Efficient bionic model for optimizing mechanical arm motion based on integrated learning operation

Abstract

The invention relates to the technical field of model design of mechanical arm motion control, and particularly discloses an efficient bionic model for optimizing mechanical arm motion based on integrated learning operation, and the bionic model estimates a fitness function through a mayfly naiad algorithm. And the fitness function is optimized to obtain initial duty ratios of various mechanical arm motion types in the mechanical arm motion type set, and a final duty ratio is determined from the initial duty ratios through multi-classifier ensemble learning, so that an optimal mechanical arm motion group in the mechanical arm motion type set is obtained. According to the bionic model, the motion control of the mechanical arm is optimized through the mayfly naiad algorithm and the integrated learning operation, so that the motion control of the mechanical arm has higher speed and higher precision.

Description

Technical Field

[0001] The invention relates to the technical field of model design for robot arm motion control. Background Art

[0002] The movement of the robot arm is highly dependent on the design and implementation of sensors and drive devices, as well as the duty cycle of these devices. Duty Ratio refers to the ratio of the time occupied by a pulse to the total time during a continuous working period. The duty ratio reflects the ratio of the time the circuit or load is turned on to the total cycle time. Therefore, the control efficiency of various device operations can be improved by optimizing the duty ratio and the power consumed by these devices can be minimized, thereby minimizing power consumption.

[0003] In the prior art, the available robot arm operation duty cycle control models are either extremely complex or inefficient. In view of this, the inventor of this case has devoted himself to research and improvement, and thus this case was produced. Summary of the invention

[0004] The purpose of the present invention is to provide an efficient bionic model based on integrated learning operation to optimize the motion control of the robot arm, which can have faster speed and higher precision.

[0005] To achieve the above-mentioned purpose, the technical solution of the present invention is: an efficient bionic model for optimizing robot arm motion based on integrated learning operation, the bionic model estimates the fitness function through the mayfly algorithm, and optimizes the fitness function to obtain the initial duty cycle of various robot arm motion types in the robot arm motion type set, and determines the final duty cycle from the initial duty cycle through multi-classifier integrated learning, thereby obtaining the optimal robot arm motion group in the robot arm motion type set.

[0006] The workflow of the bionic model is as follows:

[0007] 1) Initialize a set of optimization constants, including the total number of mayflies used for optimization NM, the total number of iterations N for which these mayflies are reconfigured i The rate L at which the mayfly learns from each other r ;

[0008] 2) Generate a set of mayflies NM through the following operations. For all robot joints and their operations, the duty cycle is randomly generated by the formula DCi=STOCH(0.1,Lr), where DC i is the duty cycle of the i-th joint that performs the set of robot operations,

[0009] According to the duty cycle DCi, through the formula Estimate the fitness value of the mayfly, where fi represents the fitness value of the mayfly, N represents the total number of operations that the robot can perform, and Effi (DC j ) represents the working efficiency of the i-th joint of the robot arm under duty cycle j,

[0010] Eff i (DC j ) can be obtained by the formula Calculation, N s Represents the total number of sub-operations successfully executed at a given duty cycle, N t Indicates the total number of sub-operations that must be performed for a given set of operations;

[0011] 3) Optimize through the fitness threshold fth, the fitness threshold fth is calculated by the formula It is concluded that when the fitness value of the mayfly is f i When the fitness threshold fth is less than the threshold value, the mayfly is discarded and regenerated through step 2), and the mayflies that are not discarded are crossed to the next set of iterations for mutation until the total number of iterations N is completed. i Choose the fitness value f i >Fitness threshold fth of the ephemera NM solution is used as the initial duty cycle.

[0012] The classifier used for controlling the duty cycle in the multi-classifier ensemble learning includes several or all classifiers integrated among naive Bayes, k-nearest neighbor, support vector machine, logistic regression and multilayer perceptron.

[0013] The parameters of the classifier configuration used to control the duty cycle are as follows:

[0014] The configuration parameters of the naive Bayes classifier include the class prior probability P and the smoothing value S V , prior probability Smoothing value S V =L r ,

[0015] The k-nearest neighbor classifier is configured with parameter k=1, which is used to classify the duty cycle of a component with the component itself.

[0016] The configuration parameters of the support vector machine classifier include the regularization constant C and the allowable error tol. The regularization constant C = L r , allowable error tol=0.001*L r

[0017] Logistic regression classifier configuration parameters include tolerance tol, weight W, and maximum number of iterations M iter , allowable error tol=0.001*L r , weight W = P, maximum number of iterations M iter =NM*L r *Ni ,

[0018] The configuration parameters of the multilayer perceptron classifier are: Total number of hidden layers = NM*L r .

[0019] The classifier used to control the duty cycle in the multi-classifier ensemble learning includes a naive Bayes, k-nearest neighbor, support vector machine, logistic regression and multi-layer perceptron. The final duty cycle is calculated by the formula cout=c(NB)*A(NB)+c(kNN)*A(kNN)+c(LR)*A(LR)+c(SVM)

[0020] *A(SVM)+c(MLP)+A(MLP)

[0021] Wherein, NB is the naive Bayes classifier, kNN is the k nearest neighbor classifier, SVM is the support vector machine classifier, LR is the logistic regression classifier, MLP is the layer perceptron classifier, and c and A represent the output categories and test accuracy of different classifiers.

[0022] By adopting the above technical solution, the beneficial effects of the present invention are as follows: the above bionic model uses the Mayfly Optimization (MO) algorithm to determine the duty cycle of various robot arm components of various types of motions, the MO model estimates the fitness function, and optimizes the fitness function to identify the robot arm motion set, and then uses the fitness function to determine which robot arm motion group is optimal, including obtaining the final duty cycle of controlling the robot arm through a combination of naive Bayes (NB), k nearest neighbor (kNN), support vector machine (SVM), logistic regression (LR) and multi-layer perceptron (MLP) classifiers, this integrated learning fusion enables the model to improve the classification performance of different robot arm operations, and the technical solution of the present invention creates a successful bionic model that can maximize the control efficiency of various device operations and minimize the power consumed. In summary, the above-mentioned purpose of the present invention is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 The present invention relates to a flow chart of an efficient bionic model for optimizing the motion of a mechanical arm based on integrated learning operations;

[0024] Figure 2 The invention discloses a schematic diagram of the design structure of a mechanical arm and a controller thereof. DETAILED DESCRIPTION

[0025] In order to further explain the technical solution of the present invention, the present invention is described in detail below through specific embodiments.

[0026] This embodiment discloses an efficient bionic model based on integrated learning operation to optimize the motion of a robotic arm. Figure 1 As shown in the figure, the bionic model estimates the fitness function through the mayfly algorithm, and optimizes the fitness function to obtain the initial duty cycle of various robot arm motion types in the robot arm motion type set, and determines the final duty cycle from the initial duty cycle through multi-classifier ensemble learning, thereby obtaining the optimal robot arm motion group in the robot arm motion type set. When the bionic model is applied, the design of the robot arm and its controller can be as follows: Figure 2 As shown, using accelerometers, potentiometers and pressure sensors, the user's robotic arm and gripper movements are monitored and controlled through deep learning models. The gripper of the robot arm responds quickly to the user's hand, and due to the integrated accelerometer and potentiometer, the robotic arm's movements react faster, so it is much simpler to position the robotic arm in the appropriate position, and it will complete basic tasks faster. This is achieved by reducing the reaction time of the robotic arm, which increases its ability to be related to the user's robotic arm. The robotic arm is remotely operated using the Internet of Things, and the MQTT communication protocol is used to facilitate data transmission. The MQTT agent receives the angle of the servo system after mathematically determining the servo system angle. Control is provided through the base and servo system, and data processing is completed by the Arduino controller.

[0027] like Figure 1 As shown, the workflow of the bionic model is as follows:

[0028] 1) Initialize a set of optimization constants, including the total number of mayflies used for optimization NM, the total number of iterations N for which these mayflies are reconfigured i The rate L at which the mayfly learns from each other r .

[0029] 2) Generate a set of mayflies NM through the following operations. For all robot joints and their operations, the duty cycle is randomly generated by the formula DCi=STOCH(0.1,Lr), where DC i is the duty cycle of the i-th joint that performs the set of robot operations,

[0030] According to the duty cycle DCi, through the formula Estimate the fitness value of the mayfly, where fi represents the fitness value of the mayfly, N represents the total number of operations that the robot can perform, and Eff i (DC j ) represents the working efficiency of the i-th joint of the robot arm under duty cycle j,

[0031] Eff i (DC j ) can be obtained by the formula Calculation, N sRepresents the total number of sub-operations successfully executed at a given duty cycle, N t Represents the total number of sub-operations that must be performed for a given set of operations; for example, weightlifting requires the robot arm to perform the following sub-operations: ① move to the correct position; ② move towards the object; ③ grab the object; ④ move up to the initial position.

[0032] Once all the mayfly NMs are generated in step 2) above, their fitness values ​​are estimated for initial optimization, and the optimization process proceeds as follows step 3).

[0033] 3) Optimize through the fitness threshold fth, the fitness threshold fth is calculated by the formula It is concluded that when the fitness value of the mayfly is f i When the fitness threshold fth is less than the threshold value, the mayfly is discarded and regenerated through step 2), and the mayflies that are not discarded are crossed to the next set of iterations for mutation until the total number of iterations N is completed. i Choose the fitness value f i >Fitness threshold fth of the mayfly NM solution to be the initial duty cycle.

[0034] The mayfly obtained in the above step 3) is used to train an integrated classifier with multi-classifier integrated learning operation optimization proposed by the present invention, and the final duty cycle is outputted through it. In the present invention, the classifier used to control the duty cycle in the multi-classifier integrated learning includes several or all classifiers integrated with naive Bayes, k-nearest neighbor, support vector machine, logistic regression and multilayer perceptron. In this embodiment, naive Bayes, k-nearest neighbor, support vector machine, logistic regression and multilayer perceptron are integrated as classifiers for controlling the duty cycle, wherein the parameters of the classifier configuration for controlling the duty cycle are as follows:

[0035] The configuration parameters of the naive Bayes classifier include the class prior probability P and the smoothing value S V , prior probability Smoothing value S V =L r ,

[0036] The k-nearest neighbor classifier is configured with parameter k=1, which is used to classify the duty cycle of a component with the component itself.

[0037] The configuration parameters of the support vector machine classifier include the regularization constant C and the allowable error tol. The regularization constant C = L r , allowable error tol=0.001*L r ,

[0038] Logistic regression classifier configuration parameters include tolerance tol, weight W, and maximum number of iterations M iter , allowable error tol=0.001*L r, weight W = P, maximum number of iterations M iter =NM*L r *N i ,

[0039] The configuration parameters of the multilayer perceptron classifier are: Total number of hidden layers = NM*L r .

[0040] The classifier used to control the duty cycle in the multi-classifier ensemble learning includes a naive Bayes, k-nearest neighbor, support vector machine, logistic regression and multi-layer perceptron. The final duty cycle is calculated by the formula cout=c(NB)*A(NB)+c(kNN)*A(kNN)+c(LR)*A(LR)+c(SVM)

[0041] *A(SVM)+c(MLP)+A(MLP)

[0042] Wherein, NB is the naive Bayes classifier, kNN is the k nearest neighbor classifier, SVM is the support vector machine classifier, LR is the logistic regression classifier, MLP is the layer perceptron classifier, and c and A represent the output categories and test accuracy of different classifiers.

[0043] The above-mentioned bionic model can improve the classification performance of different robot arm operations, and can solve the problem that the traditional methods of controlling the duty cycle of the robot arm are either inefficient or too complicated. The present invention uses the mayfly algorithm and integrated learning technology to create a successful bionic model, which can maximize the control efficiency of various device operations and minimize the power consumption.

[0044] The above embodiments and drawings do not limit the product form and style of the present invention. Any appropriate changes or modifications made by ordinary technicians in the relevant technical field should be deemed to be within the patent scope of the present invention.

Claims

1. An efficient bionic model for optimizing robot arm motion based on integrated learning operations, characterized in that: The bionic model estimates the fitness function through the mayfly algorithm, and optimizes the fitness function to obtain the initial duty cycle of various robotic arm motion types in the robotic arm motion type set. The final duty cycle is determined from the initial duty cycle through multi-classifier ensemble learning, thereby obtaining the optimal robotic arm motion group in the robotic arm motion type set.

2. The efficient bionic model for optimizing the motion of a robotic arm based on integrated learning operations as claimed in claim 1, characterized in that: The workflow of the bionic model is as follows: 1) Initialize a set of optimization constants, including the total number of mayflies used for optimization NM, the total number of iterations N for which these mayflies are reconfigured i The rate L at which the mayfly learns from each other r ; 2) Generate a set of mayflies NM through the following operations. For all robot joints and their operations, the duty cycle is randomly generated by the formula DCi=STOCH(0.1,Lr), where DC i is the duty cycle of the i-th joint that performs the set of robot operations, According to the duty cycle DCi, through the formula Estimate the fitness value of the mayfly, where fi represents the fitness value of the mayfly, N represents the total number of operations that the robot can perform, and Eff i (DC j ) represents the working efficiency of the i-th joint of the robot arm under duty cycle j, Eff i (DC j ) can be obtained by the formula Calculation, N s Represents the total number of sub-operations successfully executed at a given duty cycle, N t Indicates the total number of sub-operations that must be performed for a given set of operations; 3) Optimize through the fitness threshold fth, the fitness threshold fth is calculated by the formula It is concluded that when the fitness value of the mayfly is f i When the fitness threshold fth is less than the threshold value, the mayfly is discarded and regenerated through step 2), and the mayflies that are not discarded are crossed to the next set of iterations for mutation until the total number of iterations N is completed. i Choose the fitness value f i >Fitness threshold fth of the ephemera NM solution is used as the initial duty cycle.

3. The efficient bionic model for optimizing the motion of a robotic arm based on integrated learning operations as claimed in claim 1, characterized in that: The classifier used for controlling the duty cycle in the multi-classifier ensemble learning includes several or all classifiers integrated among naive Bayes, k-nearest neighbor, support vector machine, logistic regression and multilayer perceptron.

4. The efficient bionic model for optimizing the motion of a robotic arm based on integrated learning operations as claimed in claim 2, characterized in that: The classifier used for controlling the duty cycle in the multi-classifier ensemble learning includes several or all classifiers integrated among naive Bayes, k-nearest neighbor, support vector machine, logistic regression and multilayer perceptron.

5. An efficient bionic model for optimizing the motion of a robotic arm based on integrated learning operations as claimed in claim 3 or 4, characterized in that: The parameters of the classifier configuration used to control the duty cycle are as follows: The configuration parameters of the naive Bayes classifier include the class prior probability P and the smoothing value S V , Prior probability Smoothing value S V =L r , The k-nearest neighbor classifier is configured with parameter k=1, which is used to classify the duty cycle of a component and the component itself. The configuration parameters of the support vector machine classifier include the regularization constant C and the allowable error tol. The regularization constant C = L r , allowable error tol=0.001*L r , Logistic regression classifier configuration parameters include tolerance tol, weight W, and maximum number of iterations M iter , allowable error tol=0.001*L r , weight W = P, maximum number of iterations M iter =NM*L r *N i , The configuration parameters of the multilayer perceptron classifier are: Total number of hidden layers = NM*L r .

6. An efficient bionic model for optimizing the motion of a robotic arm based on integrated learning operations as claimed in claim 3 or 4, characterized in that: The classifier used to control the duty cycle in the multi-classifier ensemble learning includes a naive Bayes, k-nearest neighbor, support vector machine, logistic regression and multi-layer perceptron. The final duty cycle is calculated by the formula cout=c(NB)*A(NB)+c(kNN)*A(kNN)+c(LR)*A(LR)+c(SVM)*A(SVM)+c(MLP)+A(MLP) Wherein, NB is the naive Bayes classifier, kNN is the k nearest neighbor classifier, SVM is the support vector machine classifier, LR is the logistic regression classifier, MLP is the layer perceptron classifier, and c and A represent the output categories and test accuracy of different classifiers.

7. The efficient bionic model for optimizing the motion of a robotic arm based on integrated learning operations as claimed in claim 5, characterized in that: The classifier used to control the duty cycle in the multi-classifier ensemble learning includes a naive Bayes, k-nearest neighbor, support vector machine, logistic regression and multi-layer perceptron. The final duty cycle is calculated by the formula cout=c(NB)*A(NB)+c(kNN)*A(kNN)+c(LR)*A(LR)+c(SVM)*A(SVM)+c(MLP)+A(MLP) Wherein, NB is the naive Bayes classifier, kNN is the k nearest neighbor classifier, SVM is the support vector machine classifier, LR is the logistic regression classifier, MLP is the layer perceptron classifier, and c and A represent the output categories and test accuracy of different classifiers.

Images