Active compensation method for hysteresis response of hybrid platform

Through the active compensation method of hybrid platform hysteresis response, prediction model and online optimization technology are used to solve the hysteresis problem of operation and maintenance system caused by waves, and the stability and accuracy of offshore wind power operation and maintenance system are improved.

CN115639751BActive Publication Date: 2024-12-06HARBIN ENG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202211284267.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-10-20
Publication Date
2024-12-06
Estimated Expiration
2042-10-20

AI Technical Summary

Technical Problem

The offshore wind power operation and maintenance system is affected by the hysteresis caused by waves, resulting in lag in the operation and maintenance system operation and slow output of control signals, affecting the system stability and control performance.

Method used

Using the active compensation method of hybrid platform hysteresis response, the hull posture change information is obtained through sensors, and the prediction model is used to predict the posture change at the next moment, perform kinematic and dynamic analysis, obtain compensation information, correct control signals, and control the actuator movement. , cyclically compensates the motion error at every moment, and optimizes the hidden layer parameters of the prediction model online.

Benefits of technology

Effectively eliminate the hysteresis impact of sea waves on the hull, improve the stability and accuracy of the operation and maintenance system, and ensure the accurate delivery of workforce personnel or equipment to designated locations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115639751B_ABST
    Figure CN115639751B_ABST
Patent Text Reader

Abstract

The present invention provides an active compensation method for the hysteresis response of a hybrid platform, which relates to the field of marine engineering. First, the inverted current attitude change information is sent to a prediction model to obtain a hull attitude change signal at the next moment, and the influence of waves on the hull is predicted in advance. Then, inverse kinematics is calculated to obtain the compensation information of the actuator, and finally, the controller is output to the actuator to compensate for the movement of the actuator at the next moment. Before the actuator acts, the influence that the waves may cause on the hull is eliminated, and the drawbacks of the compensation hysteresis response are solved. In order to ensure the accuracy of the prediction, the operating information of the actuator at the next moment is finally inverted, and the theoretical change value of the hull attitude is reversed. The theoretical change value is subtracted from the actual change value of the hull attitude, and the difference is returned to the prediction model, and the difference is superimposed on the predicted value at the next moment, and the actuator is further corrected to achieve a better compensation effect.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of marine engineering technology, and in particular to an active compensation method for hysteresis response of a parallel-parallel platform. Background Art

[0002] With the rapid increase in population, human beings are facing global challenges such as the shortage of land resources and the deterioration of the ecological environment. In order to cope with the above crises, human beings have gradually turned their attention from land to ocean resource development and utilization. However, offshore wind turbines often break down, and maintenance personnel need to frequently take ships to the offshore platform to repair wind turbines. Therefore, safe offshore operation and maintenance work is very necessary.

[0003] The offshore wind power operation and maintenance system is able to safely deliver the operation and maintenance personnel and equipment to the wind turbine platform. Due to the influence of the waves, the hull is in a bumpy and swaying state, and the operation and maintenance system cannot accurately and smoothly deliver the operation and maintenance personnel and equipment to the wind turbine platform. Due to the huge inertia of the maintenance equipment and the hoisting weight, the operation and maintenance system lags behind the relative movement of the ship; at the same time, it is affected by adverse factors such as mechanical friction and electronic component loss, and the impact of the waves on the operation and maintenance system shows obvious nonlinear characteristics, which ultimately leads to a time lag in the execution of the operation and maintenance system, that is, there is a deviation between the actual execution action and the position that the expected action can reach; in addition, because the actuator takes a certain amount of time to execute the control signal, the signal output of the system is relatively slow, and the currently applied control action cannot be output in time, which further aggravates the generation of time lag, seriously affecting the stability of the system and greatly reducing the control performance of the system. Summary of the invention

[0004] The problem to be solved by the present invention is how to reduce the hysteresis effect caused by waves on the operation and maintenance system.

[0005] In order to solve the above problems, on the one hand, the present invention provides an active compensation method for hysteresis response of a hybrid platform, wherein the hybrid platform includes a parallel platform and a gangway, wherein the parallel platform includes an upper platform and a lower platform, and the active compensation method for hysteresis response of the hybrid platform includes:

[0006] Obtain the current attitude change information of the hull based on the feedback information of the sensor;

[0007] The current posture change information is inverted and sent to the prediction model, and combined with the motion information of the current actuator to obtain the posture change prediction information at the next moment;

[0008] Analyze the hybrid platform according to the posture change prediction information to obtain compensation information, wherein the analysis of the hybrid platform includes kinematic analysis and dynamic analysis;

[0009] Sending the compensation information to a controller, wherein the controller corrects a control signal according to the compensation information, and sends the corrected control signal to the actuator to control the actuator to move at the next moment;

[0010] Based on the movement of the actuator, obtaining the movement information of the actuator at the next moment, and feeding the movement information back to the prediction model;

[0011] The above steps are executed cyclically to compensate for the motion error of the actuator at each moment.

[0012] Furthermore, the inverting of the current posture change information is sent to the prediction model, and combined with the motion information of the current actuator to obtain the posture change prediction information at the next moment includes:

[0013] The prediction model introduces the dynamic characteristic parameters of the hybrid platform during the input and output process, and the expression is:

[0014] y p (τ+1)=f(y(τ),...,y(τ-n+1),u(τ),...,u(τ-m+1),d(τ),...,d(τ-m+1)),

[0015] Among them, y p (τ) is the predicted change of the hydraulic cylinder in the hybrid platform at time τ, y(τ) is the actual change of the hydraulic cylinder in the hybrid platform at time τ, u(τ) represents the input of the prediction model at time τ, d(τ) is the time lag of the hydraulic system of the hybrid platform at time τ, which is generated by the actual delay time of the simulation system, and m and n represent n u and m d respectively.

[0016] According to the above expression, we can know that the past n inputs and m outputs determine the output value y of the system at time τ+1 p (τ+1). The input of the Elman neural network prediction model is: the monitoring input at the previous moment, the output at the current moment of feedback, and the external interference. In the subsequent simulation process, the output y of the Elman network model p It can replace the actual output y. The trained Elman network prediction model can be independent of the mechanical system and can be expressed as follows:

[0017] y p (τ) = f(y p (τ-1),u(τ-1),d(τ-1)),

[0018] Further, analyzing the hybrid platform according to the posture change prediction information to obtain compensation information includes:

[0019] Perform the kinematic analysis on the parallel platform according to the posture change prediction information to obtain the speed and displacement of each joint of the parallel platform and the gangway;

[0020] According to the speed and the displacement, the driving force and the driving torque of the actuator in the hybrid platform are obtained, wherein the compensation information includes the speed and the displacement of each joint and the driving force and the driving torque of the actuator.

[0021] Furthermore, the kinematic analysis includes:

[0022] The homogeneous transformation matrix of the upper platform relative to the lower platform is:

[0023]

[0024] in, is the pose transformation matrix; is the displacement matrix of the upper platform relative to the lower platform;

[0025] and Coordinate transformation formula between:

[0026]

[0027] in, are the coordinates of the upper platform hinge point; is the coordinate of the hinge point of the lower platform;

[0028] The length vector of the driving rod between the upper platform and the lower platform is:

[0029]

[0030] The homogeneous transformation matrix of the end of the gangway relative to the center of the upper platform is:

[0031]

[0032]

[0033]

[0034]

[0035] Wherein, θ1 is the rotation angle of the rotary joint; θ2 is the pitch angle of the pitch joint; and d is the length of the gangway after the telescopic joint moves.

[0036] Further, the kinetic analysis includes:

[0037] The total kinetic energy of the parallel platform is the sum of the total kinetic energy of the upper platform and the total kinetic energy of the driving rod, specifically:

[0038]

[0039] Among them, the total kinetic energy of the upper platform is:

[0040]

[0041] M p is the mass moment of the upper platform, V p is the movement vector matrix of the upper platform;

[0042] Wherein, the total kinetic energy of the driving rod is:

[0043]

[0044] m1 is the mass of the cylinder part of the driving rod, and m2 is the mass of the piston part; V Gi is the velocity of the center of mass of the driving rod,

[0045] The potential energy of the i-th driving rod is:

[0046]

[0047] Where g is the acceleration due to gravity; g i is the distance from the center of mass of the hydraulic cylinder to the hinge point of the lower platform; s1 is the distance from the center of mass of the hydraulic cylinder to the hinge point of the lower platform; s2 is the distance from the telescopic center of mass to the hinge point of the upper platform; h gi is the ratio of the mass center of the telescopic cylinder to the total length of the drive rod l i is the rod length vector of the driving rod; l iz for l i The z-axis coordinate of

[0048] Furthermore, the driving force required for each mechanism in the hybrid platform is:

[0049]

[0050] Where L is the Lagrange coefficient, L = KP, K is kinetic energy, and P is potential energy; and They are partial derivative and derivative operations respectively;

[0051]

[0052] Among them, l is the horizontal distance from the center of mass of the rotating joint to the center of mass of the upper platform, d1 is the distance from the pitch hinge to the center of mass of the fixed gangway, d2 is the distance from the center of mass of the fixed gangway to the center of mass of the movable gangway, θ1 is the rotation angle of the rotating joint; θ2 is the pitch angle of the pitch joint, m3 is the mass of the upper platform; m4 is the mass of the pitch mechanism; and m5 is the mass of the telescopic mechanism.

[0053] Furthermore, the active compensation method for the hysteresis response of the hybrid platform also includes online optimization of the hidden layer parameters of the prediction model;

[0054] The online optimization of the hidden layer parameters of the prediction model includes:

[0055] Determine the range of the maximum layer delay coefficient and the number of hidden layer neurons of the prediction model;

[0056] The layer delay coefficient of the hidden layer of the prediction model is set and fixed as the maximum layer delay coefficient, and the change of the optimization target is recorded by changing the number of neurons in the hidden layer, and the layer delay coefficient that makes the optimization target produce the minimum value is taken as the final number of neurons in the hidden layer;

[0057] When the number of neurons in the hidden layer is the final number of neurons in the hidden layer, the hidden layer delay coefficient is continuously increased from 1 to the maximum layer delay coefficient, the change of the optimization target is recorded, and the hidden layer delay coefficient that makes the optimization target produce the minimum value is taken as the final hidden layer delay coefficient;

[0058] According to the optimized final number of hidden layer neurons and the final hidden layer delay coefficient, an optimized prediction model is created.

[0059] Furthermore, the total input of the hidden layer of the prediction model at the current moment is:

[0060]

[0061] Where ω ji is the weight from the input layer to the hidden layer; ip is the output of the input layer; ω jl The weight of the structural layer; lp is the output of the structural layer; τ is a time function; θ j is the threshold of the jth layer; d is the hidden layer delay coefficient.

[0062] The optimization target is the error function of the prediction model, and the error function is:

[0063]

[0064] Among them, y dkis the expected value of the output layer, and P is the number of overall samples.

[0065] Furthermore, before the online optimization of the hidden layer parameters of the prediction model, the active compensation method for the hysteresis response of the hybrid platform also includes:

[0066] With the goal of minimizing the error function, the weights or thresholds of each layer in the prediction model are adjusted through negative gradient search technology so that the actual output result is close to the expected result.

[0067] Furthermore, the weight ω from the hidden layer to the output layer of the prediction model kj The formula for adjusting (τ+1) during training is:

[0068] ω kj (τ+1)=ω kj (τ)+Δω kj (τ+1),

[0069]

[0070] Among them, Δω kj (τ+1) represents the update of the weights of the neural network of the kth output layer by the jth hidden layer; η represents the learning rate; P represents the number of samples; Represents the hidden layer output value.

[0071] The threshold θ of the output layer of the prediction model k The formula for adjusting (τ+1) during training is:

[0072] θ k (τ+1)=θ k (τ)+Δθ k (τ+1),

[0073]

[0074]

[0075] θ k represents the output layer threshold; y k (τ) represents the actual value of each node in the output layer; y dk (τ) represents the expected value of each node in the output layer; Represents the input amount of the output layer;.

[0076] The weight ω from the input layer to the hidden layer of the prediction model ji The formula for adjusting (τ+1) during training is:

[0077] ω ji (τ+1)=ωji (τ)+Δω ji (τ+1),

[0078]

[0079] Among them, Δω ji (τ+1) represents the update of the weights of the neural network of the jth hidden layer by the i-th input layer; Represents the learning signal of the hidden layer; Represents the output value of the input layer.

[0080] The threshold θ of the hidden layer of the prediction model j The formula for adjusting (τ+1) during training is:

[0081] θ j (τ+1)=θ j (τ)+Δθ j (τ+1),

[0082]

[0083]

[0084] where θ j represents the hidden layer threshold; Represents the input of the hidden layer.

[0085] The weight ω of the structural layer of the prediction model jl The formula for adjusting (τ+1) during training is:

[0086] ω jl (τ+1)=ω jl (τ)+Δω jl (τ+1),

[0087]

[0088]

[0089] where ω jl represents the weight of the structural layer, Represents the output of the hidden layer.

[0090] Compared with the prior art, the present invention has the following beneficial effects:

[0091] The present invention provides an active compensation method for the hysteresis response of a hybrid platform. First, the inverted current attitude change information is sent to the prediction model to obtain the hull attitude change signal at the next moment, and the influence of the waves on the hull is predicted in advance. Then, the inverse kinematics calculation is performed to obtain the compensation information of the actuator, and finally the controller is output to the actuator to compensate for the movement of the actuator at the next moment. Before the actuator acts, the possible influence of the waves on the hull is eliminated, and the drawbacks of the compensation hysteresis response are solved. In order to ensure the accuracy of the prediction, the operation information of the actuator at the next moment is finally inverted, and the theoretical change value of the hull attitude is reversed. The difference between the theoretical change value and the actual change value of the hull attitude is made, and the difference is returned to the prediction model, and the difference is superimposed on the predicted value at the next moment. The actuator is further corrected to achieve a better compensation effect, which promotes the operation and maintenance system to work more smoothly, so as to ensure that the work group or equipment is accurately transported to the designated location. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0093] Figure 1 A schematic diagram of the structure of a hybrid platform in an embodiment of the present invention is shown;

[0094] Figure 2 A flow chart showing an active compensation method for hysteresis response of a parallel-parallel platform in an embodiment of the present invention is shown;

[0095] Figure 3 The structure and coordinate system diagram of the six-degree-of-freedom parallel platform are shown;

[0096] Figure 4 A schematic diagram of the coordinate system of each mechanism in the hybrid platform is shown;

[0097] Figure 5 The distribution of the centroid of the drive rod is shown;

[0098] Figure 6 The Elman hidden layer parameter optimization process is shown;

[0099] Figure 7 A schematic diagram of the AMESim model for the active wave compensation system joint simulation is shown;

[0100] Figure 8 The schematic diagram of the Simulink model for the predictive control co-simulation of the active wave compensation system is shown;

[0101] Fig. 9 The graph showing the roll compensation error analysis results;

[0102] Fig.10 The graph showing the analysis result of pitch compensation error is shown;

[0103] Fig.11 A graph showing the heave compensation error analysis results is shown. DETAILED DESCRIPTION

[0104] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution 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 are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0105] like Figure 1 As shown in the figure, the hybrid wave compensation platform mechanism consists of a six-degree-of-freedom parallel platform and a three-degree-of-freedom serial gangway. In the six-degree-of-freedom parallel platform, the lower platform is connected to the operation and maintenance ship by a fixed pair, and the six driving rods are connected to the upper and lower platforms by Hooke's hinges respectively. The driving rod is realized by a hydraulic cylinder, and the hydraulic cylinder and the piston rod are connected by a cylindrical pair. In the three-degree-of-freedom serial gangway, the rotating mechanism is connected to the upper platform of the six-degree-of-freedom parallel platform through a rotating pair, the pitching mechanism is connected to the rotating mechanism through a rotating pair, and the telescopic joint is connected to the pitching mechanism through a moving pair.

[0106] Figure 2 A flow chart of an active compensation method for hysteresis response of a parallel-parallel platform in an embodiment of the present invention is shown. As described above, the parallel-parallel platform includes a parallel platform and a gangway, and the parallel platform includes an upper platform and a lower platform. The active compensation method for hysteresis response of the parallel-parallel platform includes:

[0107] Step 1: Obtain the current attitude change information of the hull according to the feedback information of the sensor; wherein the attitude change information includes: turbulent displacement changes of the hull caused by sway, pitch, bow pitch, roll, pitch, heave, etc.

[0108] Step 2: Invert the current attitude change information and send it to the prediction model, and combine it with the motion information of the current actuator to obtain the attitude change prediction information at the next moment. When the current attitude change information is inverted and sent to the prediction model, the current attitude change information is inverted, which is equivalent to eliminating the influence of the waves on the hull by reverse elimination, and using the inverted current attitude change information to predict the change of the hull attitude at the next moment.

[0109] Step 3: Analyze the hybrid platform according to the posture change prediction information to obtain compensation information, wherein the analysis of the hybrid platform includes kinematic analysis and dynamic analysis;

[0110] Step 4: Sending the compensation information to the controller, wherein the controller corrects the control signal according to the compensation information, and sends the corrected control signal to the actuator to control the actuator to move at the next moment;

[0111] Step 5: Based on the movement of the actuator, obtain the movement information of the actuator at the next moment, and feed the movement information back to the prediction model;

[0112] The above steps are executed cyclically to compensate for the motion error of the actuator at each moment.

[0113] Because the mechanical structure and hydraulic system of the hybrid wave compensation platform cannot make compensation response in time, a hysteresis response is generated. In order to eliminate the hysteresis response, the present invention firstly sends the inverted current attitude change information to the Elman neural network prediction model to obtain the hull attitude change signal at the next moment, predicts the influence of the waves on the hull in advance, and then performs inverse kinematics analysis to obtain the compensation information of the actuator, and finally outputs it to the actuator through the controller to compensate for the movement of the actuator at the next moment. Before the actuator moves, the possible influence of the waves on the hull is eliminated, and the drawbacks of the compensation hysteresis response are solved. In order to ensure the accuracy of the prediction, the operation information of the actuator at the next moment is finally inverted, and the theoretical change value of the hull attitude is reversed, and the difference between the theoretical change value and the actual change value of the hull attitude is made, and the difference is returned to the prediction model, and the difference is superimposed on the predicted value at the next moment, and the actuator is further corrected to make the compensation effect better, so as to promote the operation and maintenance system to work more smoothly, so as to ensure that the work group personnel or equipment are accurately transported to the designated location, and the operation and maintenance work of the offshore platform can also be carried out normally in bad weather.

[0114] In one embodiment of the present invention, the inverting the current posture change information and sending it to the prediction model, and combining it with the motion information of the current actuator to obtain the posture change prediction information at the next moment includes:

[0115] The prediction model introduces the dynamic characteristic parameters of the hybrid platform during the input and output process, and the expression is:

[0116] y p (τ+1)=f(y(τ),...,y(τ-n+1),u(τ),...,u(τ-m+1),d(τ),...,d(τ-m+1)),

[0117] Among them, y p(τ) is the predicted change of the hydraulic cylinder in the hybrid platform at time τ, y(τ) is the actual change of the hydraulic cylinder in the hybrid platform at time τ, u(τ) represents the input of the prediction model at time τ, d(τ) is the time lag of the hydraulic system of the hybrid platform at time τ, which is generated by the actual delay time of the simulation system, m and n represent n u and m d respectively, and f() represents y p Which factors are relevant to the prediction of (τ+1).

[0118] According to the above expression, we can know that the output value y of the system at time τ+1 is determined by the n outputs and m inputs in the past. p (τ+1). The input of the Elman neural network prediction model is: the monitoring input at the previous moment, the output at the current moment of feedback, and the external interference. In the subsequent simulation process, the output y of the Elman network model p It can replace the actual output y. The trained Elman network prediction model can be independent of the mechanical system and can be expressed as follows:

[0119] y p (τ) = f(y p (τ-1),u(τ-1),d(τ-1)),

[0120] In solving the problem of actuator hysteresis in active wave compensation system, the previous control signal will affect the current control signal output by the system, but it is impossible to determine how many previous control signals affect the current control signal, that is, the hysteresis time of the hysteresis system cannot be theoretically deduced. Therefore, based on the principle of Elman network, an active compensation method for hysteresis response of hybrid platform is proposed.

[0121] In an embodiment of the present invention, analyzing the hybrid platform according to the posture change prediction information to obtain compensation information includes:

[0122] Perform the kinematic analysis on the parallel platform according to the posture change prediction information to obtain the speed and displacement of each joint of the parallel platform and the gangway;

[0123] According to the speed and the displacement, the driving force and the driving torque of the actuator in the hybrid platform are obtained, wherein the compensation information includes the speed and the displacement of each joint and the driving force and the driving torque of the actuator.

[0124] In one embodiment of the present invention, Figure 3 As shown, firstly, the six-degree-of-freedom platform is modeled, and kinematics is calculated to establish a coordinate system Bx at the center point B of the lower platform. b y b z bThe center point A of the upper platform establishes a coordinate system Ax a y a z a The lower platform is a static platform fixed on the ship, and the upper platform is a dynamic platform, which moves with the change of the length of the driving rod. At the same time, the movement posture of the upper platform can be expressed by the A coordinate system relative to the B coordinate system. The hinge point of the upper platform is A i (i=1,2,3,4,5,6), the hinge point of the lower platform is B i (i=1,2,3,4,5,6).

[0125] The homogeneous transformation matrix of the upper platform relative to the lower platform is:

[0126]

[0127] in, is the pose transformation matrix; is the displacement matrix of the upper platform relative to the lower platform;

[0128] and Coordinate transformation formula between:

[0129]

[0130] in, are the coordinates of the upper platform hinge point; is the coordinate of the hinge point of the lower platform;

[0131] Assume that the lengths of the six connecting rods of the six-degree-of-freedom parallel platform are l i (i=1,2,3,4,5,6), the two ends of the same connecting rod correspond to the hinge points of the upper and lower platforms respectively, so the length vector of the driving rod between the upper platform and the lower platform is:

[0132]

[0133] like Figure 4 As shown, in order to accurately describe the motion mode of the mechanism, this paper uses the DH parameter method to establish the coordinate system of each joint of the three-degree-of-freedom series gangway ladder. The reference coordinate system of the three-degree-of-freedom series gangway ladder is ox o y o z o Since the three-degree-of-freedom series gangway is fixed to the upper platform of the six-degree-of-freedom parallel platform through a slewing joint, the point o is related to the upper platform coordinate system Ax a y a z aThe origin A coincides. The rotation joint coordinate system is o1-x1y1z1, and the rotation axis is z1; the pitch joint coordinate system is o2-x2y2z2, and the rotation axis is z2; the telescopic joint coordinate system is o3-x3y3z3, and the moving axis is z3. The distance from o2 to o is l, and the distance from o2 to o3 is d.

[0134] Inverse kinematics solution is to calculate the motion of each joint when the position of the end of the gangway is known. The homogeneous transformation matrix of the end of the gangway relative to the center of the upper platform is:

[0135]

[0136] The amount of motion of each joint is expressed as:

[0137]

[0138]

[0139]

[0140] Among them, θ1 is the rotation angle of the revolute joint; θ2 is the pitch angle of the pitch joint; d is the length of the gangway after the telescopic joint moves.

[0141] After calculating the inverse kinematics, the required forces and torques of each actuator are calculated, and the kinetic energy and potential energy of the upper platform are analyzed first. The dynamic modeling of the six-degree-of-freedom parallel platform is established using the Lagrangian method.

[0142] It can be seen from observation that in the six-degree-of-freedom Stewart parallel platform, the motion of the upper platform can be decomposed into translation and rotation, and the total kinetic energy of the upper platform is:

[0143] The total kinetic energy of the upper platform is:

[0144]

[0145] M p is the mass moment of the upper platform, V p is the platform's movement vector matrix.

[0146] The total kinetic energy of the driving rod is:

[0147]

[0148] m1 is the mass of the cylinder part of the driving rod, and m2 is the mass of the piston part; V Gi is the velocity of the center of mass of the driving rod,

[0149] For the calculation of the kinetic energy and potential energy of the drive rod, the motion of the drive rod cylinder and the drive rod piston is represented by the motion of their respective centers of mass, so the position of the center of mass of the drive rod needs to be determined. Figure 5 As shown, it is assumed that the mass of the cylinder part of the driving rod is m1, and the mass of the piston part is m2; the distance from the center of mass of the cylinder to the hinge point of the lower platform is s1, and the distance from the center of mass of the piston to the hinge point of the upper platform is s2; the speed of the center of mass of the driving rod is V Gi , the length of the driving rod is l i , the potential energy of the i-th driving rod is:

[0150]

[0151] Where g is the acceleration due to gravity; g i is the distance from the center of mass of the hydraulic cylinder to the hinge point of the lower platform; s1 is the distance from the center of mass of the hydraulic cylinder to the hinge point of the lower platform; s2 is the distance from the telescopic center of mass to the hinge point of the upper platform; h gi is the ratio of the center of mass of the telescopic cylinder to the total length of the drive rod l i is the rod length vector of the driving rod; l iz for l i The z-axis coordinate of .

[0152] Through the above derivation process, we can get that the total kinetic energy of the parallel platform is the sum of the total kinetic energy of the upper platform and the total kinetic energy of the driving rod, specifically:

[0153]

[0154] The Lagrangian method is also used to carry out dynamic analysis and equation derivation of the three-degree-of-freedom series gangway. According to the previous kinematic modeling, the coordinate system of the rotation joint (first joint) is o1-x1y1z1, and the rotation axis is z1; the coordinate system of the pitch joint (second joint) is o2-x2y2z2, and the rotation axis is z2; the coordinate system of the telescopic joint (third joint) is o3-x3y3z3, and the moving axis is z3. Among them, the horizontal distance from the center of mass of the rotation joint to the center of mass of the upper platform is l, the distance from the center of mass of the pitch hinge to the fixed gangway is d1, and the distance from the center of mass of the fixed gangway to the center of mass of the movable gangway is d2. θ1 is the rotation angle of the rotation joint; θ2 is the pitch angle of the pitch joint. The platform mass is m3; the mass of the pitch mechanism is m4; the mass of the telescopic mechanism is m5. The moment of inertia of the rotation joint is I1, the moment of inertia of the pitch joint is I2, and the moment of inertia of the pitch joint is I3.

[0155] The driving force required for each mechanism in the hybrid platform is:

[0156]

[0157] Where L is the Lagrange coefficient, L = KP, K is kinetic energy, and P is potential energy; and They are partial derivative and derivative operations respectively;

[0158] For the kinetic energy of the first joint:

[0159]

[0160] The kinetic energy of the second joint not only includes the rotation brought by the first joint, but also the pitch, so the kinetic energy of the second joint is:

[0161]

[0162] Similarly, for the third joint, there is not only the kinetic energy brought by the extension and contraction, but also the momentum brought by the first and second joints. The kinetic energy of the third joint is:

[0163]

[0164] The total energy of motion is:

[0165]

[0166] Assuming that the gravitational potential energy is 0 when horizontal, the gravitational potential energy of each mechanism is obtained as:

[0167] P1=0,

[0168] P2=m4gd1sinθ1,

[0169] P3=m5g(d1+d2)sinθ2,

[0170] P1 is the gravitational potential energy of the rotating mechanism; P2 is the gravitational potential energy of the rotating mechanism; P3 is the gravitational potential energy of the telescopic mechanism.

[0171] Therefore, the total gravitational potential energy: P = P1 + P2 + P3 = m4gd1sinθ1 + m5g(d1 + d2)sinθ2,

[0172] Through the above analysis, we can get:

[0173]

[0174] Among them, l is the horizontal distance from the center of mass of the rotating joint to the center of mass of the upper platform, d1 is the distance from the pitch hinge to the center of mass of the fixed gangway, d2 is the distance from the center of mass of the fixed gangway to the center of mass of the movable gangway, θ1 is the rotation angle of the rotating joint; θ2 is the pitch angle of the pitch joint, m3 is the mass of the upper platform; m4 is the mass of the pitch mechanism; and m5 is the mass of the telescopic mechanism.

[0175] Find the partial derivative and reciprocal of L

[0176] Finally, we get:

[0177]

[0178]

[0179]

[0180] The forces required for the movement of the first joint, the second joint and the third joint respectively.

[0181] In one embodiment of the present invention, the active compensation method for hysteresis response of the parallel-parallel platform further comprises online optimization of hidden layer parameters of the prediction model;

[0182] like Figure 6 As shown, the online optimization of the hidden layer parameters of the prediction model includes:

[0183] Determine the range of the maximum layer delay coefficient and the number of hidden layer neurons of the prediction model;

[0184] The layer delay coefficient of the hidden layer of the prediction model is set and fixed as the maximum layer delay coefficient, and the change of the optimization target is recorded by changing the number of neurons in the hidden layer, and the layer delay coefficient that makes the optimization target produce the minimum value is taken as the final number of neurons in the hidden layer;

[0185] When the number of neurons in the hidden layer is the final number of neurons in the hidden layer, the hidden layer delay coefficient is continuously increased from 1 to the maximum layer delay coefficient, the change of the optimization target is recorded, and the hidden layer delay coefficient that makes the optimization target produce a minimum value is taken as the final hidden layer delay coefficient.

[0186] According to the optimized final number of hidden layer neurons and the final hidden layer delay coefficient, an optimized prediction model is created.

[0187] In one embodiment of the present invention, before the online optimization of the hidden layer parameters of the prediction model, the active compensation method for the hysteresis response of the hybrid platform further includes:

[0188] With the goal of minimizing the error function, the weights or thresholds of each layer in the prediction model are adjusted through negative gradient search technology so that the actual output result is close to the expected result.

[0189] Among them, the total input of the hidden layer of the prediction model at the current moment is:

[0190]

[0191] Where ωji is the weight from the input layer to the hidden layer; O ip is the output of the input layer; ω jl The weight of the structural layer; lp is the output of the structural layer; τ is a time function; θ j is the threshold of the jth layer; d is the delay coefficient of the hidden layer. For the selection of the number of neurons in a single hidden layer, a smaller number of nodes should be selected to optimize the network structure and improve the computational efficiency while ensuring that the network learning can correctly express the input and output relationship.

[0192] The optimization target is the error function of the prediction model, and the error function is:

[0193]

[0194] Among them, y dk is the expected value of the output layer, and P is the number of overall samples;

[0195] The core idea of ​​the Elman network algorithm is to minimize the total mean square error, that is, to minimize the error function value as the goal, and to adjust the weights or thresholds of each layer in the prediction model through negative gradient search technology so that the actual output result is close to the expected result.

[0196] When adjusting the weights or thresholds of each layer, the basis for the adjustment is as follows.

[0197] Among them, the weight ω from the hidden layer to the output layer of the prediction model kj The formula for adjusting (τ+1) during training is:

[0198] ω kj (τ+1)=ω kj (τ)+Δω kj (τ+1),

[0199]

[0200] Among them, Δω kj (τ+1) represents the update of the weights of the neural network of the kth output layer by the jth hidden layer; η is the learning rate; P is the number of samples; Represents the hidden layer output value.

[0201] The threshold θ of the output layer of the prediction model k The formula for adjusting (τ+1) during training is:

[0202] θ k (τ+1)=θ k (τ)+Δθ k (τ+1),

[0203]

[0204]

[0205] θ k represents the output layer threshold; y k (τ) is the actual value of each node in the output layer; y dk (τ) is the expected value of each node in the output layer; is the input quantity of the output layer;.

[0206] The weight ω from the input layer to the hidden layer of the prediction model ji The formula for adjusting (τ+1) during training is:

[0207] ω ji (τ+1)=ω ji (τ)+Δω ji (τ+1),

[0208]

[0209] Among them, Δω ji (τ+1) represents the update of the weights of the neural network of the jth hidden layer by the i-th input layer; is the learning signal of the hidden layer; Represents the output value of the input layer.

[0210] The threshold θ of the hidden layer of the prediction model j The formula for adjusting (τ+1) during training is:

[0211] θ j (τ+1)=θ j (τ)+Δθ j (τ+1),

[0212]

[0213]

[0214] where θ j represents the hidden layer threshold; is the input of the hidden layer.

[0215] The weight ω of the structural layer of the prediction model jl The formula for adjusting (τ+1) during training is:

[0216] ω jl (τ+1)=ω jl (τ)+Δω jl (τ+1),

[0217]

[0218]

[0219] where ω jl is the weight of the structural layer, is the output of the hidden layer.

[0220] The above can realize the adjustment of the connection weights and thresholds between each layer of the Elman neural network prediction model. According to the error function, the error E n On the one hand, the generation of is the setting of the connection weights and thresholds of each layer of neurons. This problem can be adjusted by calculating the weights and thresholds. On the other hand, combined with the method of optimizing the hidden layer parameters of the Elman neural network in the previous embodiment, the prediction model can be further optimized.

[0221] In one embodiment of the present invention, in order to support the smooth execution of the above method and prediction model, the following system is designed to generate the interface module SimuCosim of AMESim (a multidisciplinary complex system modeling and simulation platform) in Simulink (a visual simulation tool, a block diagram environment) to realize the joint simulation of Simulink and AMESim. Thus, Simulink-AMESim-ADAMS electromechanical and hydraulic joint simulation is realized.

[0222] Simulink inputs the control quantity to the servo valve of AMESim through the control algorithm, and the AMESim actuator generates corresponding motion and inputs the motion quantity into ADAMS (Automatic Analysis of Mechanical System Dynamics). Thus, the motion quantity input by each joint of the mechanism in ADAMS software is finally output as the deviation between the end of the hybrid mechanism and the reference point in the compensation direction, which can reflect the control effect of the active wave compensation system. Figure 7 As shown, the AMESim model of active wave compensation system joint simulation is shown. Figure 8 Simulink model for the joint simulation of Elman predictive control for active wave compensation system.

[0223] A reference point is set at the end of the active wave compensation system hybrid mechanism to simulate the wind turbine platform in actual operation. Through simulation, the error curves of the hybrid mechanism after motion compensation and the reference point are obtained under the control of the ideal compensation and PID control algorithm and Elman predictive control algorithm, respectively. Fig. 9 , Fig.10 and Fig.11 shown.

[0224] When the system is ideally compensated, the compensation error is almost zero, and the end of the hybrid mechanism is almost relatively static with the reference point. However, when there is a time lag in the hydraulic system actuator, the compensation errors in the roll, pitch and heave directions become larger, so it is necessary to solve the time lag problem through the control system to reduce the compensation error, thereby ensuring that the hybrid mechanism remains stationary with the wind turbine platform under the interference of waves. Under the control of the two methods, the influence of waves on the hybrid mechanism has been effectively controlled. Fig. 9 is the compensation error in the roll direction. When there is no compensation, the input cycle is five seconds and the maximum amplitude is 6°. Under PID control, the maximum amplitude in the roll direction is 2°. Under Elman predictive control, the maximum amplitude in the roll direction is 1°. Fig.10 It is shown that the compensation error in the pitch direction is 1.5° in the input cycle of five seconds without compensation. Under PID control, the maximum amplitude of the pitch direction is 0.4°. Under Elman predictive control, the maximum amplitude of the pitch direction is less than 0.2°. Fig.11 The compensation error in the pitch direction is shown as 2.5m in the case of no compensation, that is, the input cycle is five seconds, the maximum amplitude is 0.4m in the heave direction under PID control, and the maximum amplitude is 0.2m in the heave direction under Elman prediction control. It can be shown that although there is a certain error between the Elman prediction value and the measured value, it does not have much impact on the determination of the control amount of the system, and the compensation effect is more effective than that of traditional PID control.

[0225] Table 1 is a comparison table of the results obtained by the control method based on Elman neural network proposed in the present invention and the traditional PID control method:

[0226] Table 1 Comparison of Elman neural network prediction control error and traditional PID control method error

[0227] No compensation error Traditional PID control error Elman predictive control error of the present invention Roll 6° 2° 1° Pitch 1.5° 0.4° 0.2° Heave 2.5m 0.4m 0.2m

[0228] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An active compensation method for hysteresis response of a parallel-parallel platform, characterized in that: The hybrid platform includes a parallel platform and a gangway, the parallel platform includes an upper platform and a lower platform, and the active compensation method for the hysteresis response of the hybrid platform includes: Obtain the current attitude change information of the hull based on the feedback information of the sensor; The current posture change information is inverted and sent to the prediction model, and combined with the motion information of the current actuator to obtain the posture change prediction information at the next moment; Analyze the hybrid platform according to the posture change prediction information to obtain compensation information, wherein the analysis of the hybrid platform includes kinematic analysis and dynamic analysis; Sending the compensation information to a controller, wherein the controller corrects a control signal according to the compensation information, and sends the corrected control signal to the actuator to control the actuator to move at the next moment; Based on the movement of the actuator, obtaining the movement information of the actuator at the next moment, and feeding the movement information back to the prediction model; Execute the above steps cyclically to compensate for the motion error of the actuator at each moment; The kinematic analysis includes: The homogeneous transformation matrix of the upper platform relative to the lower platform is: in, is the pose transformation matrix; is the displacement matrix of the upper platform relative to the lower platform; and Coordinate transformation formula between: in, are the coordinates of the hinge point of the upper platform; is the coordinate of the hinge point of the lower platform; The length vector of the driving rod between the upper platform and the lower platform is: The homogeneous transformation matrix of the end of the gangway relative to the center of the upper platform is: Wherein, θ1 is the rotation angle of the rotary joint; θ2 is the pitch angle of the pitch joint; d is the length of the gangway after the telescopic joint moves; The kinetic analysis includes: The total kinetic energy of the parallel platform is the sum of the total kinetic energy of the upper platform and the total kinetic energy of the driving rod, specifically: Among them, the total kinetic energy of the upper platform is: M p is the mass moment of the upper platform, V p is the movement vector matrix of the upper platform; Wherein, the total kinetic energy of the driving rod is: m1 is the mass of the cylinder part of the driving rod, and m2 is the mass of the piston part; V Gi is the velocity of the center of mass of the driving rod, The potential energy of the i-th driving rod is: Where g is the acceleration due to gravity; g i is the distance from the center of mass of the hydraulic cylinder to the hinge point of the lower platform; s1 is the distance from the center of mass of the hydraulic cylinder to the hinge point of the lower platform; s2 is the distance from the telescopic center of mass to the hinge point of the upper platform; h gi h is the ratio of the telescopic cylinder mass center to the total length of the drive rod gi =g i / l i ; l i is the rod length vector of the driving rod; l iz for l i The z-axis coordinate of .

2. The active compensation method for hysteresis response of a parallel-parallel platform according to claim 1, characterized in that: The inverting of the current posture change information is sent to the prediction model, and combined with the motion information of the current actuator to obtain the posture change prediction information at the next moment includes: The prediction model introduces the dynamic characteristic parameters of the hybrid platform during the input and output process, and the expression is: y p (τ+1)=f(y(τ),...,y(τ-n+1),u(τ),...,u(τ-m+1),d(τ),...,d(τ-m+1)) Among them, y p (τ) is the predicted change of the hydraulic cylinder in the hybrid platform at time τ, y(τ) is the actual change of the hydraulic cylinder in the hybrid platform at time τ, u(τ) represents the input of the prediction model at time τ, d(τ) is the time lag of the hydraulic system of the hybrid platform at time τ, which is generated by the actual delay time of the simulation system, and m and n represent n u and m d respectively.

3. The active compensation method for hysteresis response of a parallel-parallel platform according to claim 1, characterized in that: The analyzing the hybrid platform according to the posture change prediction information to obtain compensation information includes: Perform the kinematic analysis on the parallel platform according to the posture change prediction information to obtain the speed and displacement of each joint of the parallel platform and the gangway; According to the speed and the displacement, the driving force and the driving torque of the actuator in the hybrid platform are obtained, wherein the compensation information includes the speed and the displacement of each joint and the driving force and the driving torque of the actuator.

4. The active compensation method for hysteresis response of a parallel-parallel platform according to claim 1, characterized in that: The driving force required for each mechanism in the hybrid platform is: Where L is the Lagrange coefficient, L = KP, K is kinetic energy, and P is potential energy; and They are partial derivative and derivative operations respectively; Among them, l is the horizontal distance from the center of mass of the rotating joint to the center of mass of the upper platform, d1 is the distance from the pitch hinge to the center of mass of the fixed gangway, d2 is the distance from the center of mass of the fixed gangway to the center of mass of the movable gangway, θ1 is the rotation angle of the rotating joint; θ2 is the pitch angle of the pitch joint, m3 is the mass of the upper platform; m4 is the mass of the pitch mechanism; and m5 is the mass of the telescopic mechanism.

5. The active compensation method for hysteresis response of a parallel-parallel platform according to claim 1, characterized in that: It also includes online optimization of the hidden layer parameters of the prediction model; The online optimization of the hidden layer parameters of the prediction model includes: Determine the range of the maximum layer delay coefficient and the number of hidden layer neurons of the prediction model; The layer delay coefficient of the hidden layer of the prediction model is set and fixed as the maximum layer delay coefficient, and the change of the optimization target is recorded by changing the number of neurons in the hidden layer, and the layer delay coefficient that makes the optimization target produce the minimum value is taken as the final number of neurons in the hidden layer; When the number of neurons in the hidden layer is the final number of neurons in the hidden layer, the hidden layer delay coefficient is continuously increased from 1 to the maximum layer delay coefficient, the change of the optimization target is recorded, and the hidden layer delay coefficient that makes the optimization target produce the minimum value is taken as the final hidden layer delay coefficient; According to the optimized final number of hidden layer neurons and the final hidden layer delay coefficient, an optimized prediction model is created.

6. The active compensation method for hysteresis response of a parallel-parallel platform according to claim 5, characterized in that: The total input of the hidden layer of the prediction model at the current moment is: Where ω ji is the weight from the input layer to the hidden layer; ip is the output of the input layer; ω jl The weight of the structural layer; lp is the output of the structural layer; τ is a time function; θ j is the threshold of the jth layer; d is the delay coefficient of the hidden layer; The optimization target is the error function of the prediction model, and the error function is: Among them, y dk is the expected value of the output layer, and P is the number of overall samples.

7. The active compensation method for hysteresis response of a parallel-parallel platform according to claim 5, characterized in that: Before the online optimization of the hidden layer parameters of the prediction model, the method further includes: With the goal of minimizing the error function, the weights or thresholds of each layer in the prediction model are adjusted through negative gradient search technology so that the actual output result is close to the expected result.

8. The active compensation method for hysteresis response of a parallel-parallel platform according to claim 7, characterized in that: The weight ω from the hidden layer to the output layer of the prediction model kj The formula for adjusting (τ+1) during training is: oh kj (τ+1)=ω kj (t)+Do kj (t+1), Among them, Δω kj (τ+1) represents the update of the weights of the neural network of the kth output layer by the jth hidden layer; η represents the learning rate; P represents the number of samples; Represents the output value of the hidden layer; The threshold θ of the output layer of the prediction model k The formula for adjusting (τ+1) during training is: i k (τ+1)=θ k (t)+Δθ k (t+1), θ k represents the output layer threshold; y k (τ) represents the actual value of each node in the output layer; y dk (τ) represents the expected value of each node in the output layer; Represents the input amount of the output layer; The weight ω from the input layer to the hidden layer of the prediction model ji The formula for adjusting (τ+1) during training is: oh ji (τ+1)=ω ji (t)+Do ji (t+1), Among them, Δω ji (τ+1) represents the update of the weights of the neural network of the jth hidden layer by the i-th input layer; Represents the learning signal of the hidden layer; Represents the output value of the input layer; The threshold θ of the hidden layer of the prediction model j The formula for adjusting (τ+1) during training is: i j (τ+1)=θ j (t)+Δθ j (t+1), where θ j represents the hidden layer threshold; Represents the input amount of the hidden layer; The weight ω of the structural layer of the prediction model jl The formula for adjusting (τ+1) during training is: oh jl (τ+1)=ω jl (t)+Do jl (t+1), where ω jl represents the weight of the structural layer, Represents the output of the hidden layer.