Lower limb exoskeleton control method based on fuzzy weight model prediction under complex terrain

Abstract

The invention discloses a lower limb exoskeleton control method based on fuzzy weight model prediction in a complex terrain. According to the dynamic weight MPC algorithm based on the fuzzy logic, firstly, the relation between the ground characteristics and the weight coefficient of the MPC algorithm is established, and the weight coefficient in the MPC algorithm is dynamically adjusted by using the fuzzy logic on the basis, so that the joint trajectory tracking precision of the lower limb exoskeleton on the hard road surface is remarkably improved, and the joint trajectory tracking precision of the lower limb exoskeleton on the hard road surface is improved. The environment adaptability to soft and granular complex terrains is enhanced, the walking assisting stability of the lower limb exoskeleton on the complex terrains is effectively improved, the self-adaptive balance of precision and control smoothness is achieved, the robustness to sudden disturbance is greatly improved, and on the premise that the performance is guaranteed through dynamic weight adjustment, the stability of the lower limb exoskeleton is greatly improved. Meaningless high consumption is avoided, and meanwhile, the workload of off-line parameter debugging and field testing is remarkably reduced through a dynamic weight updating strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of exoskeleton control, and specifically, to a lower limb exoskeleton control method based on a fuzzy weight model prediction under complex terrains. Background Art

[0002] In real life, people often need to cross various complex terrains, which not only interfere with the stability of gait but also increase the energy consumption of walking. The current research on lower limb exoskeleton assistive devices mainly focuses on optimizing gait stability and reducing metabolic costs under hard pavement conditions, and its control strategies generally adopt quasi-stiffness control with fixed parameters or local joint impedance optimization methods based on reinforcement learning. However, there are significant deficiencies when facing non-hard pavements (such as sand and muddy soft soil).

[0003] The literature "Z. Wang, B. Wang, Y. Zhou, M. Chen, B. Chen and J. Zhang, 'Model Predictive Control for Attitude Tracking of Rehabilitation Exoskeleton Robots,' 2024 3rd Conference on Fully Actuated System Theory and Applications (FASTA), Shenzhen, China, 2024, pp. 1352 - 1357." proposed an attitude tracking controller for lower limb rehabilitation exoskeleton robots using model predictive control with fixed parameters. This method relies on the assumption of a rigid ground, does not consider the needs of users on non-hard pavements, and ignores the dynamic reaction forces generated by the deformation of granular or soft surfaces, resulting in an increase in the prediction error of joint torques. The fixed-parameter MPC algorithm cannot achieve stable control in such ground environments. At the same time, a single control mode cannot adapt to the differential effects of different terrains on gait kinematics. For example, the hip flexion angle increases due to foot sinking on soft surfaces, while the knee joint stiffness needs to be increased on hard pavements to maintain balance. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the technical problem to be solved by the present invention is to provide a lower limb exoskeleton control method based on a fuzzy weight model prediction under complex terrains.

[0005] The technical solution for the present invention to solve the above technical problem is to provide a lower limb exoskeleton control method based on a fuzzy weight model prediction under complex terrains, which is characterized in that the method includes the following steps:

[0006] Step 1: Based on the disturbing torque of the ground characteristics, establish the dynamic model of the interaction between the lower limb exoskeleton and the ground environment, and then linearize the dynamic model; then discretize the linearized dynamic model and transform it into a state space equation.

[0007] Step 2: According to the state space equation, establish the cost function of the MPC algorithm; then define the target optimization problem according to the cost function of the MPC.

[0008] Step 3: Based on the gain scheduling theory, design the relationship between the output error weight coefficient Q and the friction coefficient μ and the softness / hardness degree S, and the relationship between the control input weight coefficient R and the friction coefficient μ and the softness / hardness degree S. Then, design the fuzzy logic semantics and membership functions from the relationships, thereby establishing fuzzy logic rules, and then use the fuzzy logic rules to achieve dynamic adjustment of the weights.

[0009] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0010] (1) The fuzzy weight model prediction designed in the present invention, that is, the dynamic weight MPC algorithm based on fuzzy logic, first establishes the relationship between the ground characteristics and the weight coefficients of the MPC algorithm, and on this basis, uses fuzzy logic to dynamically adjust the weight coefficients in the MPC algorithm. This not only significantly improves the joint trajectory tracking accuracy of the lower limb exoskeleton on hard pavements, but also enhances the environmental adaptability to soft and granular complex terrains, effectively improving the stability of the lower limb exoskeleton in assisting walking on complex terrains, achieving an adaptive balance between accuracy and control smoothness, greatly improving the robustness to sudden disturbances, and the adjustment of dynamic weights avoids unnecessary high consumption while ensuring performance. At the same time, the dynamic weight update strategy significantly reduces the workload of offline parameter debugging and on-site tests.

[0011] (2) The present invention explicitly introduces the external environmental disturbance (i.e., the disturbing torque of the ground characteristics) into the dynamic model of the lower limb exoskeleton, linearizes and discretizes the nonlinear continuous model at each equilibrium point, and thus converts it into a standard state space equation. This method not only fully considers real disturbances such as road surface reaction forces in the control design process, significantly improving the robustness and tracking accuracy of the lower limb exoskeleton system, but also can quickly solve the MPC optimization problem online with the help of the simplified linear discrete model to meet the real-time requirements. This design is convenient for seamless integration and deployment with the MPC framework, thereby greatly improving the assisting walking performance of the lower limb exoskeleton under complex and changeable terrains.

[0012] (3) Based on the gain scheduling theory, the present invention designs fuzzy logic rules for the friction coefficient and hardness of the ground and the weight coefficients in the MPC algorithm, and dynamically adjusts the weight coefficients (i.e., the output error weight coefficient Q and the control input weight coefficient R) in the MPC algorithm in real time, enabling the lower limb exoskeleton to adaptively switch the joint flexion angle according to the friction coefficient and hardness, and smoothly transition between different terrains. This method not only significantly improves the joint trajectory tracking accuracy of the lower limb exoskeleton on hard pavements, but also enhances the environmental adaptability to soft and granular complex terrains, effectively improving the stability of the lower limb exoskeleton during assisted walking on non-hard terrains. Description of the Drawings

[0013] Figure 1 is the overall flowchart of the present invention;

[0014] Figure 2 is the simplified model diagram of the unilateral leg system of the lower limb exoskeleton of the present invention;

[0015] Figure 3 is the membership function curve diagram of the friction coefficient of the present invention;

[0016] Figure 4 is the membership function curve diagram of the hardness of the present invention;

[0017] Figure 5 is the output angle trajectory tracking curve diagram of the hip joint and knee joint of the lower limb exoskeleton obtained by the fixed low-weight MPC algorithm in Embodiment 1 of the present invention;

[0018] Figure 6 is the output angle trajectory tracking error curve diagram of the hip joint and knee joint of the lower limb exoskeleton obtained by the fixed low-weight MPC algorithm in Embodiment 1 of the present invention;

[0019] Figure 7 is the output angle trajectory tracking curve diagram of the hip joint and knee joint of the lower limb exoskeleton obtained by the fixed high-weight MPC algorithm in Embodiment 1 of the present invention;

[0020] Figure 8 is the output angle trajectory tracking error curve diagram of the hip joint and knee joint of the lower limb exoskeleton obtained by the fixed high-weight MPC algorithm in Embodiment 1 of the present invention;

[0021] Figure 9 is the output angle trajectory tracking curve diagram of the hip joint and knee joint of the lower limb exoskeleton obtained by the MPC algorithm with dynamic weights in Embodiment 1 of the present invention;

[0022] Figure 10 is the output angle trajectory tracking error curve diagram of the hip joint and knee joint of the lower limb exoskeleton obtained by the MPC algorithm with dynamic weights in Embodiment 1 of the present invention. Detailed Embodiment

[0023] Specific embodiments of the present invention are given below. The specific embodiments are only used to further illustrate the present invention in detail and do not limit the protection scope of the present invention.

[0024] The present invention provides a lower limb exoskeleton control method based on fuzzy weight model prediction under complex terrain (hereinafter referred to as the method), which is characterized in that the method includes the following steps:

[0025] Step 1: Establish a dynamic model of the interaction between the lower limb exoskeleton and the ground environment based on the disturbance torque of the ground characteristics (i.e., the friction coefficient μ of the ground and the hardness S of the ground), and then linearize the dynamic model; then discretize the linearized dynamic model and transform it into a state space equation;

[0026] Preferably, step 1 is specifically:

[0027] S11: Establish the ground reaction force according to the ground characteristics:

[0028] Preferably, in step S11, the ground reaction force F in the vertical direction z is:

[0029]

[0030] In formula (1), f is the ground compression amount, which is related to the hardness S and the load; is the ground compression rate, and c is the damping coefficient;

[0031] The ground reaction force F in the horizontal direction x is:

[0032] F x = μ·F z (2)

[0033] In formula (2), μ is the friction coefficient.

[0034] S12: According to the ground reaction force in step S11, fuse the ground disturbance torque to establish a dynamic model of the interaction between the lower limb exoskeleton and the ground environment, as shown in formula (3):

[0035]

[0036] In formula (3), q = [q1, q2] T is the joint angle of the dynamic model, q1 and q2 are the hip joint angle and the knee joint angle respectively; are the angular velocity and angular acceleration of the lower limb exoskeleton joints respectively; τ = [τ1, τ2] T is the joint torque, τ1 and τ2 are the hip joint torque and the knee joint torque respectively; τ extThe disturbing torque generated by the ground on the lower limb exoskeleton is represented as τ ext = J T [F x , F z T , J is the Jacobian matrix; M(q) is the inertia matrix of the dynamic model, which is a symmetric positive definite matrix, M(q) ∈ R n ×n , is the Coriolis and centrifugal force matrix of the dynamic model, C 11 = -m2l1l2q2sinq2, C 22 = 0; G(q) is the gravity matrix of the dynamic model, G(q) ∈ R n×1 , G(q) = (G1, G2) T ,

[0037] In equation (3), for the convenience of analysis, the lower limb exoskeleton system is simplified to the lower limb exoskeleton single - leg system (as Figure 2 shown), m1 represents the mass of the thigh rod, m2 represents the mass of the calf rod, l1 represents the length of the thigh rod, l2 represents the length of the calf rod, and g represents the acceleration due to gravity;

[0038] S13. Linearize the dynamic model, and then use a zero - order hold to discretize the linearized dynamic model and convert it into a state - space equation.

[0039] Preferably, in step S13, the specific steps of linearizing the dynamic model are as follows:

[0040] (1) Define Convert equation (3) into equation (4):

[0041]

[0042] Equation (4) is in a non - linear form and can be simplified to Since the non - linear model has a large amount of calculation and high computational complexity, the non - linear model is linearized: The motion range of the lower limb exoskeleton system is divided into three stages: the standing stage, the support stage, and the swing stage. At each stage, the equilibrium points q0 and τ0 are selected, and then equation (4) is Taylor - expanded at the equilibrium points q0 and τ0, and the first - order terms are retained to obtain equation (5):

[0043]

[0044] In equation (5), B​d = [0; M(q0) -1 T ;

[0045] (2) For the sake of simplifying the analysis, according to the definition of the equilibrium point, rewrite Equation (5) as u = τ.

[0046] Preferably, in step S13, the state - space equation is:

[0047]

[0048] In Equation (6),

[0049] Step 2: According to the state - space equation, establish the cost function of the MPC (Model Predict Control) algorithm; then define the objective optimization problem according to the cost function of MPC;

[0050] Preferably, step 2 is specifically:

[0051] S21: Establish the cost function of the MPC algorithm: The control objective is that the joint output angle q tracks the joint desired angle q d = [q 1d , q 2d T , where q 1d and q 2d are the hip joint desired angle and the knee joint desired angle respectively. Define the joint angle error E as shown in Equation (7):

[0052] E = q - q d (7)

[0053] In order to make the joint output angle q more accurately track the given joint desired angle q d , it is necessary to adjust the relevant joint torques. Accordingly, define the cost function of the MPC algorithm as Equation (8):

[0054]

[0055] In Equation (8), E(k + i) is the predicted state error at time k + i, U(k + i) is the predicted control signal at time k + i, N is the control time - domain length; Q is the output error weight coefficient, which is used to adjust the deviation between the predicted joint output angle and its expected value; R is the control input weight coefficient, which is used to adjust the input joint torque;

[0056] S22: Define the objective optimization problem according to the cost function of the MPC algorithm:

[0057] ​​By dynamically adjusting the output error weight coefficient Q and the control input weight coefficient R to adapt to the changes in complex terrain environments; the output error weight coefficient Q is a diagonal matrix, and the diagonal elements α1 and α2 correspond to the output error weight factors of the hip joint and the knee joint respectively, and the expression is Q = diag[α1, α2]; the control input weight coefficient R is a diagonal matrix, and the diagonal elements β1 and β2 correspond to the control input weight factors of the hip joint and the knee joint respectively, and the expression is R = diag[β1, β2]. Thus, the objective optimization problem of the MPC algorithm is constructed as shown in Equation (9):

[0058]

[0059] In Equation (9), Δq1(i) = q 1d (i) - q1(i), Δq2(i) = q 2d (i) - q2(i), Δτ1(i) = τ1(i + 1) - τ1(i), Δτ2(i) = τ2(i + 1) - τ2(i); Δq1 is the hip joint angle tracking difference, Δq2 is the knee joint angle tracking difference, Δτ1 is the hip joint input torque change rate, Δτ2 is the knee joint input torque change rate; i represents the time;

[0060] This objective optimization problem is a convex optimization problem and has an optimal solution, which can be obtained by the solution function in the simulation software (MATLAB in this embodiment).

[0061] Step 3: Based on the gain scheduling theory, design the relationship between the output error weight coefficient Q and the friction coefficient μ and the softness / hardness degree S, and the relationship between the control input weight coefficient R and the friction coefficient μ and the softness / hardness degree S. Then, design the fuzzy logic semantics and membership functions from the relationships, thereby establishing fuzzy logic rules, and then use the fuzzy logic rules to achieve dynamic adjustment of the weights.

[0062] Preferably, Step 3 is specifically as follows:

[0063] S31: Based on the gain scheduling theory, define the comprehensive road surface evaluation coefficient R road as the scheduling variable, which reflects the dynamic characteristics of the road surface, as shown in Equation (10):

[0064] R road = λ·μ + (1 - λ)·S, λ ∈ [0, 1] (10)

[0065] In Equation (10), λ is the correction factor, which is set according to the prominent characteristics of different road surfaces; μ is the friction coefficient; S is the softness / hardness degree;

[0066] S32: By means of linear interpolation, the comprehensive road surface evaluation coefficient R roadMap to the output error weight coefficient Q and the control input weight coefficient R, corresponding to the output error weight factors α1 and α2 and the control input weight factors β1 and β2 respectively, so as to realize the dynamic adjustment of the two weight factors; the linear interpolation method is shown in Equation (11):

[0067]

[0068] In Equation (11), Q max and Q min respectively represent the maximum and minimum output error weight coefficients, and R max and R min respectively represent the maximum and minimum control input weight coefficients; according to Equation (11), the correlations between the output error weight factors α1 and α2 and the control input weight factors β1 and β2 and the ground characteristics can be obtained, and the fuzzy logic semantics of hard pavements, soft pavements, and granular pavements can be designed accordingly;

[0069] Preferably, in step S32, the fuzzy logic semantics of hard pavements, soft pavements, and granular pavements are specifically:

[0070] On a hard pavement, the friction coefficient μ is relatively large, and the hardness and softness degree S is also relatively large, indicating that the lower limb exoskeleton-assisted walking is subject to less external interference, and the control strategy for the lower limb exoskeleton tends to reduce energy consumption and improve movement efficiency;

[0071] On a soft pavement, the friction coefficient μ is medium, and the hardness and softness degree S is small, which indicates that the lower limb exoskeleton-assisted walking is subject to large external interference, and it is necessary to greatly increase the auxiliary torque of the lower limb exoskeleton to quickly achieve the tracking of the joint trajectory;

[0072] On a granular pavement, the friction coefficient μ is relatively small, and the hardness and softness degree S is also relatively small, which indicates that the lower limb exoskeleton-assisted walking is subject to large external interference, and the control strategy for the lower limb exoskeleton is more inclined to increase the auxiliary torque, improve the tracking efficiency of the joint desired trajectory, and balance the stability of the human body.

[0073] S33. Design the membership function to map the friction coefficient μ and the hardness and softness degree S to the membership degree of fuzzy concepts;

[0074] Preferably, in step S33, a membership function is designed by combining a trigonometric function and a Gaussian function; the mathematical expression of the trigonometric function is a piecewise linear function, which has a low computational complexity, and its linear change enables it to respond quickly in the boundary region. Based on the advantages of the trigonometric function, the two ends of the membership function are designed as trigonometric functions so that when the friction coefficient μ and the softness / hardness degree S are extremely low or extremely high, it can quickly transition from low to medium to respond to the abrupt terrain environment; the output change of the Gaussian function is smooth, which can better simulate the gradual uncertainty in the terrain environment, and its smoothness also makes it insensitive to input noise. Therefore, the middle region of the membership function is designed as a Gaussian function. Such a design can most effectively reflect the complex terrain environment, achieve a smooth transition of the complex terrain, and reduce the influence of input noise.

[0075] Preferably, in step S33, in this embodiment, the membership function curve of the friction coefficient μ is as Figure 3 shown, and the membership function curve of the softness / hardness degree S is as Figure 4 shown.

[0076] S34. According to the fuzzy logic semantics and the membership function, design fuzzy logic rules, and then based on these fuzzy logic rules, implement dynamic adjustment of the weights.

[0077] Preferably, step S34 is specifically: according to the fuzzy logic semantics, obtain the basic expression of the fuzzy logic rules; then map the basic expression of the fuzzy logic rules to the membership degree through the membership function to obtain the fuzzy logic rules of the friction coefficient μ and the softness / hardness degree S and the output error weight coefficient Q, as well as the fuzzy logic rules of the friction coefficient μ and the softness / hardness degree S and the control input weight coefficient R; then according to the fuzzy logic rules, obtain the output error weight coefficient Q and the control input weight coefficient R; then map the output error weight coefficient Q and the control input weight coefficient R to the output error weight factor α1 and α2 and the control input weight factor β1 and β2 in the objective optimization problem of the MPC algorithm to complete the dynamic adjustment of the weights, so as to realize the adaptive control of the MPC algorithm for the lower limb exoskeleton system.

[0078] Preferably, in step S34, the basic expression of the fuzzy logic rules is: when the friction coefficient μ is large and the softness / hardness degree S is also large, the output error weight coefficient Q needs to be set to a small value, and the control input weight coefficient R needs to be set to a large value; when the friction coefficient μ is small and the softness / hardness degree S is also small, the output error weight coefficient Q needs to be set to a large value, and the control input weight coefficient R needs to be set to a small value; when the friction coefficient μ and the softness / hardness degree S are both at a medium level, the output error weight coefficient Q needs to be set to a small value, and the control input weight coefficient R needs to be set to a large value.

[0079] Preferably, in step S34, the fuzzy logic rules of the friction coefficient μ, the hardness S, and the output error weight coefficient Q are shown in Table 1, and the fuzzy logic rules of the friction coefficient μ, the hardness S, and the control input weight coefficient R are shown in Table 2;

[0080] Table 1

[0081]

[0082] Table 2

[0083]

[0084] In Tables 1 and 2, {L, VL, M, VH, H} are the defined fuzzy sets, which respectively correspond to {small, relatively small, medium, relatively large, large} in fuzzy logic semantics.

[0085] Example 1:

[0086] Set the initial values of the joint output angle q and the joint torque τ to 0; set the desired joint angle h, and use the knee and hip joint angles in the standard human walking database. Select [0, 1] and [1, 10] as the basic domains of the friction coefficient μ and the hardness S. Set Scenario 1: The friction coefficient μ is 0.7 - 0.9, and the hardness S is 6 - 7 to simulate an asphalt road surface; set Scenario 2: The friction coefficient μ is 0.3 - 0.5, and the hardness S is 1 - 3 to simulate a sandy road surface. Define the parameters of the lower limb exoskeleton: the mass of the thigh rod m1 = 2.776 kg, the mass of the calf rod m2 = 0.726 kg, the length of the thigh rod l1 = 0.4 m, the length of the calf rod l2 = 0.4 m, and the acceleration due to gravity g = 9.8 m / s 2 。

[0087] Then conduct a simulation experiment:

[0088] Establish an experimental scenario to simulate the walking process of an experimenter from one scenario to another. Define that the scenario switch is achieved in 3.5 seconds. Before 3.5 seconds, use the parameters of Scenario 1 to simulate the walking process on the asphalt road surface, and after 3.5 seconds, use the parameters of Scenario 2 to simulate the walking process on the sandy road surface, simulating the process of an experimenter wearing the lower limb exoskeleton walking from Scenario 1 to Scenario 2. Use the fixed weight MPC algorithm and the MPC algorithm based on fuzzy logic dynamic weight adjustment of the present invention to track the joint output angles of the lower limb exoskeleton respectively.

[0089] Analysis of experimental results: From Figure 5 and Figure 6 it can be seen that if the weight factor meeting the requirements of Scenario 1 is adopted, in Scenario 2, due to the relatively low setting of the output error weight factor, the lower limb exoskeleton system cannot effectively track the desired joint angle trajectory.

[0090] It can be seen from Figure 7 and Figure 8 that when the output error weight factor is increased to meet the control requirements of Scenario 2, two typical problems will occur in the lower limb exoskeleton system under Scenario 1: First, the excessive weight factor causes the controller to saturate, resulting in oscillation at the joints of the lower limb exoskeleton system; Second, due to the controller generating excessive compensation torque, obvious overshoot will occur during the response phase.

[0091] It can be seen from Figure 9 and Figure 10 that the method of the present invention can dynamically adjust the weight coefficient according to the ground characteristics, so that the lower limb exoskeleton system can still stably track the desired joint angle trajectory when the ground characteristics change. Therefore, the present invention realizes the lower limb exoskeleton assisted walking control under complex and variable terrains, solves the stability and energy consumption problems of the fixed weight MPC algorithm, and proves the effectiveness of the present invention.

[0092] The scenarios of this experiment are realized by parameters simulating the actual scenarios. Therefore, under actual working conditions, the present invention can also achieve good tracking effects.

[0093] Matters not described in the present invention are applicable to the prior art.

Claims

1. A lower limb exoskeleton control method based on fuzzy weight model prediction under complex terrain, characterized in that The method includes the following steps: Step 1: Establish a dynamic model for the interaction between the lower limb exoskeleton and the ground environment based on the disturbing torque of the ground characteristics, and then linearize the dynamic model; then discretize the linearized dynamic model to transform it into a state-space equation; Step 2: According to the state-space equation, establish the cost function of the MPC algorithm; then define the target optimization problem according to the cost function of the MPC; Step 3: Based on the gain scheduling theory, design the relationship between the output error weight coefficient Q and the friction coefficient μ and the softness / hardness degree S, and the relationship between the control input weight coefficient R and the friction coefficient μ and the softness / hardness degree S. Then, design the fuzzy logic semantics and membership functions from the relationships to establish fuzzy logic rules, and use the fuzzy logic rules to realize the dynamic adjustment of the weights.

2. The lower limb exoskeleton control method based on fuzzy weight model prediction under complex terrain according to claim 1, characterized in that Specifically, Step 1 is as follows: S11: Establish the ground reaction force according to the ground characteristics; S12: According to the ground reaction force in Step S11, fuse the ground disturbing torque to establish a dynamic model for the interaction between the lower limb exoskeleton and the ground environment, as shown in Equation (3): In Equation (3), q = [q1, q2] T is the joint angle of the dynamic model, where q1 and q2 are the hip joint angle and the knee joint angle respectively; are the angular velocity and angular acceleration of the lower limb exoskeleton joints respectively; τ = [τ1, τ2] T is the joint torque, where τ1 and τ2 are the hip joint torque and the knee joint torque respectively; τ ext is the disturbance torque generated by the ground on the lower limb exoskeleton, expressed as τ ext = J T [F x , F z T , where J is the Jacobian matrix; M(q) is the inertia matrix of the dynamic model, which is a symmetric positive definite matrix, M(q) ∈ R n×n , is the Coriolis and centrifugal force matrix of the dynamic model, C 22 = 0; G(q) is the gravity matrix of the dynamic model, G(q) ∈ R n×1 , G(q) = (G1, G2) T , ​ In Equation (3), for the convenience of analysis, the lower limb exoskeleton system is simplified to a single-leg system of the lower limb exoskeleton. m1 represents the mass of the thigh rod, m2 represents the mass of the calf rod, l1 represents the length of the thigh rod, l2 represents the length of the calf rod, and g represents the acceleration due to gravity; S13: Linearize the dynamic model, and then use a zero-order hold to discretize the linearized dynamic model to transform it into a state-space equation.

3. The lower limb exoskeleton control method based on fuzzy weight model prediction under complex terrain according to claim 2, wherein In step S11, the ground reaction force F in the vertical direction z is as follows: In Equation (1), f is the ground compression amount, which is related to the hardness S and the load; is the ground compression rate, and c is the damping coefficient; The ground reaction force F in the horizontal direction x is as follows: F x = μ·F z (2) In Equation (2), μ is the friction coefficient.

4. The lower limb exoskeleton control method based on fuzzy weight model prediction under complex terrain according to claim 2, characterized in that In Step S13, the specific steps for linearizing the dynamic model are as follows: (1) Definition Convert Equation (3) to Equation (4): Equation (4) is in a non-linear form and is simplified to Next, linearization is performed on the non-linear model: The range of motion of the lower limb exoskeleton system is divided into three stages: the standing stage, the support stage, and the swing stage. At each stage, the equilibrium points q0 and τ0 are selected, and then Equation (4) is expanded using Taylor series at the equilibrium points q0 and τ0, and the first-order terms are retained to obtain Equation (5): In formula (5), B d = [0; M(q0) -1 T ;​ (2) According to the definition of the equilibrium point, rewrite Equation (5) as u = τ; In Step S13, the state-space equation is: In formula (6), 5. The lower limb exoskeleton control method based on fuzzy weight model prediction under complex terrain according to claim 1, characterized in that Specifically, Step 2 is as follows: S21. Establish the cost function of the MPC algorithm: The control objective is that the joint output angle q tracks the joint desired angle q d =[q 1d ,q 2d T , where q 1d and q 2d are the desired hip joint angle and the desired knee joint angle respectively. Define the joint angle error E as shown in Equation (7):​ E = q - q d (7) In order to make the joint output angle \(q\) more accurately track the given desired joint angle \(q_d\) d , it is necessary to adjust the relevant joint torques, and the cost function of the MPC algorithm is defined as in Equation (8): In Equation (8), E(k + i) is the predicted state error at time k + i, U(k + i) is the predicted control signal at time k + i, and N is the control time domain length; Q is the output error weight coefficient, which is used to adjust the deviation between the predicted joint output angle and its expected value; R is the control input weight coefficient, which is used to adjust the input joint torque; S22: Define the target optimization problem according to the cost function of the MPC algorithm: Adapt to the changes in the complex terrain environment by dynamically adjusting the output error weight coefficient Q and the control input weight coefficient R; the output error weight coefficient Q is a diagonal matrix, and the diagonal elements α1 and α2 correspond to the output error weight factors of the hip joint and the knee joint respectively, and the expression is Q = diag[α1, α2]; the control input weight coefficient R is a diagonal matrix, and the diagonal elements β1 and β2 correspond to the control input weight factors of the hip joint and the knee joint respectively, and the expression is R = diag[β1, β2]. Thus, the target optimization problem of the MPC algorithm is constructed as shown in Equation (9): In Equation (9), Δq1(i) = q 1d (i) - q1(i), Δq2(i) = q 2d (i) - q2(i), Δτ1(i) = τ1(i + 1) - τ1(i), Δτ2(i) = τ2(i + 1) - τ2(i); △q1 is the hip joint angle tracking difference, △q2 is the knee joint angle tracking difference, △τ1 is the hip joint input torque change rate, △τ2 is the knee joint input torque change rate; i represents the time This target optimization problem is a convex optimization problem and has an optimal solution.

6. The lower limb exoskeleton control method based on fuzzy weight model prediction under complex terrain according to claim 1, characterized in that Specifically, Step 3 is as follows: S31. Based on the gain scheduling theory, define the comprehensive pavement evaluation coefficient R road as the scheduling variable, which reflects the dynamic characteristics of the pavement, as shown in Equation (10): R road = λμ + (1 - λ)·S, λ ∈ [0, 1] (10) In Equation (10), λ is the correction factor; μ is the friction coefficient; S is the softness / hardness degree; S32. Through linear interpolation, map the comprehensive road surface evaluation coefficient R road to the output error weight coefficient Q and the control input weight coefficient R, corresponding to the output error weight factors α1 and α2 and the control input weight factors β1 and β2 respectively, so as to realize the dynamic adjustment of the two weight factors; the linear interpolation method is shown in Equation (11): In formula (11), Q max and Q min respectively represent the maximum and minimum output error weight coefficients, and R max and R min respectively represent the maximum and minimum control input weight coefficients; according to formula (11), the correlations between the output error weight factors α1 and α2 and the control input weight factors β1 and β2 and the ground characteristics can be obtained, and thus the fuzzy logic semantics of hard pavements, soft pavements and granular pavements can be designed; S33: Design the membership function to map the friction coefficient μ and the softness / hardness degree S to the membership degree of fuzzy concepts; S34. Design fuzzy logic rules according to the fuzzy logic semantics and membership functions, and then implement the dynamic adjustment of weights based on these fuzzy logic rules.

7. The lower limb exoskeleton control method based on fuzzy weight model prediction under complex terrain according to claim 6, characterized in that In step S32, the specific fuzzy logic semantics of the hard road surface, soft road surface, and granular road surface are as follows: On the hard road surface, the friction coefficient μ is relatively large, and the hardness / softness degree S is also relatively large, indicating that the lower limb exoskeleton-assisted walking is subject to less external interference, and the control strategy for the lower limb exoskeleton tends to reduce energy consumption and improve movement efficiency. On the soft road surface, the friction coefficient μ is medium, and the hardness / softness degree S is small, which indicates that the lower limb exoskeleton-assisted walking is subject to large external interference, and it is necessary to greatly increase the auxiliary torque of the lower limb exoskeleton to quickly achieve the tracking of the joint trajectory. On the granular road surface, the friction coefficient μ is relatively small, and the hardness / softness degree S is also relatively small, which indicates that the lower limb exoskeleton-assisted walking is subject to large external interference, and the control strategy for the lower limb exoskeleton is more inclined to increase the auxiliary torque, improve the tracking efficiency of the desired joint trajectory, and balance the stability of the human body.

8. The lower limb exoskeleton control method based on fuzzy weight model prediction under complex terrain according to claim 6, characterized in that, In step S33, a method combining trigonometric functions and Gaussian functions is used to design the membership function; the two ends of the membership function are designed as trigonometric functions, and the middle region of the membership function is designed as a Gaussian function.

9. The lower limb exoskeleton control method based on fuzzy weight model prediction under complex terrain according to claim 6, characterized in that Step S34 is specifically as follows: According to the fuzzy logic semantics, obtain the basic expression of the fuzzy logic rules; then map the basic expression of the fuzzy logic rules to the membership degree through the membership function to obtain the fuzzy logic rules of the friction coefficient μ and the hardness / softness degree S and the output error weight coefficient Q, as well as the fuzzy logic rules of the friction coefficient μ and the hardness / softness degree S and the control input weight coefficient R; then obtain the output error weight coefficient Q and the control input weight coefficient R according to the fuzzy logic rules; then map the output error weight coefficient Q and the control input weight coefficient R to the output error weight factor α1 and α2 and the control input weight factor β1 and β2 in the target optimization problem of the MPC algorithm to complete the dynamic adjustment of weights, thereby realizing the adaptive control of the MPC algorithm for the lower limb exoskeleton system.

10. The lower limb exoskeleton control method based on fuzzy weight model prediction under complex terrain according to claim 9, characterized in that In step S34, the basic expression of the fuzzy logic rules is: when the friction coefficient μ is large and the hardness / softness degree S is also large, the output error weight coefficient Q needs to be set to a small value, and the control input weight coefficient R needs to be set to a large value; when the friction coefficient μ is small and the hardness / softness degree S is also small, the output error weight coefficient Q needs to be set to a large value, and the control input weight coefficient R needs to be set to a small value; when the friction coefficient μ and the hardness / softness degree S are both at a medium level, the output error weight coefficient Q needs to be set to a small value, and the control input weight coefficient R needs to be set to a large value.

Images