A method for calculating adaptive control coefficients

Through the calculation and update speed proportional control coefficients through the three-layer neural network, the problem of poor classic PID control effect in multi-robot cooperative transportation is solved, more efficient formation control and flexibility is achieved, and enterprise equipment costs are reduced.

CN117826592BActive Publication Date: 2024-11-26JIANGSU JINLING INST OF INTELLIGENT MFG CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202311808863.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-12-26
Publication Date
2024-11-26
Estimated Expiration
2043-12-26

AI Technical Summary

Technical Problem

In the multi-robot collaborative transport scenario, classic PID control is difficult to achieve ideal formation control effects, and the multi-robot system is complex and difficult to abstract accurate mathematical models, resulting in insufficient formation control accuracy and flexibility, which increases the cost of enterprise equipment.

Method used

A three-layer neural network is used to adaptive online adjustment of the speed proportional control coefficient. By calculating the reference speed and position deviation of the robot, the neural network connection weight matrix is ​​corrected to realize the calculation and update of the adaptive control coefficient.

Benefits of technology

It greatly reduces debugging time, improves formation control accuracy and flexibility, reduces enterprise equipment costs, and is suitable for formation control at different pilot robot operating speeds, which is in line with the actual application scenarios of the engineering.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117826592B_ABST
    Figure CN117826592B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for calculating an adaptive control coefficient, the method comprising the following steps: step 1, calculating the reference speed and reference position of the following robot according to the running speed and formation constraints of the pilot robot; step 2, calculating the position deviation according to the reference position and the actual position; step 3, using a three-layer neural network to calculate the speed proportional control coefficient and calculating the center speed of the following robot in combination with the reference speed; step 4, correcting the neural network connection weight matrix. The method enables the speed control parameters to adapt to the changes in parameters such as the running speed and position of the formation, greatly reducing the time for manual parameter adjustment and improving the control accuracy of the formation, so that the formation cooperative transportation is more in line with the actual application scenario of the engineering.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of intelligent industrial logistics, and in particular to a control parameter calculation method with adaptive characteristics. Background Art

[0002] Robots are automated equipment that integrates advanced technologies from multiple disciplines, such as mechanics, electronics, automatic control, and artificial intelligence. Mobile robots have been applied to various industrial production occasions, replacing manual labor to perform some transfer tasks, and have shown great advantages in transfer efficiency and transfer safety. However, the transportation method using a single mobile robot still has some limitations in the transfer scenario of oversized and overweight parts: the increase in load demand has led to a significant increase in the manufacturing and maintenance costs of mobile robots; the "special machine for special purpose" large mobile robots have a low degree of reuse, which increases the equipment cost of the enterprise and is not conducive to the space planning of the factory. In this scenario, if multiple mobile robots are used for collaborative transfer, the transfer flexibility and multi-product adaptability can be greatly improved, and the equipment cost of the enterprise can be reduced. However, multiple mobile robots have high requirements for the control accuracy of the formation; and multiple robots are complex systems, and it is difficult to abstract an accurate mathematical model, and the use of classic PID control cannot achieve ideal control effects. Therefore, it is necessary to use intelligent control technology with self-learning, self-adaptation, and self-optimization to achieve better results in formation control. Summary of the invention

[0003] In order to achieve better control effect for multi-robot collaborative transport, the present invention aims to provide a method for calculating an adaptive control coefficient, which constructs a three-layer neural network comprising an input layer, a hidden layer and an output layer to realize adaptive online adjustment of the speed control coefficient.

[0004] The present application is implemented by the following scheme: a method for calculating an adaptive control coefficient, the method comprising the following steps:

[0005] Step 1: Calculate the reference speed and reference position of the following robot according to the running speed and formation constraints of the leading robot;

[0006] Step 2, calculating the position deviation according to the reference position and the actual position;

[0007] Step 3: Use a three-layer neural network to calculate the speed proportional control coefficient and calculate the center speed of the following robot in combination with the reference speed;

[0008] Step 4: Correct the neural network connection weight matrix.

[0009] Furthermore, the calculation method of the reference speed in step 1 is specifically as follows:

[0010] Step 1.1, the formation constraint is to design a formation. Under the formation constraint, the reference position of the follower robot is the position coordinate of the origin of the follower robot coordinate system in the leader robot coordinate system, using [x d y d ]express;

[0011] Step 1.2, the speed of the pilot robot uses the x-speed v x , y to v y , the vehicle angular velocity ω and the rotation center [x c y c ] four parameters are represented; according to the reference position of the following robot, the rotation radius R of the following robot is obtained s for:

[0012]

[0013] Step 1.3, calculate the reference speed v of the following robot d :

[0014] v d =ω·R s

[0015] Step 1.4, calculate the reference velocity direction θ of the following robot:

[0016] θ=atan((y c -y d ) / (x c -x d ))+π / 2

[0017] Step 1.5: Calculate the reference speed of the following robot in two axes

[0018]

[0019]

[0020] Furthermore, the posture deviation calculation method in step 2 is specifically as follows:

[0021] Assume that the actual position coordinates of the follower robot in the leader robot coordinate system are [x r y r ]; then the position deviation of the two axes is e x 、e y for:

[0022] e x =x d -x r

[0023] e y =y d-y r .

[0024] Furthermore, the calculation method of the proportional control coefficient in step 3 is specifically as follows:

[0025] Step 3.1, use the three-layer BP neural network parameter adjustment model, which are input layer, hidden layer and output layer; use Represents the neural network input matrix;

[0026] Step 3.2, use represents the weight matrix connecting the input layer and the hidden layer, and uses f(·) to represent the hidden layer activation function:

[0027]

[0028] use represents the weight matrix connecting the hidden layer and the output layer, and uses g(·) to represent the output layer activation function:

[0029]

[0030] Step 3.3, calculate the hidden layer input I (2) With output O (2) :

[0031]

[0032] O (2) =f(I (2) )

[0033] Step 3.4, calculate the output layer input I (3) With output O (3) :

[0034]

[0035] O (3) =g(I (3) )

[0036] Among them, (3) is the proportional control coefficient, then the following robot should have an x-axis center speed for:

[0037] u x =O (3) ·e x ;

[0038]

[0039] where u x Compute outputs for neural networks;

[0040] Step 3.5, use Represents the neural network input matrix, using the connection weight matrix Repeat steps 3.2 to 3.4 to obtain the center velocity of the y-axis of the following robot. for:

[0041] u y =O (3) ·e y

[0042]

[0043] where u y Compute the output for the neural network.

[0044] Furthermore, the method for correcting the connection weight matrix in step 4 is specifically as follows:

[0045] Step 4.1, calculate the symbolic function f sgn :

[0046]

[0047] in The x-axis coordinate of the actual position of the robot in the previous control cycle. Calculate the output of the neural network for the last control cycle;

[0048] Step 4.2, calculate the output layer activation function derivative g'(x) and the hidden layer activation function derivative f'(x):

[0049]

[0050]

[0051] Step 4.3, update the weight matrix connecting the hidden layer and the output layer. The new matrix uses It means that the calculation method is:

[0052] Δ=e x ·e x ·f sgn ·g'(I (3) )

[0053]

[0054] in is the weight matrix connecting the hidden layer and the output layer in the previous cycle, and Δ is the calculation process quantity;

[0055] Step 4.4, update the weight matrix connecting the input layer and the hidden layer. The new matrix uses It means that the calculation method is:

[0056]

[0057] in The weight matrix connecting the input layer and the hidden layer in the previous cycle;

[0058] Step 4.5, calculate the symbolic function f sgn :

[0059]

[0060] in The y-axis coordinate of the actual position of the robot in the previous control cycle, Calculate the output of the neural network for the previous control cycle; replace the corresponding x-axis related variables with the y-axis, repeat steps 4.2 to 4.4, and calculate the connection weight matrix between the hidden layer and the output layer The weight matrix connecting the input layer and the hidden layer

[0061] Beneficial effects:

[0062] The present invention utilizes the arbitrary nonlinear system expression capability of a neural network and realizes the calculation and output of optimal control parameters by learning the system performance during operation.

[0063] The present invention can greatly reduce the debugging time. The same initial value of the weight matrix can be used at different operating speeds of the pilot robot, eliminating the step of manually modifying parameters, which is more in line with actual engineering application scenarios.

[0064] The speed control parameter output of the present invention is based on the learning of the formation system by a neural network, and the weight matrix is ​​continuously updated and iterated according to the formation operation effect. Compared with the traditional fixed-value PID control, a better formation control effect can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a schematic diagram of the overall flow of the adaptive control coefficient calculation method of this application. DETAILED DESCRIPTION

[0066] The present invention will be described in further detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention. Based on the present invention, other embodiments obtained by ordinary technicians in the field without making any creative work all fall within the scope of protection of the present invention.

[0067] like Figure 1 As shown, the present application provides a method for calculating an adaptive control coefficient, the method comprising the following steps:

[0068] Step 1: Calculate the reference speed and reference position of the following robot according to the running speed and formation constraints of the leading robot;

[0069] Step 2, calculating the position deviation according to the reference position and the actual position;

[0070] Step 3: Use a three-layer neural network to calculate the speed proportional control coefficient and calculate the center speed of the following robot in combination with the reference speed;

[0071] Step 4: Correct the neural network connection weight matrix.

[0072] Furthermore, the calculation method of the reference speed in step 1 is specifically as follows:

[0073] Step 1.1, the formation constraint is to design a formation. Under the formation constraint, the reference position of the follower robot is the position coordinate of the origin of the follower robot coordinate system in the leader robot coordinate system, using [x d y d ]express;

[0074] Step 1.2, the speed of the pilot robot uses the x-speed v x , y to v y , the vehicle angular velocity ω and the rotation center [x c y c ] four parameters are represented; according to the reference position of the following robot, the rotation radius R of the following robot is obtained s for:

[0075]

[0076] Step 1.3, calculate the reference speed v of the following robot d :

[0077] v d =ω·R s

[0078] Step 1.4, calculate the reference velocity direction θ of the following robot:

[0079] θ=atan((y c -y d ) / (x c - xd ))+π / 2

[0080] Step 1.5: Calculate the reference speed of the following robot in two axes

[0081]

[0082]

[0083] Furthermore, the posture deviation calculation method in step 2 is specifically as follows:

[0084] Assume that the actual position coordinates of the follower robot in the leader robot coordinate system are [x r y r ]; then the position deviation of the two axes is e x 、e y for:

[0085] e x =x d -x r

[0086] e y =y d -y r .

[0087] Furthermore, the calculation method of the proportional control coefficient in step 3 is specifically as follows:

[0088] Step 3.1, use the three-layer BP neural network parameter adjustment model, which are input layer, hidden layer and output layer; use Represents the neural network input matrix;

[0089] Step 3.2, use represents the weight matrix connecting the input layer and the hidden layer, and uses f(·) to represent the hidden layer activation function:

[0090]

[0091] use represents the weight matrix connecting the hidden layer and the output layer, and uses g(·) to represent the output layer activation function:

[0092]

[0093] Step 3.3, calculate the hidden layer input I (2) With output O (2) :

[0094]

[0095] O (2) =f(I (2) )

[0096] Step 3.4, calculate the output layer input I (3) With output O (3) :

[0097]

[0098] O (3) =g(I (3) )

[0099] Among them, (3)is the proportional control coefficient, then the following robot should have an x-axis center speed for:

[0100] u x =O (3) ·e x ;

[0101]

[0102] where u x Compute outputs for neural networks;

[0103] Step 3.5, use Represents the neural network input matrix, using the connection weight matrix Repeat steps 3.2 to 3.4 to obtain the center velocity of the y-axis of the following robot. for:

[0104] u y =O (3) ·e y

[0105]

[0106] where u y Compute the output for the neural network.

[0107] Furthermore, the method for correcting the connection weight matrix in step 4 is specifically as follows:

[0108] Step 4.1, calculate the symbolic function f sgn :

[0109]

[0110] in The x-axis coordinate of the actual position of the robot in the previous control cycle. Calculate the output of the neural network for the last control cycle;

[0111] Step 4.2, calculate the output layer activation function derivative g'(x) and the hidden layer activation function derivative f'(x):

[0112]

[0113]

[0114] Step 4.3, update the weight matrix connecting the hidden layer and the output layer. The new matrix uses It means that the calculation method is:

[0115] Δ=e x ·e x ·fsgn ·g'(I (3) )

[0116]

[0117] in is the weight matrix connecting the hidden layer and the output layer in the previous cycle, and Δ is the calculation process quantity;

[0118] Step 4.4, update the weight matrix connecting the input layer and the hidden layer. The new matrix uses It means that the calculation method is:

[0119]

[0120] in The weight matrix connecting the input layer and the hidden layer in the previous cycle;

[0121] Step 4.5, calculate the symbolic function f sgn :

[0122]

[0123] in The y-axis coordinate of the actual position of the robot in the previous control cycle, Calculate the output of the neural network for the previous control cycle; replace the corresponding x-axis related variables with the y-axis, repeat steps 4.2 to 4.4, and calculate the connection weight matrix between the hidden layer and the output layer The weight matrix connecting the input layer and the hidden layer

[0124] Take a two-car formation as an example. The formation constraint is that the following robot is 3 meters behind the leading robot. Then the reference position of the following robot [x d y d ] is [-3 0].

[0125] The pilot robot receives the command running speed v x =1.5m / s,v y =0.5m / s; rotation center [x c y c ] is [-1.265 3.795]; the angular velocity is ω=0.395. Correspondingly, the rotation radius R of the following robot s =4.1726, reference speed y d =1.649, reference speed direction θ=335°. The reference speeds of the two axes are

[0126] Assume that the actual position coordinates of the follower robot in the leader robot coordinate system are [x r yr ] is [-3.5 0.8], then the corresponding position deviations of the two axes are e x =0.5, e y =-0.8.

[0127] For the x-axis, the neural network input matrix O can be calculated from the above parameters. (1) =[1.494 0.5]; Value of the weight matrix connecting the input layer and the hidden layer:

[0128]

[0129] The weight matrix connecting the hidden layer and the output layer is:

[0130]

[0131] For the x-axis, the hidden layer input I is calculated (2) =[0.6988 0.4494].

[0132] For the x-axis, the hidden layer output O is calculated (2) =[0.6036 0.4214].

[0133] For the x-axis, the output layer input I is calculated (3) =0.5125.

[0134] For the x-axis, the output layer output O is calculated (3) =0.7359.

[0135] For the x-axis, the neural network calculates the output u x =0.368, then the following robot should have a center speed of x-axis

[0136] For the x-axis, update the weight matrix and calculate Δ=0.0972, then:

[0137]

[0138] For the y-axis, the neural network input matrix O can be calculated from the above parameters. (1) =[0.697 0.8]; Value of the weight matrix connecting the input layer and the hidden layer:

[0139]

[0140] The weight matrix connecting the hidden layer and the output layer is:

[0141]

[0142] For the y-axis, the hidden layer input I is calculated (2)=[1.339 1.009].

[0143] For the y-axis, the hidden layer output O is calculated (2) =[0.872 0.765].

[0144] For the y-axis, the output layer input I is calculated (3) =0.8397.

[0145] For the y-axis, the output layer output O is calculated (3) =0.8428.

[0146] For the y-axis, the neural network calculates the output u y =-0.6743, the following robot should have a y-axis center speed

[0147] For the y-axis, update the weight matrix and calculate Δ = 0.265, then:

[0148]

[0149] By repeating the above cycle, the connection weight matrix will be continuously updated according to the position deviation of the following unit and the reference speed, and the calculated output of the solved neural network will also be continuously adjusted to achieve the effect of self-adaptation and self-learning.

[0150] The method provided in this application is applicable to the adaptive control coefficient calculation method of multi-robot collaborative control. Mobile robots have been widely used in heavy industrial occasions, mostly for the handling of large-size and large-mass workpieces; and the method of using multiple mobile robots to form a formation for collaborative transportation can further improve the flexibility of mobile robot transportation and the adaptability of multiple products. In order to obtain appropriate speed control parameters and achieve a high-precision formation control effect, a long period of parameter debugging and exploration is required for each formation member at different operating speeds. Considering that robots are complex systems, and the formation system composed of multiple groups of mobile robots is more difficult to abstract an accurate mathematical model, the present invention introduces a neural network into the speed control coefficient calculation module of the follower robot, and proposes a control coefficient calculation method based on a neural network. This method enables the speed control parameters to adapt to changes in parameters such as the formation running speed and formation position, greatly reducing the time for manual parameter adjustment and improving the formation control accuracy, so that the formation collaborative transportation is more in line with the actual application scenario of the project. The application scenario of this patent is multi-robot collaborative transportation, where multiple mobile robots form a formation and maintain the formation during operation. The calculation output is the speed control coefficient of the robot member, which can be used to adjust the robot running speed so that the formation formation is well maintained. The control parameters are used to control the speed of the robots during formation operation.

[0151] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the scope of improvement of the present invention.

Claims

1. A method for calculating an adaptive control coefficient, characterized in that: The method comprises the following steps: Step 1: Calculate the reference speed and reference position of the following robot according to the running speed and formation constraints of the leading robot; Step 2, calculating the position deviation according to the reference position and the actual position; Step 3: Use a three-layer neural network to calculate the speed proportional control coefficient and calculate the center speed of the following robot in combination with the reference speed; Step 4: Correct the neural network connection weight matrix. The method for correcting the connection weight matrix in step 4 is as follows: Step 4.1, calculate the symbolic function f sgn : in The x-axis coordinate of the actual position of the robot in the previous control cycle. Calculate the output of the neural network for the last control cycle; Step 4.2, calculate the output layer activation function derivative g'(x) and the hidden layer activation function derivative f'(x): Step 4.3, update the weight matrix connecting the hidden layer and the output layer. The new matrix uses It means that the calculation method is: △=e x ·e x ·f sgn ·g‘(I (3) ) in is the weight matrix connecting the hidden layer and the output layer in the previous cycle, and Δ is the calculation process quantity; Step 4.4, update the weight matrix connecting the input layer and the hidden layer. The new matrix uses It means that the calculation method is: in The weight matrix connecting the input layer and the hidden layer in the previous cycle; Step 4.5, calculate the symbolic function f sgn : in The y-axis coordinate of the actual position of the robot in the previous control cycle. Calculate the output of the neural network for the previous control cycle; replace the corresponding x-axis related variables with the y-axis, repeat steps 4.2 to 4.4, and calculate the connection weight matrix between the hidden layer and the output layer The weight matrix connecting the input layer and the hidden layer 2. The method for calculating an adaptive control coefficient according to claim 1, characterized in that: The calculation method of the reference speed in step 1 is as follows: Step 1.1, the formation constraint is to design a formation. Under the formation constraint, the reference position of the follower robot is the position coordinate of the origin of the follower robot coordinate system in the leader robot coordinate system, using [x d y d ]express; Step 1.2, the speed of the pilot robot is the x-speed υ x ,y to v y , the vehicle angular velocity ω and the rotation center [x c y c ] four parameters are represented; according to the reference position of the following robot, the rotation radius R of the following robot is obtained s for: Step 1.3, calculate the reference speed υ of the following robot d : u d =ω·R s Step 1.4, calculate the reference velocity direction θ of the following robot: θ=atan((y c -y d ) / (x c -x d ))+π / 2 Step 1.5: Calculate the reference speed of the following robot in two axes 3. The method for calculating an adaptive control coefficient according to claim 1, characterized in that: The specific method for calculating the posture deviation in step 2 is: Assume that the actual position coordinates of the follower robot in the leader robot coordinate system are [x r y r ]; then the position deviation of the two axes is e x 、e y for: e x =x d -x r and y =and d -and r 。 4. The method for calculating an adaptive control coefficient according to claim 1, characterized in that: The specific calculation method of the proportional control coefficient in step 3 is: Step 3.1, use the three-layer BP neural network parameter adjustment model, which are input layer, hidden layer and output layer; use Represents the neural network input matrix; Step 3.2, use represents the weight matrix connecting the input layer and the hidden layer, and f(·) represents the hidden layer activation function: use represents the weight matrix connecting the hidden layer and the output layer, and uses g(·) to represent the output layer activation function: Step 3.3, calculate the hidden layer input I (2) With output O (2) : O (2) =f(I (2) ) Step 3.4, calculate the output layer input I (3) With output O (3) : O (3) =g(I (3) ) Among them, (3) is the proportional control coefficient, then the following robot should have an x-axis center speed for: u x =The (3) ·and x ; where u x Compute outputs for neural networks; Step 3.5, use Represents the neural network input matrix, using the connection weight matrix Repeat steps 3.2 to 3.4 to obtain the center velocity of the y-axis of the following robot. for: u y =The (3) ·and y where u y Compute the output for the neural network.

Citation Information

Patent Citations

  • Mobile robot formation control method based on leader-follow

    CN105527960A

  • Missile-borne electric steering engine rudder deflection angle position tracking method based on BP-PID neural network

    CN113504721A

  • Coded data identification method and system for air formation

    CN115061501A

  • A velocity type neural controller for speed control of a digital speed regulator

    KR1020090027096A

  • Vehicle speed tracking control method and apparatus, and device and storage medium

    WO2022227720A1