CN101122777A - Large condenser underwater operation environment two-joint robot control method - Google Patents
Large condenser underwater operation environment two-joint robot control method Download PDFInfo
- Publication number
- CN101122777A CN101122777A CNA2007100357659A CN200710035765A CN101122777A CN 101122777 A CN101122777 A CN 101122777A CN A2007100357659 A CNA2007100357659 A CN A2007100357659A CN 200710035765 A CN200710035765 A CN 200710035765A CN 101122777 A CN101122777 A CN 101122777A
- Authority
- CN
- China
- Prior art keywords
- neural network
- fuzzy
- robot
- network
- joints
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Abstract
The utility model discloses a control method for controlling a two-joint robot of an underwater-operation condenser. Processes are as following: {1} A controller structure of a fuzzy Kovski neural network is built up and each sub-network represents a serve controller of a joint; {2} Every parameter of the control system and weighted value of network is initialized; an error of an angular displacement of two joints and rate of error is obtained based on the current angular displacement of two joints and expected motion angle displacement; {3}The error and the rate of error are used as the input of the fuzzy Kovski neural network controller and obtain moment exerted on motor points of two joints by fuzzy Kovski neural network controller. Tthe moment is used to control an electric motor, making the joints reach expected elements of a fix. The invention can overcome the difficulty in self-modeling of the robots and uncertainty of environmental disturbance. The utility model can not only sum up experience in controlling of human experts but also have self-study function of neural network controller, thereby reaching a higher controlling degree of accuracy.
Description
Technical Field
The invention mainly relates to the field of control of a two-joint robot, in particular to a control method of the two-joint robot under a large condenser underwater operation environment.
Background
The real development work of the robot started in the 40 th 20 th century, and the first industrial robot came out in 1959 in united states department of america, which marked the beginning of the robot entering industrial applications. Up to now, about 80% of the robots used in practice are located in the industrial field, belonging to industrial robots, which have: the robot has the characteristics of simple structure, known working environment, preset working of the robot and the like, so that the control of the robot is determined to belong to a teaching and reproducing type, the working path and the motion parameters of the robot need to be taught by an operator through a hand handle or realized through on-line programming, and the operation demonstrated by the operator can be basically repeated. With the development of sensor technology, computer technology, network and other technologies, the robot has an external sensor, has certain sensing capability on a working object and an external environment, and forms closed-loop control by using the sensed information as feedback. At present, the robot not only has the sensing capability to the outside, but also can carry out complex logic thinking and decision, and complete intelligent operation. Therefore, as the robot technology is developed, the robot control technology is also rapidly developed.
The problem with controlling a robot is to enable the position of each joint or end effector of the robot to track a given trajectory or settle at a specified position with a desired dynamic quality. Two difficulties with this problem exist: firstly, how to realize the stability of a closed-loop error system and make the track tracking error approach to zero as soon as possible; another is how to suppress the interference, reducing as much as possible the effect of the interfering signal on tracking in.
As a control object, the robot itself is a time-varying, strongly coupled multiple-input multiple-output nonlinear system, and its control is very complicated. The intelligent two-joint robot control method in the large condenser underwater operation environment provided by the patent is to take the robot of the patent as a control object and study the control method in the operation process on the basis of the patents of a condenser copper pipe two-joint type online cleaning robot (publication number: CN 1945196) and an underwater operation robot for online cleaning a heat exchanger (grant number: CN 2631712Y). Because the robot works in an underwater environment, signal measurement and system modeling are inaccurate, and complicated environment changes and external uncertain disturbance influence caused by water flow (speed, strength), water temperature, water pressure and the like cannot be added, a precise and complete motion model of the robot cannot be obtained, so that the control precision of the robot cannot be ensured or is difficult to ensure by a conventional control method and a modern control method, and the problems that the robot is too complicated or cannot ensure the control instantaneity of the robot and the like exist. Therefore, on the premise of meeting the control precision of the robot, the design of the intelligent two-joint robot control method which is simple, reliable, good in real-time performance, convenient to realize and capable of working underwater is a key technical problem for ensuring the normal work of the robot.
Disclosure of Invention
The invention aims to solve the problems that: aiming at the technical problems in the prior art, the invention provides a control method of a two-joint robot under the large-scale condenser underwater operation environment, which is based on a fuzzy Gaussian neural network control method, can overcome the influences caused by the difficulty in modeling of the robot per se, uncertainty of environmental disturbance and the like, can synthesize the control experience of human experts, and has the self-learning function of a neural network controller, so that the control method of the two-joint robot under the large-scale condenser underwater operation environment with higher control precision is achieved.
In order to solve the technical problems, the solution provided by the invention is as follows: a control method of a two-joint robot in a large condenser underwater operation environment is characterized by comprising the following steps:
(1) Establishing a fuzzy Gaussian neural network controller structure, dividing the whole network into two subnets, wherein each subnet represents a servo controller of a joint, and considering the coupling effect between the joints, the output u of the two subnets 1 And u 2 Multiplying the output y of the whole network by corresponding influence factors respectively 1 And y 2 ;
(2) Initializing all parameters and network weights of a control system, taking a control period T value, and acquiring angular displacement (theta) of two current joints of the robot through a shaft encoder 1 、θ 2 ) And calculating the expected motion angular displacement (theta d) of the two joints through the expected positioning coordinates 1 、θd 2 ) Thereby obtaining the error of the angular displacement of the two joints and the error change rate e 1 =θd 1 -θ 1 、 e 2 =θd 2 -θ 2 、
(3) Will (e) 1 、e 2 、ec 1 、ec 2 ) As the input of the fuzzy Gaussian neural network controller, the moment (t) applied to the two joint motion points is obtained by the fuzzy Gaussian neural network controller 1 、t 2 ) Wherein (t) 1 、t 2 ) Satisfies the relation: t is t 1 =y 1 ,t 2 =y 2 According to the moment (t) 1 、t 2 ) And controlling the motors of the two joints on the two-joint robot to move to reach the expected positioning coordinates.
The fuzzy Gaussian neural network controller in the step (1) is sequentially divided into 5 layers in the following formulas k in j (i) 、 k out j (i) The input and the output of the j-th neuron of the i-th network of the k-th sub-network are respectively represented as follows:
(1) calculating input; layer 1 will input the error and the rate of change of the error (e) 1 、e 2 、ec 1 、ec 2 ) Introducing network, and each input universe of discourse is [ -1,1 [ -1];
Wherein k x i An ith input representing a kth subnet;
(2) fuzzification; fuzzifying the input at the layer 2, wherein 3 fuzzy language word sets { N, Z, P } = { "negative", "zero", "positive" } are corresponding to each input, the membership function adopts a Gaussian basis function, the central values corresponding to { N, Z, P } are { -1,0,1} respectively, the widths are {0.5,0.5 },
in the formula (I), the compound is shown in the specification, k A i j representing the j language word set corresponding to the ith input in the k subnet, k a ij 、 k b ij are respectively as k A i j The center value and the width of (c);
(3) cross multiplication; layer 3 represents "and" operation, where multiplication is used instead of small operations,
(4) defuzzification; layer 4 represents the defuzzification process, where a weight average decision method is used,
in the formula (I), the compound has the following structure, k ω ij (3) the physical meaning of the weight value of the network is the central value of the language word set corresponding to the output of each control rule;
(5) coupling treatment; layer 5 represents the coupling between the joints,
wherein, ω is kl (4) The network weight reflects the coupling between the joints.
In the step (2), the fuzzy Gaussian neural network controller completes the off-line learning, and the steps are as follows:
(1) initializing each weight omega of the network by adopting random function kl (4) 、 k ω ij (3) And parameters of Gaussian function k a ij 、 k b ij ;
(3) Computing network output y by a fuzzy Gaussian neural network 1 And y 2 ;
(4) Calculating J according to the following formula (7) off ;
In the formula (I), the compound is shown in the specification,for the sample output, i.e. the desired output of the network, y l Is actually output by the network;
(5) if J is off If the weight is less than or equal to 0.01, the off-line learning is finished to obtain the weight omega kl (4) 、 k ω ij (3) And parameters of Gaussian function k a ij 、 k b ij ;
(6) If J is off If the value is greater than 0.01, the parameter ω is calculated and updated by the following equations (7) to (10) kl (4) 、 k ω ij (3) And k a ij 、 k b ij and returning to the step (3) to enter the calculation process until the weight omega meeting the requirement is obtained kl (4) 、 k ω ij (3) Parameters of the sum Gaussian function k a ij 、 k b ij :
Wherein eta 1 、η 2 、η 3 And η 4 Is the learning rate.
After the fuzzy Gaussian neural network controller completes offline learning, online learning can be carried out, and in order to ensure the real-time performance of the control system, only the parameter omega is subjected to online learning kl (4) And k ω ij (3) the method carries out 'online learning' and comprises the following steps:
(1) parameter omega obtained by off-line learning method kl (4) 、 k ω ij (3) And k a ij 、 k b ij assigning to a fuzzy Gaussian neural network;
(2) according to the control input of the actual operation of the robot, the action of the robot is controlled through a fuzzy Gaussian neural network to obtain the actual output theta of the robot 1 、θ 2 ;
(3) Desired output θ according to reference d1 、θ d2 And theta 1 、θ 2 Calculating J by using the following formula (12) on
(4) If J is on If the weight is less than or equal to 0.005, the on-line learning is ended to obtain the weight omega kl (4) 、 k ω ij (3) ;
(5) If J is on If the value is greater than 0.005, the parameter ω is calculated and updated by the following equations (12) to (17) kl (4) 、 k ω ij (3) And returning to the step (2) to enter a calculation process until a weight omega meeting the requirement is obtained kl (4) 、 k ω ij (3) :
k=1,2;i=1,2,3;j=1,2,3;l=1,2(16)
Wherein eta is 1 、η 2 、η 3 And η 4 For the learning rate, ε is a very small positive number.
Compared with the prior art, the invention has the advantages that:
1. the invention adopts the fuzzy Gaussian neural network control method, combines the neural network with the fuzzy control, can better overcome the respective defects of the neural network and the fuzzy control, not only can lead the fuzzy control to have self-learning capability, but also can lead the neural network to have reasoning and induction capability, and simultaneously can lead the structure and the weight of the network to have definite physical significance;
2. compared with the traditional robot control method, the method has the function of learning and setting the on-line parameters of the controller according to the change of the environment under the underwater severe environment, so that the control of the robot has strong robustness;
3. the method solves the problem of control of the two-joint robot due to uncertainty of external parameters and unknown interference, and can realize accurate operation control of underwater environment.
Drawings
FIG. 1 is a schematic structural diagram of a two-joint robot control system based on a fuzzy Gaussian neural network;
FIG. 2 is a schematic diagram of a fuzzy Gaussian neural network controller;
FIG. 3 is a schematic diagram of the initial shape and distribution of membership functions;
fig. 4 is a schematic diagram of a trajectory tracking control curve of the joint 1;
fig. 5 is a schematic view of a tracking error curve of the joint 1;
FIG. 6 is a schematic diagram of a trajectory tracking control curve for joint 2;
FIG. 7 is a schematic view of a tracking error curve for the joint 2;
FIG. 8 is a schematic flow chart of the operation of a two-joint robot control system of a fuzzy Gaussian neural network;
FIG. 9 is a flow diagram of a fuzzy Gaussian neural network input-output process;
FIG. 10 is a schematic flow diagram of an offline learning algorithm;
FIG. 11 is a flow diagram of an online learning algorithm.
Detailed Description
The invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.
As shown in fig. 8, the method for controlling a two-joint robot in a large condenser underwater operation environment of the present invention comprises the steps of:
(1) Establishing a fuzzy Gaussian neural network controller structure, dividing the whole network into two subnetworks, wherein each subnet represents a servo controller of a joint, and considering the coupling effect between the joints, the output u of the two subnetworks 1 And u 2 Multiplying the output y of the whole network by corresponding influence factors respectively 1 And y 2 ;
(2) Initializing all parameters and network weight of a control system, taking a control period T value, and acquiring angular displacement (theta) of two current joints of the robot through a shaft encoder 1 、θ 2 ) And calculating the expected angular displacement (theta d) of the two joints according to the expected positioning coordinates 1 、θd 2 ) Thereby obtaining the error of the angular displacement of the two joints and the error change rate e 1 =θ d1 -θ 1 、 e 2 =θ d2 -θ 2 、 ;
(3) Will (e) 1 、e 2 、ec 1 、ec 2 ) As the input of the fuzzy Gaussian neural network controller, the moment (t) applied to the two joint motion points is obtained by the fuzzy Gaussian neural network controller 1 、t 2 ) Wherein (t) 1 、t 2 ) Satisfies the relation: t is t 1 =y 1 ,t 2 =y 2 According to the moment (t) 1 、t 2 ) And controlling the motors of the two joints on the two-joint robot to move to reach the expected positioning coordinates.
As shown in fig. 1, 2 and 9, the fuzzy gaussian neural network controller is sequentially divided into 5 layers in the above step (1), in the following formulas k in j (i) 、 k out j (i) Respectively represents the input and the output of the j-th neuron of the i-th network of the k-th sub-network:
(1) calculating input; layer 1 will input the error and the rate of change of the error (e) 1 、e 2 、ec 1 、ec 2 ) Is introduced into the network, each timeAll the input domains are [ -1,1 [ ]];
Wherein k x i An ith input representing a kth subnet;
(2) fuzzification; fuzzifying the input at the layer 2, wherein 3 fuzzy language word sets { N, Z, P } = { "negative", "zero", "positive" } are corresponding to each input, the membership function adopts a Gaussian basis function, the central values corresponding to { N, Z, P } are { -1,0,1} respectively, the widths are {0.5,0.5 },
in the formula (I), the compound is shown in the specification, k A i j represents the jth language word set corresponding to the ith input in the kth sub-network, k a ij 、 k b ij are respectively as k A ij The center value and the width of (c);
(3) cross multiplication; layer 3 represents "and" operation, where multiplication is used instead of small operations,
(4) defuzzification; layer 4 represents the defuzzification process, where a weight average decision method is used,
in the formula (I), the compound is shown in the specification, k ω ij (3) the physical meaning of the weight value of the network is the central value of the language word set corresponding to the output of each control rule;
(5) coupling treatment; layer 5 represents the coupling between the joints,
wherein, ω is kl (4) The network weight reflects the coupling between the joints.
In FIG. 1, θ d1 、θ d2 Is the desired position of the two joints, θ 1 、θ 2 Is the actual position of the two joints, e 1 、e 2 Is a positional error of the two joints, e 1 、e 2 Obtaining error change rate ec after differentiation 1 、ec 2 。t 1 、t 2 To act on two jointsThe torque of (1). In fig. 1, a fuzzy gaussian neural network is used as a joint servo controller. The following symbols exist in the following relationsComprises the following steps: 1 x 1 =e 1 , 1 x 2 =ec 1 , 2 x 1 =e 2 , 2 x 2 =ec 2 ,t 1 =y 1 ,t 2 =y 2 。
as shown in fig. 9, in the step (2), the fuzzy gaussian neural network controller completes the "off-line learning", which includes the following steps:
(1) initializing each weight omega of the network by adopting random function kl (4) 、 k ω ij (3) And parameters of Gaussian function k a ij 、 k b ij ;
(3) Computing network output y by fuzzy Gaussian neural network 1 And y 2 ;
(4) Calculating J according to the following formula (7) off ;
In the formula (I), the compound is shown in the specification,for the sample output, i.e. the desired output of the network, y l Is actually output for the network;
(5) if J off If the weight is less than or equal to 0.01, the off-line learning is finished to obtain the weight omega kl (4) 、 k ω ij (3) And parameters of Gaussian function k a ij 、 k b ij ;
(6) If J off If the value is greater than 0.01, the parameter ω is calculated and updated by the following equations (8) to (11) kl (4) 、 k ω ij (3) And k a ij 、 k b ij and returning to the step (3) to enter the calculation process until the weight omega meeting the requirement is obtained kl (4) 、 k ω ij (3) Parameters of the sum Gaussian function k a ij 、 k b ij :
Wherein eta is 1 、η 2 、η 3 And η 4 As a learning rate
In a specific embodiment, the specific steps of "offline learning" are: in the off-line learning stage, firstly, selectingSet of samplesData (a) ) Then training the network with the sample data, the training value including omega kl (4) 、 k ω ij (3) And parameters of Gaussian functions k a ij 、 k b ij . The calculation is performed by iterating equations (8) to (11).
The training objective function is:
in the formula (I), the compound is shown in the specification,for the sample output, i.e. the network expected output, y l Is the actual output of the network. Then there are:
as shown in fig. 10, after the fuzzy gaussian neural network controller completes the "offline learning", the "online learning" can be performed, which includes the following steps:
(1) parameter omega obtained by off-line learning method kl (4) 、 k ω ij (3) And k a ij 、 k b ij assigning to a fuzzy Gaussian neural network;
(2) according to the control input of the actual operation of the robot, the action of the robot is controlled by the fuzzy Gaussian neural network to obtain the actual output theta of the robot 1 、θ 2 ;
(3) Desired output θ according to reference d1 、θ d2 And theta 1 、θ 2 Using the following formula (1)8) Calculating J on
(4) If J is on If the weight is less than or equal to 0.005, the on-line learning is ended to obtain the weight omega kl (4) 、 k ω ij (3) ;
(5) If J is on If it is greater than 0.005, the parameter ω is calculated and updated by the following equations (19) to (23) kl (4) 、 k ω ij (3) And returning to the step (2) to enter a calculation process until a weight omega meeting the requirement is obtained kl (4) 、 k ω ij (3) :
Wherein eta 1 、η 2 、η 3 And η 4 For the learning rate, ε is a very small positive number.
The operation of the present invention will be described in detail with reference to a specific application example. The control method of the intelligent two-joint robot is applied to the control of the underwater condenser cleaning robot, and the specific parameters are as follows: m is 1 =10kg,m 2 =2kg and l 1 =1.1m,l 2 =0.8m. Initial condition is θ 1 (0)=0rad,θ 2 (0)=1rad, The desired trajectory is θ d1 (t)=sin(2πt),θ d2 (T) = cos (2 π T), sampling period T is 0.0005s. Suppose that the friction term and the disturbance term are respectively Weight of network k ω ij (3) ,ω ij (4) The initial values of (a) are shown in tables 1, 2 and 3, respectively.
TABLE 1 weight values 1 ω ij (3) Initial value of
1 ω 11 (3) | 1 ω 12 (3) | 1 ω 13 (3) | 1 ω 21 (3) | 1 ω 22 (3) | 1 ω 23 (3) | 1 ω 31 (3) | 1 ω 32 (3) | 1 ω 33 (3) |
-1 | -1 | 0 | -1 | 0 | 1 | 0 | 1 | 1 |
TABLE 2 weight values 2 ω ij (3) Initial value of
2 ω 11 (3) | 2 ω 12 (3) | 2 ω 13 (3) | 2 ω 21 (3) | 2 ω 22 (3) | 2 ω 23 (3) | 2 ω 31 (3) | 2 ω 32 (3) | 2 ω 33 (3) |
-1 | -1 | 0 | -1 | 0 | 1 | 0 | 1 | 1 |
TABLE 3 weight values ω ij (4) Initial value of
ω 11 (4) | ω 12 (4) | ω 21 (4) | ω 22 (4) |
1 | 0 | 0 | 1 |
As shown in fig. 4 to 7, the results of the proposed fuzzy gaussian neural network control method are given and compared with the conventional fuzzy control results. Fig. 4 shows the trajectory tracking curve of the joint 1, in the three curves of the figure — the curve representing the angular displacement expectation of the joint 1; represents the curve of the angular displacement variation of the joint 1 when a fuzzy gaussian controller is used; the method comprises the following steps of (1) of (8230); 823030indicates the curve of angular displacement change of the joint 1 when a traditional fuzzy logic controller is adopted; as can be seen from the figure, the fuzzy gaussian based controller is better able to track the desired curve to the maximum extent of angular displacement of the joint 1, and therefore is superior in performance to the conventional fuzzy logic controller.
As shown in fig. 5, a tracking error curve of the joint 1 is given, of which two curves, an error curve representing the angular displacement of the joint 1 when a fuzzy gaussian based controller is used is represented; represents an error curve of the angular displacement of the joint 1 when a conventional fuzzy logic controller is used; as can be seen from the figure, with the fuzzy gaussian controller, the error is small and can be made to approach zero value quickly.
As shown in fig. 6, trajectory tracking curves of the joint 2 are given, among the three curves in the figure — a curve representing the angular displacement expectation of the joint 2; represents the angular displacement variation curve of the joint 2 when the fuzzy Gaussian controller is adopted; the change in angular displacement of the joint 2 is expressed by a conventional fuzzy logic controller; as can be seen, the fuzzy gaussian based controller better tracks the desired curve for maximum angular displacement of the joint 2 and therefore performs better than conventional fuzzy logic controllers.
As shown in fig. 7, a tracking error curve of the joint 2 is given, of which two curves — an error curve representing the angular displacement of the joint 2 when a fuzzy gaussian based controller is used is represented; an error curve representing the angular displacement of the joint 2 when using a conventional fuzzy logic controller; as can be seen from the figure, with a fuzzy gaussian based controller, the error is small and can be made to approach zero very quickly. As can be seen from the control result, the fuzzy Gaussian neural network can be well used for controlling the robot, and the performance of the fuzzy Gaussian neural network is greatly improved compared with that of a fuzzy controller.
Claims (4)
1. A control method of a two-joint robot in a large condenser underwater operation environment is characterized by comprising the following steps:
(1) Establishing a fuzzy Gaussian neural network controller structure, dividing the whole network into two subnetworks, wherein each subnet represents a servo controller of a joint, and considering the coupling effect between the joints, the output u of the two subnetworks 1 And u 2 Multiplying the output y of the whole network by corresponding influence factors respectively 1 And y 2 ;
(2) Initializing all parameters and network weights of a control system, taking a control period T value, and acquiring angular displacement (theta) of two current joints of the robot through a shaft encoder 1 、θ 2 ) And calculating the expected motion angular displacement (theta) of the two joints through the expected positioning coordinates d1 、θ d2 ) Thereby obtaining the angular position of the two jointsError of shift and error rate of change e 1 =θ d1 -θ 1 、 e 2 =θ d2 -θ 2 、
(3) Will (e) 1 、e 2 、ec 1 、ec 2 ) As the input of the fuzzy Gaussian neural network controller, the moment (t) applied to the two joint motion points is obtained by the fuzzy Gaussian neural network controller 1 、t 2 ) Wherein (t) 1 、t 2 ) Satisfies the relation: t is t 1 =y 1 ,t 2 =y 2 According to the moment (t) 1 、t 2 ) And controlling the motors of the two joints on the two-joint robot to move to reach the expected positioning coordinates.
2. The method for controlling a two-joint robot in a large condenser underwater operation environment according to claim 1, wherein the fuzzy Gaussian neural network controller in the step (1) is sequentially divided into 5 layers in the following formulas k in j (i) 、 k out j (i) The input and the output of the j-th neuron of the i-th network of the k-th sub-network are respectively represented as follows:
(1) calculating input; layer 1 will input the error and the rate of change of the error (e) 1 、e 2 、ec 1 、ec 2 ) Introducing network, and each input universe of discourse is [ -1,1 [ -1];
Wherein k x i An ith input representing a kth subnet;
(2) fuzzification; fuzzifying the input at the layer 2, wherein 3 fuzzy language word sets { N, Z, P } = { "negative", "zero", "positive" } are corresponding to each input, the membership function adopts a Gaussian basis function, the central values corresponding to { N, Z, P } are { -1,0,1} respectively, the widths are {0.5,0.5 },
in the formula (I), the compound is shown in the specification, k A i j representing the j language word set corresponding to the ith input in the k subnet, k a ij 、 k b ij are respectively as k A i j The center value and the width of (c);
(3) cross multiplication; layer 3 represents "and" operation, where multiplication is used instead of small operations,
(4) defuzzification; layer 4 represents the defuzzification process, where a weight average decision method is used,
in the formula (I), the compound is shown in the specification, k ω ij (3) the physical meaning of the weight value of the network is the central value of the language word set corresponding to the output of each control rule;
(5) coupling treatment; layer 5 represents the coupling between the joints,
wherein, ω is kl (4) The network weight reflects the coupling between the joints.
3. The method for controlling a two-joint robot in a large condenser underwater operation environment according to claim 1 or 2, wherein in the step (2), the fuzzy Gaussian neural network controller is enabled to complete off-line learning, and the method comprises the following steps:
(1) initializing each weight omega of network kl (4) 、 k ω ij (3) And parameters of Gaussian function k a ij 、 k b ij ;
(3) Computing network output y by a fuzzy Gaussian neural network 1 And y 2 ;
(4) Calculating J according to the following formula (7) off ;
In the formula (I), the compound is shown in the specification,for the sample output, i.e. the desired output of the network, y l Is actually output by the network;
(5) if J off If the weight is less than or equal to 0.01, the off-line learning is finished to obtain the weight omega kl (4) 、 k ω ij (3) And parameters of Gaussian function k a ij 、 k b ij ;
(6) If J is off If the value is greater than 0.01, the parameter ω is calculated and updated by the following equations (8) to (11) kl (4) 、 k ω ij (3) And k a ij 、 k b ij and returning to the step (3) to enter the calculation flow,until a weight omega meeting the requirements is obtained kl (4) 、 k ω ij (3) Parameters of the sum Gaussian function k a ij 、 k b ij :
Wherein eta 1 、η 2 、η 3 And η 4 Is the learning rate.
4. The control method of the two-joint robot in the large condenser underwater operation environment according to claim 3, characterized in that: after the fuzzy Gaussian neural network controller completes off-line learning, the online learning can be carried out, and only the parameter omega is subjected to the real-time performance of the control system kl (4) And k ω ij (3) the method carries out 'on-line learning' and comprises the following steps:
(1) parameter omega obtained by off-line learning method kl (4) 、 k ω ij (3) And k a ij 、 k b ij assigning a value to the fuzzy Gaussian neural network;
(2) according to the control input of the actual operation of the robot, the action of the robot is controlled by the fuzzy Gaussian neural network to obtain the actual output theta of the robot 1 、θ 2 ;
(3) Desired output θ according to reference d1 、θ d2 And theta 1 、θ 2 By the following formulaEquation (12), calculation of J on
(4) If J is on If the weight is less than or equal to 0.005, the on-line learning is ended to obtain the weight omega kl (4) 、 k ω ij (3) ;
(5) If J is on If it is greater than 0.005, the parameter ω is calculated and updated by the following equations (12) to (17) kl (4) 、 k ω ij (3) And returning to the step (2) to enter a calculation process until a weight omega meeting the requirement is obtained kl (4) 、 k ω ij (3) :
Wherein eta is 1 、η 2 、η 3 And η 4 For the learning rate, ε is a very small positive number.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CNB2007100357659A CN100498600C (en) | 2007-09-18 | 2007-09-18 | Large condenser underwater operation environment two-joint robot control method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CNB2007100357659A CN100498600C (en) | 2007-09-18 | 2007-09-18 | Large condenser underwater operation environment two-joint robot control method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN101122777A true CN101122777A (en) | 2008-02-13 |
CN100498600C CN100498600C (en) | 2009-06-10 |
Family
ID=39085134
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CNB2007100357659A Expired - Fee Related CN100498600C (en) | 2007-09-18 | 2007-09-18 | Large condenser underwater operation environment two-joint robot control method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN100498600C (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101833333B (en) * | 2009-12-23 | 2011-09-14 | 湖南大学 | Unknown environment route planning method of underwater cleaning robot |
CN101551661B (en) * | 2009-05-12 | 2013-04-24 | 广东工业大学 | Control method for multi-robot system |
CN103279039A (en) * | 2013-05-17 | 2013-09-04 | 安徽工业大学 | Robot neural network type computed torque controller training platform and training method |
CN103927583B (en) * | 2014-04-28 | 2017-06-20 | 莫小钧 | Artificial neural network |
CN107972036A (en) * | 2017-12-25 | 2018-05-01 | 厦门大学嘉庚学院 | Industrial robot kinetic control system and method based on TensorFlow |
CN109202861A (en) * | 2017-06-30 | 2019-01-15 | 沈阳新松机器人自动化股份有限公司 | A kind of medicament allotment robot integrated platform |
CN109623819A (en) * | 2018-12-25 | 2019-04-16 | 哈工大机器人集团股份有限公司 | The acquisition methods and device of a kind of robot harmonic drive joint actual torque value |
CN110109363A (en) * | 2019-05-28 | 2019-08-09 | 重庆理工大学 | A kind of Neural Network Adaptive Control method that wheeled mobile robot is formed into columns |
CN115648227A (en) * | 2022-12-27 | 2023-01-31 | 珞石(北京)科技有限公司 | Robot motion trajectory neural network fuzzy control optimization method |
-
2007
- 2007-09-18 CN CNB2007100357659A patent/CN100498600C/en not_active Expired - Fee Related
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101551661B (en) * | 2009-05-12 | 2013-04-24 | 广东工业大学 | Control method for multi-robot system |
CN101833333B (en) * | 2009-12-23 | 2011-09-14 | 湖南大学 | Unknown environment route planning method of underwater cleaning robot |
CN103279039A (en) * | 2013-05-17 | 2013-09-04 | 安徽工业大学 | Robot neural network type computed torque controller training platform and training method |
CN103927583B (en) * | 2014-04-28 | 2017-06-20 | 莫小钧 | Artificial neural network |
CN109202861A (en) * | 2017-06-30 | 2019-01-15 | 沈阳新松机器人自动化股份有限公司 | A kind of medicament allotment robot integrated platform |
CN107972036A (en) * | 2017-12-25 | 2018-05-01 | 厦门大学嘉庚学院 | Industrial robot kinetic control system and method based on TensorFlow |
CN107972036B (en) * | 2017-12-25 | 2021-04-27 | 厦门大学嘉庚学院 | Industrial robot dynamics control system and method based on TensorFlow |
CN109623819A (en) * | 2018-12-25 | 2019-04-16 | 哈工大机器人集团股份有限公司 | The acquisition methods and device of a kind of robot harmonic drive joint actual torque value |
CN110109363A (en) * | 2019-05-28 | 2019-08-09 | 重庆理工大学 | A kind of Neural Network Adaptive Control method that wheeled mobile robot is formed into columns |
CN115648227A (en) * | 2022-12-27 | 2023-01-31 | 珞石(北京)科技有限公司 | Robot motion trajectory neural network fuzzy control optimization method |
Also Published As
Publication number | Publication date |
---|---|
CN100498600C (en) | 2009-06-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101122777A (en) | Large condenser underwater operation environment two-joint robot control method | |
CN110238839B (en) | Multi-shaft-hole assembly control method for optimizing non-model robot by utilizing environment prediction | |
CN110450156B (en) | Optimal design method of self-adaptive fuzzy controller of multi-degree-of-freedom mechanical arm system | |
Yang et al. | Adaptive fuzzy finite-time coordination control for networked nonlinear bilateral teleoperation system | |
CN110450155B (en) | Optimal design method for controller of multi-degree-of-freedom mechanical arm system | |
CN103433924A (en) | High-accuracy position control method for serial robot | |
Li et al. | Fuzzy adaptive fault tolerant time-varying formation control for nonholonomic multirobot systems with range constraints | |
CN108717262B (en) | Special-shaped curved surface tracking method and system based on moment feature learning neural network | |
CN103399488B (en) | Multiple Model Control Method based on self study | |
CN113885548B (en) | Multi-quad-rotor unmanned helicopter output constraint state consistent game controller | |
CN114942593A (en) | Mechanical arm self-adaptive sliding mode control method based on disturbance observer compensation | |
Hu | Research on robot fuzzy neural network motion system based on artificial intelligence | |
Chen et al. | Neuroadaptive tracking control of affine nonlinear systems using echo state networks embedded with multiclustered structure and intrinsic plasticity | |
Qun | Intelligent control technology of agricultural greenhouse operation robot based on fuzzy PID path tracking algorithm. | |
CN109176529B (en) | Self-adaptive fuzzy control method for coordinated movement of space robot | |
CN113359626B (en) | Finite time hierarchical control method for multi-robot system | |
CN115344047A (en) | Robot switching type predictive control trajectory tracking method based on neural network model | |
CN114012733A (en) | Mechanical arm control method for scribing PC (personal computer) component mold | |
CN110355780A (en) | The output torque calculation method of integrated flexible joint of robot | |
CN108155836A (en) | Based on global online heuristic dynamic programming permanent magnet synchronous motor vector control method | |
CN116774576A (en) | Underwater vehicle dynamics black box modeling method based on neural network indirect estimation | |
Zhao et al. | Robotic peg-in-hole assembly based on reversible dynamic movement primitives and trajectory optimization | |
CN113848954A (en) | Improved unmanned vehicle servo control system of multi-neuron neural network PID algorithm | |
Jing et al. | Research on neural network PID adaptive control with industrial welding robot in multi-degree of freedom | |
CN112380655A (en) | Robot inverse kinematics solving method based on RS-CMSA algorithm |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20090610 Termination date: 20180918 |