A method for optimizing a control system of a robot joint module

Abstract

The application provides an optimization method of a robot joint module control system, which comprises a response performance test of a robot joint module, a performance test taking a load mass m, an output speed v of the joint module, a proportional coefficient P, an integral coefficient I and a differential coefficient D as test variables, and obtaining a joint module response performance data set; a response performance prediction of the robot joint module, building a prediction model, and training the prediction model through the response performance data set; a response performance optimization of the joint module, and obtaining optimized control parameters in the trained prediction model. Beneficial effects: the application realizes the prediction of the response performance of the robot joint module, solves the low efficiency problem of the traditional method in the optimization of the optimal control parameters of the robot joint module, simultaneously obtains the optimal control parameters under different response performance requirements, effectively improves the optimization efficiency and precision of the control parameters of the robot joint module, and provides an effective theoretical basis for performance optimization.

Description

Technical Field

[0001] This invention relates to an optimization method for a control system, and more particularly to an optimization method for a robot joint module control system, belonging to the field of robot joint control technology. Background Technology

[0002] As artificial intelligence and robotics are increasingly integrated into people's lives and work, robot joint modules are essential core motion components for articulated robots to achieve movement. Therefore, the performance of robot joint modules directly affects the robot's transmission accuracy, operating speed, vibration, and noise levels. The control method used for a single robot joint module is PID control, with proportional coefficient P, integral coefficient I, and derivative coefficient D as its main control parameters. The main response performance indicators of a robot joint module are overshoot and response time. Factors affecting the response performance of a robot joint module include load, output speed, and control parameters (proportional coefficient P, integral coefficient I, and derivative coefficient D). Load and output speed are determined by the operating conditions of the robot joint module, while the control parameters are set manually. Therefore, the quality of the control parameters directly determines the performance of the robot joint module. In summary, optimizing the control parameters of the robot joint module and improving its response performance can effectively improve the overall performance of the robot.

[0003] Currently, traditional debugging methods are mostly applied to the entire robot, using experimental methods to debug and calibrate control parameters. This makes it difficult to accurately obtain the optimal control parameters for individual joint modules, and the optimization efficiency is low due to the need for a large number of experiments.

[0004] Chinese patent CN 108227479 B discloses a PID control method and system for a multi-joint robot. By comprehensively adjusting the overall gain, proportional gain, integral gain, and derivative gain, it achieves faster system response and facilitates quicker device stabilization. However, this prior art does not consider the impact of PID control parameters on control performance under different speeds and loads, and has deficiencies in parameter selection and outlier handling, failing to meet the requirements for high-precision control. Summary of the Invention

[0005] Purpose of the Invention: The purpose of this invention is to address the problems existing in the prior art by providing an optimization method for a robot joint module control system. This method combines response performance testing of the robot joint module with intelligent models, enabling both prediction of the robot joint module's response performance and optimization of the joint module's control parameters.

[0006] Technical solution: An optimization method for a robot joint module control system, comprising the following steps:

[0007] Step S1: Conduct a response performance test on the robot joint module. Using the mass m of the weight representing the load, the output speed v of the joint module, the proportional coefficient P of the joint module control system, the integral coefficient I of the joint module control system, and the differential coefficient D of the joint module control system as test variables, design a 5-factor response performance test and obtain the joint module response performance dataset.

[0008] Step S2: Predict the response performance of the robot joint module, build a prediction model, and train the prediction model using the joint module response performance dataset obtained in step S1.

[0009] Step S3: Optimize the response performance of the joint module, establish the response performance index of the robot joint module, take the values ​​of the 5 factors in step S1 evenly to form a parameter combination, and substitute the parameter combination into the prediction model trained in step S2 to obtain the minimum value of the response performance index as the optimized control parameter.

[0010] This invention, taking into account the influence of PID control parameters on the control performance under different speeds and loads, combines the response performance test of the robot joint module with an intelligent model. This allows for both prediction of the response performance of the robot joint module and optimization of the control parameters of the joint module.

[0011] In order to investigate the influence of PID control parameters on the control performance of the joint module under different speeds and loads, the specific test steps for the response performance of the robot joint module in step S1 are as follows:

[0012] Step S1.1: Set up a test bench for the response performance of the robot joint module. Select the robot joint module 1 for testing and fix it to the base 2 through the fixing flange 7. The base 2 is fixedly installed on the plate 5. The output flange 6 of the robot joint module 1 is equipped with a connecting rod 3. The other end of the connecting rod 3 is equipped with a weight 4 of mass m. The mass m of the weight 4 is changed according to the test requirements. Connect the robot joint module 1 to the host computer debugger. The host computer controls the robot joint module 1 according to the speed and load designed in the test and adjusts the PID parameter values.

[0013] Step S1.2, Experimental Design: Design an orthogonal experimental table with 5 factors and n levels (n≥5) using the mass m of the weight representing the load, the output speed v of the joint module, the proportional coefficient P, the integral coefficient I, and the differential coefficient D as experimental variables.

[0014] Step S1.3: Using the joint module response performance dataset, and following the orthogonal experimental table in Step S1.2, conduct experiments on the robot joint module response performance test bench built in Step S1.1 to obtain the robot joint module velocity response curves under different experimental parameters, and extract the joint module response performance index from the velocity response curves: overshoot O. shoot and response time R time Record and store the data to form a dataset of joint module response performance.

[0015] A joint module is an integrated mechanical device that enables mechanical structures to rotate, swing, and translate in different directions, thereby completing specific functions or tasks. Joint modules are widely used in various mechanical equipment and devices, such as in the construction of industrial robots. Different designs can be implemented depending on the application scenario. Using a constructed experimental platform, joint module operation tests were conducted according to a designed orthogonal experimental table. The tests collected relevant data such as response time and overshoot under different influencing factors. The data processing can provide a dataset for training predictive models and allows analysis of the relationship between system overshoot and response speed under different loads and operating speeds.

[0016] In a preferred embodiment, to address the issue of the model getting trapped in local optima, the specific steps for optimizing the joint module response performance in step S2 are as follows:

[0017] Step S2.1: Establish an MA-BPNN mayfly-neural network prediction model, which includes an MA mayfly model and a BPNN backpropagation neural network model. The MA mayfly model provides optimized weights and thresholds for the BPNN backpropagation neural network model.

[0018] Step S2.2: Train the MA-BPNN mayfly neural network prediction model. First, import the joint module response performance dataset obtained in step S1.2 into the model. The initial mayfly population size is N. Then, update the positions of male and female mayflies and calculate their fitness values. Then, iteratively generate offspring mayflies and update their positions. If the fitness value of the offspring mayfly is less than F... v If the condition is met, the iteration ends; otherwise, the iteration continues to generate offspring mayflies.

[0019] The globally optimal parameters generated by the mayfly are used as the weights and thresholds of the neural network model for prediction and error calculation until the error after iteration is less than E. u At that time, the output overshoot is O shoot and response time R time The prediction results indicate that the MA-BPNN prediction model has been trained.

[0020] Optimizing control parameters for mechanical joint modules is a highly complex nonlinear problem, influenced by multiple factors such as load, operating speed, and PID control parameters. Establishing an effective mathematical relationship to describe the connections between these factors is extremely difficult. Therefore, employing machine learning methods to explore the relationship between these influencing factors and the response performance of the control system is an effective approach. Currently, various machine learning methods have been introduced into the research of joint module motion control, among which artificial neural networks have proven to achieve good results in control parameter optimization. However, the aforementioned models may get trapped in local optima. To address this issue, the MA-BPNN control parameter optimization model is adopted, which comprehensively utilizes the powerful nonlinear mapping capability and flexible structure of BP neural networks, as well as the strong fault tolerance of MA search models for errors and noise, effectively avoiding the predicament of getting trapped in local optima during the computation process.

[0021] In a preferred embodiment, to predict the response performance of the robot joint module, the training steps for the MA-BPNN mayfly-neural network prediction model in step S2.2 are as follows:

[0022] Step S2.2.1: Initialize the model. Import the joint module response performance dataset obtained in Step 1.3 into the model. Set the number of male and female mayflies in the mayfly model to N. mayflies Each mayfly's position represents a set of weights and threshold parameters for a backpropagation (BP) neural network, and its velocity represents the step size of these parameters during the update process. The weights and thresholds of the BP neural network are initialized. The positions and velocities of the mayflies are initialized.

[0023] Step S2.2.2: Update the velocity and position of the male mayfly individuals. Update the velocity v of the i-th male mayfly in the j-dimensional search space after t+1 iterations using formulas (1) and (2). t+1 mij and position x t+1 ij ,

[0024]

[0025]

[0026] In the formula, t is the t-th iteration, x t ij It represents the position of the i-th male mayfly in the j-th dimension of the search space after the t-th iteration, v. t mij It represents the velocity of the i-th male mayfly in the j-th dimension of the search space after the t-th iteration. It is the historical best position of the i-th male mayfly in the j-th dimension of the search space. ζ is the optimal position of the population in the j-th dimension of the search space, ζ is the dynamic inertia coefficient, a1 and a2 are the attraction coefficients of male mayflies, β is the visibility coefficient, and r is the population optimum. P This represents the current location of the male mayfly and... The distance between them, r G This represents the current location of the male mayfly and The distance between them, da is the dance coefficient, used to describe the process of attracting female mayflies, r c It is a random number, r c ∈[-1,1], F3(x) is the fitness function, as shown in Formula 3:

[0027]

[0028] In the formula, x represents the position of the mayfly, and y represents the position of the mayfly. actual This is the actual output value; y predicte This is the predicted output value; N is the number of samples in the sample set.

[0029] Step S2.2.3: Update the velocity and position of the female mayfly individuals. Update the velocity v of the i-th female mayfly in the j-dimensional search space after t+1 iterations using formulas (4) and (5). t+1 fij and position y t+1 ij ,

[0030]

[0031]

[0032] In the formula, y t ij It represents the position of the i-th female mayfly in the j-th dimension of the search space after the t-th iteration. is the velocity of the i-th female mayfly in the j-th dimension search space after the t-th iteration, a3 is the attraction coefficient of the female mayfly, and r mf is the distance between female and male mayflies, and fl is the random flight coefficient;

[0033] Step S2.2.4: Generate offspring mayflies; update the mayfly population according to formulas (6) and (7):

[0034] offspring1=r l ·male+(1-r l )·female (6)

[0035] offspring2=r l ·female+(1-r l )·male (7)

[0036] In the formula, offspring1 and offspring2 represent two offspring mayflies; r l r is a random number l ∈[-1,1]; male and female represent male and female mayflies, respectively;

[0037] Step S2.2.5: Fitness value determination; F v Let F3(x) be the target fitness value, if F3(x) ≤ F v If the iteration ends, proceed to step S2.2.2 and iterate again.

[0038] Step S2.2.6: Output the global optimal position of the mayfly and obtain the optimized weights and thresholds;

[0039] Step S2.2.7: Forward propagation of the BPNN, the input neurons are the mass m of the weight, the output velocity v of the joint module, the proportional coefficient P, integral coefficient I, and differential coefficient D of the joint module control system, and the output neuron is the overshoot O. shoot and response time R time The input layer data is passed to the neurons in the hidden layer through optimized weights and thresholds; the activation function F1(x) = (e^(x-1)) is applied to the neurons in the hidden layer. x -e -x ) / (e x +e -x ), calculate θ of the g-th neuron in the hidden layer according to formula (8). g The data from the hidden layer is then passed to the neurons in the output layer via weights and thresholds, and an activation function F2(x) = (e^(x-1) / x) is applied to the neurons in the output layer. x ) / (e x +e -x ), calculate Y of the k-th neuron in the output layer according to formula (9) k ;

[0040]

[0041] In the formula, X h W represents the h-th neuron in the input layer. hg It is the weight between the h-th neuron in the input layer and the g-th neuron in the hidden layer; T g It is the threshold of the g-th neuron in the hidden layer; n hide It is the number of nodes in the hidden layer;

[0042]

[0043] In the formula, T k It is the threshold of the k-th neuron in the output layer; W gkIt is the weight between the g-th neuron in the hidden layer and the k-th neuron in the output layer; n out This is the number of nodes in the output layer;

[0044] Step S2.2.8: Calculate the prediction error E u ; Calculate E according to formula (10) u The value;

[0045] E u =|y actual -y predicted | (10)

[0046] Step S2.2.9: Error backpropagation; Use the backpropagation model to adjust the weights and thresholds to reduce Eu;

[0047] Step S2.2.10: Training iteration. By repeatedly executing steps S2.2.7, S2.2.8, and S2.2.9, the weights and thresholds are adjusted until Eu ≤ Ev, where Ev is the target prediction error value. Output O. shoot and R time The prediction results indicate that the MA-BPNN prediction model has been trained.

[0048] This method enables the prediction of the response performance of robot joint modules, solving the problem of low efficiency in optimizing the optimal control parameters of robot joint modules using traditional methods. Furthermore, by adjusting the weighting coefficients in the response performance indicators, optimal control parameters can be obtained under different response performance requirements, effectively improving the optimization efficiency and accuracy of robot joint module control parameters and providing a sound theoretical basis for robot performance optimization.

[0049] In a preferred embodiment, to optimize the control based on the optimization model, the specific steps for optimizing the joint module response performance in step S3 are as follows:

[0050] Step S3.1: Establish the robot joint module response performance index K

[0051] K=α1O shoot +α2R time (11)

[0052] In the formula, α1 is the overshoot weighting coefficient, α2 is the response time weighting coefficient, and α1+α2=1;

[0053] Step S3.2: For the five factors affecting the robot joint module: the mass of the weight m, the output speed v of the joint module, the proportional coefficient P, the integral coefficient I, and the differential coefficient D, each factor is uniformly selected from n (n≥5) values ​​within its respective range, forming n 5 Group parameter combinations;

[0054] Step S3.3: Take n from step S3.2 5 Each parameter combination is fed into the MA-BPNN model trained in step S2, and the response performance index K of each parameter combination is calculated. The parameter combination with the minimum K value, namely the proportional coefficient P, integral coefficient I, and derivative coefficient D, is the optimal control parameter after optimization.

[0055] By optimizing the control parameters of the joint module, the optimal values ​​of the control parameters under different speeds and loads can be obtained, thus solving the problem of joint module control parameter optimization.

[0056] Beneficial Effects: The proposed method for optimizing robot joint module control parameters combines orthogonal experiments on robot joint module response performance with an intelligent model (MA-BPNN model), enabling the prediction of robot joint module response performance and solving the problem of low efficiency in optimizing optimal control parameters of robot joint modules using traditional methods. Furthermore, by adjusting the weighting coefficients in the response performance indicators, optimal control parameters under different response performance requirements can be obtained, effectively improving the optimization efficiency and accuracy of robot joint module control parameters and providing a sound theoretical basis for robot performance optimization. Attached Figure Description

[0057] Figure 1 This is the overall flowchart of the present invention;

[0058] Figure 2 This is a schematic diagram of the structure of the robot joint module response performance test bench of the present invention;

[0059] Figure 3 This is a schematic diagram of the robot joint module structure of the present invention;

[0060] Figure 4 This is a speed response curve of the robot joint module of the present invention;

[0061] Figure 5 This is a flowchart of the MA-BPNN prediction model of the present invention;

[0062] Figure 6 This is a comparison of the prediction results of the MA-BPNN prediction model of this invention with those of the traditional BP neural network model. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] like Figure 1 As shown, the specific implementation process of the present invention is as follows: S1 Conduct response performance tests on the robot joint module; S2 Predict the response performance of the robot joint module; S3 Optimize the response performance of the joint module.

[0065] S1 specifically includes the following steps:

[0066] Step S1.1: Construct a test bench for the response performance of the robot joint module. For example... Figure 2 and Figure 3 As shown, the robot joint module 1 is fixed to the base 2 by the fixing flange 7. The base 2 is installed on the plate 5 by bolts. The output flange 6 of the robot joint module is equipped with a connecting rod 3. The other end of the connecting rod 3 is equipped with a weight 4 with a weight of m. The weight m of the weight 4 can be replaced according to the test conditions.

[0067] Step S1.2: Design an orthogonal experimental table. Using the mass m of the weight, the output speed v of the joint module, the proportional coefficient P of the joint module control system, the integral coefficient I of the joint module control system, and the differential coefficient D of the joint module control system as experimental variables, design an orthogonal experimental table with 5 factors and 5 levels, totaling 25 sets.

[0068] Step 1.3: Following the orthogonal experimental setup in Step S1.2, conduct 25 sets of tests on the robot joint module response performance test bench built in Step 1.1. Obtain the robot joint module velocity response curves under different test parameters, and extract the joint module's response performance index from the velocity response curves: overshoot O. shoot and response time R time ,like Figure 4 As shown. Record and store 25 sets of m, v, P, I, O. shoot R time The values ​​are used to form a dataset of joint module response performance.

[0069] Step S2 specifically includes the following steps:

[0070] Step S2.1: Establish the MA-BPNN mayfly neural network prediction model. This model mainly includes the MA mayfly model and the BPNN backpropagation neural network model, which provide optimized weights and thresholds.

[0071] Step S2.2: Train the MA-BPNN prediction model. For example... Figure 5 The training steps shown are as follows:

[0072] Step S2.2.1: Initialize the model. Import the joint module response performance dataset obtained in step S1.3 into the model. Set the number of male and female mayflies in the mayfly model to N.mayflies = 25. The position of each mayfly represents a set of weights and threshold parameters for a backpropagation (BP) neural network, and the velocity represents the step size of these parameters during the update process. Initialize the weights and thresholds of the BP neural network. Initialize the positions and velocities of the mayflies.

[0073] Step S2.2.2: Update the velocity and position of the male mayfly individuals. Update the velocity v of the i-th male mayfly in the j-th dimension search space after t+1 iterations using formulas (1) and (2). t+1 mij and position x t+1 ij .

[0074]

[0075]

[0076] In the formula, t is the t-th iteration; x t ij It represents the position of the i-th male mayfly in the j-th dimension of the search space after the t-th iteration; v t mij It is the velocity of the i-th male mayfly in the j-th dimension search space after the t-th iteration; It is the historical best position of the i-th male mayfly in the j-th dimension of the search space; ζ is the optimal position of the population in the j-th dimension of the search space; ζ is the dynamic inertia coefficient; a1 and a2 are the attraction coefficients of male mayflies, a1 = 0.2, a2 ​​= 0.3; β is the visibility coefficient, β = 2; r P This represents the current location of the male mayfly and The distance between them; r G This represents the current location of the male mayfly and The distance between them; da is the dance coefficient, used to describe the process of attracting female mayflies; r c It is a random number, r c ∈[-1,1]; F3(x) is the fitness function, as shown in Formula 3.

[0077]

[0078] In the formula, x represents the position of the mayfly, and y represents the position of the mayfly. actual This is the actual output value; y predicte is the predicted output value; N is the number of samples.

[0079] Step S2.2.3: Update the velocity and position of the female mayfly. Update the velocity v of the i-th female mayfly in the j-th dimension search space after t+1 iterations using formulas (4) and (5). t+1 fij and position yt+1 ij .

[0080]

[0081]

[0082] In the formula, y t ij It is the position of the i-th female mayfly in the j-th dimension search space after the t-th iteration; is the velocity of the i-th female mayfly in the j-th dimension search space after the t-th iteration; a3 is the attraction coefficient of the female mayfly, a3 = 0.2; r mf is the distance between female and male mayflies; fl is the random flight coefficient.

[0083] Step S2.2.4: Generate offspring mayflies. Update the mayfly population according to formulas (6) and (7).

[0084] offspring1=r l ·male+(1-r l )·female (6)

[0085] offspring2=r l ·female+(1-r l )·male (7)

[0086] In the formula, offspring1 and offspring2 represent two offspring mayflies; r l r is a random number l ∈[-1,1]; male and female represent male and female mayflies, respectively.

[0087] Step S2.2.5: Fitness value determination. F v F represents the target fitness value. v =0.001, if F3(x)≤F v If the iteration fails, the iteration ends; otherwise, proceed to step 2.2.2 and iterate again.

[0088] Step S2.2.6: Output the global optimal position of the mayfly and obtain the optimized weights and thresholds.

[0089] Step S2.2.7: Forward propagation of the BPNN. The input neurons are the mass m of the weight, the output velocity v of the joint module, the proportional coefficient P, integral coefficient I, and differential coefficient D of the joint module control system. The output neuron is the overshoot O. shoot and response time R timeThe input layer data is passed to the neurons in the hidden layer through optimized weights and thresholds. An activation function F1(x) = (e^(-x / x)) is applied to the neurons in the hidden layer. x -e -x ) / (e x +e -x ), calculate θ of the g-th neuron in the hidden layer according to formula (8). g The data from the hidden layer is then passed to the neurons in the output layer via weights and thresholds, and an activation function F2(x) = (e^(x-1) / x) is applied to the neurons in the output layer. x ) / (e x +e -x ), calculate Y of the k-th neuron in the output layer according to formula (9) k .

[0090]

[0091] In the formula, X h W represents the h-th neuron in the input layer. hg It is the weight between the h-th neuron in the input layer and the g-th neuron in the hidden layer; T g It is the threshold of the g-th neuron in the hidden layer; n hide It represents the number of nodes in the hidden layer.

[0092]

[0093] In the formula, T k It is the threshold of the k-th neuron in the output layer; W gk It is the weight between the g-th neuron in the hidden layer and the k-th neuron in the output layer; n out It represents the number of nodes in the output layer.

[0094] Step S2.2.8: Calculate the prediction error E u Calculate E according to formula (10). u The value of .

[0095] E u =|y actual -y predicted | (10)

[0096] Step S2.2.9: Error Backpropagation. Use the backpropagation model to adjust the weights and thresholds to reduce E. u .

[0097] Step S2.2.10: Training Iteration. Adjust the weights and threshold by repeatedly executing steps 2.2.7, 2.2.8, and 2.2.9 until E... u ≤E v Up to now, E v Let E be the target prediction error value.v =0.0001. Output O shoot and R time The prediction results indicate that the MA-BPNN prediction model has been trained.

[0098] A comparison of the prediction results of the trained MA-BPNN prediction model and the traditional BP neural network model is shown below. Figure 6 As shown. Compared to the traditional BP neural network model, the MA-BPNN prediction model improves the coefficient of certainty by 2.06% and O(n)%. shoot The mean squared error was reduced by 46.92%, R time The mean squared error was reduced by 43.75%, and the running time was reduced by 0.33 seconds. The MA-BPNN prediction model proposed in this invention outperforms the traditional BP neural network model in both prediction error and running efficiency.

[0099] S3 specifically includes the following steps:

[0100] Step S3.1: Establish the robot joint module response performance index K.

[0101] K=α1O shoot +α2R time (11)

[0102] In the formula, α1 is the overshoot weighting coefficient, α1=0.6; α2 is the response time weighting coefficient, α2=0.4.

[0103] Step S3.2: For the five factors affecting the robot joint module: the mass of the weight m, the output speed v of the joint module, the proportional coefficient P of the joint module control system, the integral coefficient I of the joint module control system, and the differential coefficient D of the joint module control system, the experimental variables are taken. Each factor is evenly assigned 5 values ​​within its respective value range, forming 3125 sets of parameter combinations.

[0104] Step S3.3: Substitute the 3125 sets of parameters from step 3.2 into the MA-BPNN model trained in step S2, calculate the response performance index K for each set of parameters, and obtain the optimal control parameters after optimization by taking the proportional coefficient P, integral coefficient I, and derivative coefficient D that have the minimum K value.

[0105] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An optimization method for a robot joint module control system, characterized in that, Includes the following steps: Step S1: Conduct a response performance test on the robot joint module. Using the mass m of the weight representing the load, the output speed v of the joint module, the proportional coefficient P of the joint module control system, the integral coefficient I of the joint module control system, and the differential coefficient D of the joint module control system as test variables, design a 5-factor response performance test and obtain the joint module response performance dataset. Step S2: Predict the response performance of the robot joint module, build a prediction model, and train the prediction model using the joint module response performance dataset obtained in step S1. Step S3: Optimize the response performance of the joint module, establish the response performance index of the robot joint module, take the values ​​of the 5 factors in step S1 evenly to form a parameter combination, and substitute the parameter combination into the prediction model trained in step S2 to obtain the minimum value of the response performance index as the optimized control parameter. The specific test steps for the response performance of the robot joint module in step S1 are as follows: Step S1.1: Build a test bench for the response performance of the robot joint module. Select the robot joint module (1) for the test and fix it on the base (2) through the fixing flange (7). The base (2) is fixedly installed on the plate (5). The output flange (6) of the robot joint module (1) is equipped with a connecting rod (3). The other end of the connecting rod (3) is equipped with a weight (4) with a mass of m. The mass m of the weight (4) is changed according to the test requirements. Connect the robot joint module (1) to the host computer debugger. The host computer controls the robot joint module (1) according to the speed and load designed in the test and adjusts the PID parameter value. Step S1.2, Experimental Design: Design an orthogonal experimental table with 5 factors and n levels (n≥5) using the mass m of the weight representing the load, the output speed v of the joint module, the proportional coefficient P, the integral coefficient I, and the differential coefficient D as experimental variables. Step S1.3: Using the joint module response performance dataset, and following the orthogonal experimental table in Step S1.2, conduct experiments on the robot joint module response performance test bench built in Step S1.1 to obtain the robot joint module velocity response curves under different experimental parameters, and extract the joint module response performance index from the velocity response curves: overshoot O. shoot and response time R time Record and store the data to form a joint module response performance dataset; The specific steps for predicting the joint module response performance in step S2 are as follows: Step S2.1: Establish an MA-BPNN mayfly-neural network prediction model, which includes an MA mayfly model and a BPNN backpropagation neural network model. The MA mayfly model provides optimized weights and thresholds for the BPNN backpropagation neural network model. Step S2.2: Train the MA-BPNN prediction model. First, import the joint module response performance dataset obtained in step S1.3 into the model. The initial population size of mayflies is N. Then, update the positions of male and female mayflies and calculate their fitness values. Then, iteratively generate offspring mayflies and update their positions. If the fitness value of the offspring mayflies is less than F... v If the condition is met, the iteration ends; otherwise, the iteration continues to generate offspring mayflies. The globally optimal parameters generated by the mayfly are used as the weights and thresholds of the neural network model for prediction and error calculation until the error after iteration is less than E. u At that time, the output overshoot is O shoot and response time R time The prediction results indicate that the MA-BPNN prediction model has been trained.

2. The optimization method for the robot joint module control system according to claim 1, characterized in that, The training steps for the MA-BPNN mayfly neural network prediction model in step S2.2 are as follows: Step S2.2.1: Initialize the model. Import the joint module response performance dataset obtained in Step 1.3 into the model. Set the number of male and female mayflies in the mayfly model to N. mayflies Each mayfly's position represents a set of weights and threshold parameters for a backpropagation (BP) neural network, and its velocity represents the step size of these parameters during the update process. The weights and thresholds of the BP neural network are initialized. The positions and velocities of the mayflies are initialized. Step S2.2.2: Update the velocity and position of the male mayfly individuals. Update the velocity v of the i-th male mayfly in the j-dimensional search space after t+1 iterations using formulas (1) and (2). t+1 mij and position x t+1 ij , (1) In the formula, ζ is the dynamic inertia coefficient, t is the t-th iteration, and x... t ij It represents the position of the i-th male mayfly in the j-th dimension of the search space after the t-th iteration, v. t mij P is the velocity of the i-th male mayfly in the j-th dimension of the search space after the t-th iteration, where a1 and a2 are the attraction coefficients of the male mayfly. bestij G is the historical best position of the i-th male mayfly in the j-th dimension of the search space. bestj It is the optimal position of the population in the j-th dimension of the search space, e is the natural constant, β is the visibility coefficient, and r P The current location of the male mayfly is represented by P. bestij The distance between them, r G The current location of the male mayfly is represented by G. bestj The distance between them, da is the dance coefficient, used to describe the process of attracting female mayflies, r c It is a random number, r c ∈[-1, 1], F3(x) is the fitness function. (2) In the formula, t is the t-th iteration, x t ij It represents the position of the i-th male mayfly in the j-th dimension of the search space after the t-th iteration, v. t +1 mij It is the velocity of the i-th male mayfly in the j-th dimension search space after the (t+1)-th iteration; As shown in Formula 3: (3) In the formula, x represents the position of the mayfly, and y represents the position of the mayfly. actual This is the actual output value; y predicte It predicts the output value; N is the size of the sample set; Step S2.2.3: Update the velocity and position of the female mayfly individuals. Update the velocity v of the i-th female mayfly in the j-dimensional search space after t+1 iterations using formulas (4) and (5). t+1 fij and position y t+1 ij , (4) In the formula, ζ is the dynamic inertia coefficient, and y t ij It represents the position of the i-th female mayfly in the j-th dimension of the search space after the t-th iteration, v. t fij is the velocity of the i-th female mayfly in the j-th dimension search space after the t-th iteration, a3 is the attraction coefficient of the female mayfly, and r mf is the distance between female and male mayflies, and fl is the random flight coefficient; (5) In the formula, y t ij It represents the position of the i-th female mayfly in the j-th dimension of the search space after the t-th iteration, v. t+1 fij It is the velocity of the i-th female mayfly in the j-th dimension search space after the (t+1)-th iteration; Step S2.2.4: Generate offspring mayflies; update the mayfly population according to formulas (6) and (7): (6) (7) In the formula, offspring1 and offspring2 represent two offspring mayflies; r l r is a random number l ∈[-1, 1]; male and female represent male and female mayflies, respectively; Step S2.2.5: Fitness value determination; F v Let F3(x) be the target fitness value, if F3(x) ≤ F v If the iteration ends, proceed to step S2.2.2 and iterate again. Step S2.2.6: Output the global optimal position of the mayfly and obtain the optimized weights and thresholds; Step S2.2.7: Forward propagation of the BPNN, the input neurons are the mass m of the weight, the output velocity v of the joint module, the proportional coefficient P, integral coefficient I, and differential coefficient D of the joint module control system, and the output neuron is the overshoot O. shoot and response time R time The input layer data is passed to the neurons in the hidden layer through optimized weights and thresholds; the activation function F1(x)=(e^x) is applied to the neurons in the hidden layer. x -e -x ) / (e x +e -x ), calculate θ of the g-th neuron in the hidden layer according to formula (8). g The data from the hidden layer is then passed to the neurons in the output layer via weights and thresholds, and an activation function F2(x) = (e^(x-1)) is applied to the neurons in the output layer. x ) / (e x +e -x ), calculate Y of the k-th neuron in the output layer according to formula (9) k ; (8) In the formula, X h W represents the h-th neuron in the input layer. hg It is the weight between the h-th neuron in the input layer and the g-th neuron in the hidden layer; T g It is the threshold of the g-th neuron in the hidden layer; n hide It is the number of nodes in the hidden layer; (9) In the formula, T k It is the threshold of the k-th neuron in the output layer; W gk It is the weight between the g-th neuron in the hidden layer and the k-th neuron in the output layer; n out This is the number of nodes in the output layer; Step S2.2.8: Calculate the prediction error E u ; Calculate E according to formula (10) u The value; (10) In the formula, y actual This is the actual output value; y predicte It predicts the output value; Step S2.2.9: Error backpropagation; Use the backpropagation model to adjust the weights and thresholds to reduce Eu; Step S2.2.10: Training iteration. By repeatedly executing steps S2.2.7, S2.2.8, and S2.2.9, the weights and thresholds are adjusted until Eu ≤ Ev, where Ev is the target prediction error value. Output O. shoot and R time The prediction results indicate that the MA-BPNN prediction model has been trained.

3. The optimization method for the robot joint module control system according to claim 1 or 2, characterized in that, The specific steps for optimizing the joint module response performance in step S3 are as follows: Step S3.1: Establish the robot joint module response performance index K （11） In the formula, This is the overshoot weighting coefficient. For response time weighting coefficients, ; Step S3.2: For the five factors affecting the robot joint module: the mass of the weight m, the output speed v of the joint module, the proportional coefficient P, the integral coefficient I, and the differential coefficient D, each factor is uniformly selected from n (n≥5) values ​​within its respective range, forming n 5 Group parameter combinations; Step S3.3: Take n from step S3.2 5 Each parameter combination is fed into the MA-BPNN model trained in step S2, and the response performance index K of each parameter combination is calculated. The parameter combination with the minimum K value, namely the proportional coefficient P, integral coefficient I, and derivative coefficient D, is the optimal control parameter after optimization.

Images