the process only depends on the exoskeleton and its host computer (PC terminal, mobile terminal, MCU, etc.) without the intervention of a specific test environment or equipment.

Description of drawings

The accompanying drawings described here are used to provide a further understanding of the present invention and constitute a part of the application. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations to the present invention. In the attached picture:

Fig. 1 is a schematic diagram of a parameter setting method of an ankle exoskeleton PD controller in the present invention.

detailed description

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

Example 1

In this embodiment, the scene is that a healthy person walks freely wearing an ankle exoskeleton at an average pace of 1.25m/s. The following describes in detail how to use the present invention to adjust the parameters of the PD controller for the ankle exoskeleton during the walking process. process.

As shown in accompanying drawing 1, the man-machine-in-the-loop ankle exoskeleton PD controller parameter tuning method disclosed in this embodiment includes the following steps:

Step S1. Select the initial PD controller parameters, and establish an initial sample set for parameter tuning of the exoskeleton PD controller by recording the assist force control error of the exoskeleton under the current PD controller parameters. The process is as follows:

Step S11, stipulate that the initial definition domains of the proportional coefficient K _p and the differential coefficient K _d of the PD controller are [0,6] and [0,12] respectively, and the definition domains of K _p and K _d are respectively set to 0.1 and 0.2 Step length is used for mechanical sampling to obtain the lattice to be searched {X＝[K _p ,K _d ], K _p ∈{0,0.1,0.2,…,6},K _d ∈{0,0.2,0.4,…, 12}} Do mechanical sampling in it, determine 9 uniformly distributed initial sample points X ₁ =[1,3], X ₂ =[1,6], X ₃ =[1,9], X ₄ =[3 ,3], X ₅ =[3,6], X ₆ =[3,9], X ₇ =[5,3], X ₈ =[5,6], X ₉ =[5,9];

Step S12: Take an unselected initial sample point from the 9 initial sample points in the above-mentioned mechanical sampling, set it as the parameter of the PD controller of the current ankle joint exoskeleton, stipulate that the wearer should perform walking exercise in the current working condition, and complete Complete 12 gait cycles, and record the control error e _m during the process, m=1,2,...,12;

Step S13: Perform uniform spline interpolation on the recorded control errors in each gait cycle to obtain 12 control error sequences with 100 elements {e _m,i ,m=1,2,...,12| i=1,2,...,100};

Step S14, calculate an evaluation index E _m for the control error sequence of each gait cycle, E _m represents the evaluation index of the mth gait cycle, and the calculation method is as follows

Among them, e _{m, i} represent the i-th element of the m-th gait cycle of the control error, F _i is the corresponding target ankle auxiliary torque when the control error occurs, and 12 evaluation indices E _m ,m=1,2, ...,12, average the 12 evaluation indexes to obtain the average evaluation index

其中

表示第n个初始样本点的平均评估指数

Step S15, repeat step S12 to step S14 until all 9 initial sample points are selected, and finally obtain the average evaluation index corresponding to the initial sample point X _n

in

Indicates the average evaluation index of the nth initial sample point

In this embodiment, the initial sample set of exoskeleton PD controller parameter tuning established in step S1 is shown in Table 1.

Table 1. Parameter tuning initial sample set table of embodiment 1

initial sample order K_p K_d E. 1 1 3 44.714 2 1 6 38.507 3 1 9 34.888 4 3 3 20.247 5 3 6 21.047 6 3 9 21.906 7 5 3 16.286 8 5 6 18.293 9 5 9 20.204

Step S2, use the existing sample set including the initial sample set as training data, import it into the Gaussian regression algorithm for training, repeatedly solve the optimal sample point until the convergence condition is met, and determine the optimal proportional coefficient K _p and differential coefficient K _d The value is used as the parameter tuning result of the ankle exoskeleton PD controller, and the process is as follows:

作为因变量，并基于最大似然法，使用牛顿法或共轭梯度法等非线性数值优化算法来寻优高斯回归算法的超参数，使之最匹配现有样本集的真值分布和噪声分布，然后回归得到自变量与因变量的拟合曲面；Step S21. Import all existing sample points including the initial sample points into the Gaussian regression algorithm for training, wherein the current PD controller parameter X _n = [K _p , K _d ] is used as an independent variable, and the average evaluation index

As a dependent variable, and based on the maximum likelihood method, use nonlinear numerical optimization algorithms such as Newton method or conjugate gradient method to optimize the hyperparameters of the Gaussian regression algorithm so that it best matches the true value distribution and noise distribution of the existing sample set , and then regression to obtain the fitting surface of the independent variable and the dependent variable;

Step S22. Retrieve the predicted dependent variable of each point on the lattice one by one on the fitted surface obtained by the trained Gaussian regression algorithm, and obtain the current optimal sample point x opt where the minimum value of the predicted dependent variable is located x _opt = [K _popt , K _dopt ], where K _dopt and K _dopt are the current optimal parameter values of the proportional coefficient K _p and the differential coefficient K _d of the PD controller respectively;

Step S23, the current optimal parameter values K _popt and K _dopt are used as the new proportional coefficient K _p and differential coefficient K _d as the parameters of the current ankle exoskeleton PD controller, and the wearer is required to perform in the current working condition Walking movement, complete M complete gait cycles, and record the control error em during the process, _m =1,2,...,M;

为当前最优样本点x_opt的平均评估指数；Step S24, repeating steps S13 to S14, the obtained average evaluation index

is the average evaluation index of the current optimal sample point x _opt ;

最小者为最优，平均评估指数

最小者的比例系数K_p和微分系数K_d取值作为踝关节外骨骼PD控制器的参数整定结果。S25. Record the current optimal sample point x _opt = [k _popt , k _dot ] as a new sample point, and form an existing sample set together with the initial sample point. If any sample point in the existing sample set is the same as If the Euclidean distance of the new sample point is less than a certain threshold, then select the average evaluation index between the point and the new sample point

The smallest is the best, the average evaluation index

The value of the smallest proportional coefficient K _p and differential coefficient K _d is used as the parameter tuning result of the ankle exoskeleton PD controller.

In this embodiment, step S25 also includes:

最小者为最优，平均评估指数

最小者的比例系数K_p和微分系数K_d取值作为踝关节外骨骼PD控制器的参数整定结果；Step S25.1. If the new sample point is not on the boundary of the domain of definition, and the Euclidean distance to any point in the sample set is less than 1, it means that convergence has been achieved. Select the point between the arbitrary point and the new sample point Inter-average evaluation index

The smallest is the best, the average evaluation index

The value of the smallest proportional coefficient K _p and differential coefficient K _d is used as the parameter tuning result of the ankle exoskeleton PD controller;

Step S25.2. If the new sample point is not on the boundary of the definition domain, and the Euclidean distance from any point in the sample set is not less than 1, then start from step S21 and then execute the process steps in order;

Step S25.3. If the new sample point is on the domain boundary of the proportional coefficient K _p and the differential coefficient K _d , expand the domain to a certain distance outside the boundary, if it is on the boundary of K _p , extend it by 1, if it is on the boundary of K p On the K _d boundary, 2 is extended to form new domains [0, p] and [0, d], and then the process steps are executed sequentially from step S21.

In this embodiment, the process sample set obtained in the step S2 process is shown in Table 2,

Table 2. The parameter tuning process sample set of embodiment 1

Process Sample Order K_p K_d E. 1 4.9 0 23.750 2 4.9 12 19.329 3 <u>4.5</u> <u>3.8</u> <u>14.281</u> 4 4.4 3.6 17.706

最后选取过程样本次序3的样本点为最优点，该点比例系数K_p和微分系数K_d取值作为踝关节外骨骼PD控制器的参数整定结果If the Euclidean distance between the sample point of process sample order 4 and the sample point of process sample order 3 is less than 1, it means that it has converged, and compare the average evaluation index of process sample order 3 and process sample order 4

Finally, the sample point of process sample order 3 is selected as the optimal point, and the value of the proportional coefficient K _p and differential coefficient K _d of this point is used as the parameter tuning result of the ankle exoskeleton PD controller

Example 2

In this embodiment, the same experimental scene as in Embodiment 1 is adopted, and the parameter tuning process of the PD controller for the ankle exoskeleton in this embodiment will be introduced in detail below.

Step S1. Select the initial PD controller parameters, and establish an initial sample set for parameter tuning of the exoskeleton PD controller by recording the assist force control error of the exoskeleton under the current PD controller parameters. The process is as follows:

Step S11, stipulate that the initial domains of the proportional coefficient K _p and the differential coefficient K _d of the PD controller are both [0,6], perform mechanical sampling on the domains of K _p and K _d with a step size of 0.1, and obtain The searched lattice {X=[K _p ,K _d ], K _p ∈{0,0.1,0.2,…,6},K _d ∈{0,0.1,0.2,…,6}} is mechanically sampled in it , determine six uniformly distributed initial sample points X ₁ =[1,3], X ₂ =[1,6], X ₃ =[3,3], X ₄ =[3,6], X ₅ =[ 5,3], X ₆ =[5,6];

Step S12: Take an unselected initial sample point from the 6 initial sample points in the above-mentioned mechanical sampling, set it as the parameter of the PD controller of the current ankle joint exoskeleton, and stipulate that the wearer should perform walking exercise in the current working condition, and complete Complete 12 gait cycles, and record the control error e _m during the process, m=1,2,...,12;

Step S13: Perform uniform spline interpolation on the recorded control errors in each gait cycle to obtain 12 control error sequences with 100 elements {e _m,i ,m=1,2,...,12| i=1,2,...,100};

具体方法可以参照实施例1的步骤S14；Step S14, calculating an evaluation index E _m for the control error sequence of each gait cycle, and averaging to obtain the average evaluation index

The specific method can refer to step S14 of embodiment 1;

Step S15, repeat step S12 to step S14 until all 6 initial sample points are selected, and finally obtain the average evaluation index corresponding to the initial sample point X _n

In this embodiment, the initial sample set for parameter tuning of the exoskeleton PD controller established in step S1 is shown in Table 3.

Table 3. The parameter tuning initial sample set of embodiment 2

Step S2, use the existing sample set including the initial sample set as training data, import it into the Gaussian regression algorithm for training, repeatedly solve the optimal sample point until the convergence condition is met, and determine the optimal proportional coefficient K _p and differential coefficient K _d value as the parameter tuning result of the ankle exoskeleton PD controller, the specific content can refer to step S2 in the embodiment, in the present embodiment, the process sample set obtained in the step S2 process is as shown in Table 4,

Table 4. The parameter tuning process sample set table of embodiment 2

最后选取过程样本次序5的样本点为最优点，该点比例系数K_p和微分系数K_d取值作为本实施例的踝关节外骨骼PD控制器的参数整定结果。本实施例更改了实施例1中的初始样本数据，通过人机在环调试的踝关节外骨骼PD控制器参数整定方法，经过数次的参数整定迭代过程最终都得到了最优的PD控制器参数，进一步证明了本发明的技术效果。If the Euclidean distance between the sample point of process sample order 5 and the sample point of process sample order 6 is less than 1, it means that it has converged, and compare the average evaluation index of process sample order 5 and process sample order 5

Finally, the sample point of process sample sequence 5 is selected as the optimal point, and the values of the proportional coefficient K _p and the differential coefficient K _d of this point are used as the parameter setting results of the ankle exoskeleton PD controller in this embodiment. In this embodiment, the initial sample data in Embodiment 1 is changed, and the optimal PD controller is finally obtained after several iterations of parameter setting through the man-machine-in-the-loop debugging ankle exoskeleton PD controller parameter setting method. Parameters further prove the technical effect of the present invention.

To sum up, the present embodiment discloses a method for parameter setting of a man-machine-in-the-loop ankle exoskeleton PD controller. Aiming at the technical problem of _Kp and Kd parameters of the PD controller when different wearers and walking conditions will affect the optimal control effect, the human-machine-in-the-loop parameter tuning method of this embodiment can be used for different wearers and walking conditions _. The parameters of the PD controller of the ankle exoskeleton are adjusted under the walking condition to achieve the optimal control effect. The tuning process is adjusted online by man-machine-in-the-loop, without the intervention of specific test environments or equipment. The man-machine-in-the-loop parameter setting method of this embodiment only relies on the calculation of the exoskeleton and its host computer (PC terminal, mobile terminal, MCU, etc.), and does not rely on other specific external test environments and equipment, which can be fast and targeted The parameters of the ankle exoskeleton PD controller are tuned accordingly.

The above-mentioned embodiment is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited by the above-mentioned embodiment, and any other changes, modifications, substitutions, combinations, Simplifications should be equivalent replacement methods, and all are included in the protection scope of the present invention.

Claims (4)

1. An ankle joint exoskeleton PD controller parameter tuning method of man-machine in-loop debugging, it is characterized in that, described parameter tuning method comprises the steps: Step S1, select initial PD controller parameters, and establish an initial sample set for exoskeleton PD controller parameter tuning by recording the assist force control error of the exoskeleton under the current PD controller parameters; Step S2, use the existing sample set including the initial sample set as training data, import it into the Gaussian regression algorithm for training, repeatedly solve the optimal sample point until the convergence condition is met, and determine the optimal proportional coefficient K _p and differential coefficient K _d The value is used as the parameter tuning result of the ankle exoskeleton PD controller. 2. a kind of human-machine-in-the-loop ankle exoskeleton PD controller parameter tuning method according to claim 1, is characterized in that, described step S1 process is as follows: S11. It is stipulated that the initial definition domains of the proportional coefficient K _p and the differential coefficient K _d of the PD controller are [0, p ₀ ] and [0, d ₀ ] respectively, in [0, p ₀ ] and [0, d ₀ ] Sampling at the step size of a ₀ and b ₀ respectively, as the initial sample set, to obtain the initial sample point matrix {X=[K _p , K _d ], K _p ∈ {0, a ₀ , 2a ₀ ,..., p ₀ }, K _d ∈ {0, b ₀ , 2b ₀ ,..., d ₀ }}, do mechanical sampling in the initial sample point lattice, and determine N uniformly distributed initial sample points X _n , n=1 ,2,...,N; S12. Take an unselected initial sample point from the N initial sample points in the above-mentioned mechanical sampling, and set it as the parameter of the PD controller of the current ankle joint exoskeleton, and stipulate that the wearer can walk in the current working condition to complete the complete M gait cycles of , and record the control error em during the process, _m =1, 2,..., M; S13. Perform uniform spline interpolation on the recorded control error in each gait cycle to obtain M control error sequences {e _{m, i} , m=1, 2, ..., M with 100 elements |i=1, 2,..., 100}, where e _{m, i} represent the i-th element of the m-th gait cycle of the control error; S14. Calculate the evaluation index E _m for the control error sequence of each gait cycle, where E _m represents the evaluation index of the mth gait cycle, and the calculation method is as follows:

Among them, F _i is the corresponding target ankle joint auxiliary moment when the control error occurs, and M evaluation indexes E _m are calculated, m=1, 2,..., M, and then the average evaluation index is obtained by averaging the M evaluation indexes

S15. Repeat steps S12 to S14 until N initial sample points are selected, and finally obtain the average evaluation index corresponding to the initial sample point X _n

n=1,2,...,N, where

Indicates the average evaluation index of the nth initial sample point

3. a kind of man-machine-in-the-loop ankle joint exoskeleton PD controller parameter tuning method according to claim 2, is characterized in that, described step S2 process is as follows: S21. Import all existing sample points including the initial sample points into the Gaussian regression algorithm for training, wherein the parameter X _n = [K _p , K _d ] of the current PD controller is used as an independent variable, and the average evaluation index

As a dependent variable, and based on the maximum likelihood method, use nonlinear numerical optimization algorithms such as Newton method or conjugate gradient method to optimize the hyperparameters of the Gaussian regression algorithm so that it best matches the true value distribution and noise distribution of the existing sample set , and then regression to obtain the fitting surface of the independent variable and the dependent variable; S22. Retrieve the predicted dependent variable of each point on the lattice one by one on the fitted surface obtained by the trained Gaussian regression algorithm, and obtain the current optimal sample point where the minimum value of the predicted dependent variable is located x _opt = [K _popt , K _dopt ], where K _popt and K _dopt are the current optimal parameter values of the proportional coefficient K _p and the differential coefficient K _d of the PD controller respectively; S23. Set the current optimal parameter values K _popt and K _dopt as the new proportional coefficient K _p and differential coefficient K _d as the parameters of the current ankle exoskeleton PD controller, and stipulate that the wearer can walk in the current working condition Movement, complete M complete gait cycles, and record the control error em during the process, _m =1, 2,..., M; S24. Repeat steps S13 to S14 to obtain the average evaluation index

is the average evaluation index of the current optimal sample point x _opt ; S25. Record the current optimal sample point x _opt = [k _popt , k _dopt ] as a new sample point, and form the existing sample set together with the initial sample point. If any sample point in the existing sample set is the same as If the Euclidean distance of the new sample point is less than a certain threshold, then select the average evaluation index between the point and the new sample point

The smallest is the best, the average evaluation index

The value of the smallest proportional coefficient K _p and differential coefficient K _d is used as the parameter tuning result of the ankle exoskeleton PD controller. 4. a kind of man-machine-in-the-loop ankle joint exoskeleton PD controller parameter tuning method according to claim 3, is characterized in that, also comprises in the described step S25: Step S25.1. If the new sample point is not on the boundary of the definition domain, and the Euclidean distance to any point in the sample set is less than a certain threshold, it means that convergence has been achieved, and the arbitrary point and the new sample point are selected. average evaluation index

The smallest is the best, the average evaluation index

The value of the smallest proportional coefficient K _p and differential coefficient K _d is used as the parameter tuning result of the ankle exoskeleton PD controller; Step S25.2. If the new sample point is not on the boundary of the defined domain, and the Euclidean distance from any point in the sample set is not less than a certain threshold, then start from step S21 and then execute the process steps in order; Step S25.3, if the new sample point is on the domain boundary of the proportional coefficient _Kp and the differential coefficient Kd, expand the domain to a certain distance Δl outside the boundary to form a new domain [0, _p ] and [ 0, d], start from step S21 and then execute the process steps in sequence.

Images