Motor parameter identification method and system

Abstract

The invention relates to the field of motor control, and particularly discloses a motor parameter identification method and system, and the method comprises the steps: building a to-be-identified parameter model based on a motor dynamics model considering external disturbance and nominal deviation; gaussian process regression training is carried out on the training data of the to-be-identified parameter model to obtain prior distribution; and obtaining posterior distribution of test data of the to-be-identified parameter model based on the prior distribution, and performing parameter identification on the to-be-identified parameter model based on the posterior distribution to obtain a target parameter model. Priori distribution is established through Gaussian process regression training, posterior distribution under a given input condition is presumed in combination with the prior distribution, parameter identification is performed on the to-be-identified parameter model based on the posterior distribution, and the obtained target parameter model has statistical significance and relatively high generalization ability. High-dimension, small-sample and non-linear term motor parameter identification can be realized with relatively high precision, and the efficiency and stability of motor system operation are improved.

Description

Technical Field

[0001] The present invention relates to the field of motor control technology, and in particular to a motor parameter identification method and system. Background Art

[0002] In the process of modern industrial automation and intelligent manufacturing, the performance and reliability of motors, as key actuators, directly affect the efficiency and stability of the entire system. Therefore, the identification of motor parameters is particularly important. Traditional motor parameter identification methods mainly include linear regression based on experimental data, least squares method and Kalman filtering. These methods perform well when dealing with linear systems and simple nonlinear systems, but when facing complex nonlinear dynamic systems, the accuracy of motor parameter identification is often not high, which affects the efficiency and stability of the motor system.

[0003] Therefore, how to provide a motor parameter identification method that can accurately identify the parameters of a complex nonlinear dynamic motor system has become a technical problem that needs to be solved urgently by those skilled in the art. Summary of the invention

[0004] The present invention provides a motor parameter identification method and a motor parameter identification system, which solve the problem in the related art that it is difficult to accurately identify complex nonlinear dynamic motor system parameters, thereby affecting the efficiency and stability of the motor system operation.

[0005] As a first aspect of the present invention, a method for identifying motor parameters is provided, comprising:

[0006] Construct a motor dynamics model that takes into account external disturbance factors and deviations from the nominal values ​​of motor parameters;

[0007] Construct the parameter model to be identified according to the motor dynamics model;

[0008] Acquire training data and test data of the parameter model to be identified, wherein the training data and the test data both include displacement, velocity, acceleration, control input, true value of viscous friction coefficient and true value of mover mass of the motor, the training data include training input data and training output data, and the test data include test input data and test output data;

[0009] Perform Gaussian process regression training on the training data to obtain the prior distribution of the training data;

[0010] Obtaining the posterior distribution of the test data based on the prior distribution of the training data, and performing parameter identification on the parameter model to be identified based on the posterior distribution of the test data to obtain a parameter identification result;

[0011] A target parameter model is obtained according to the parameter identification results, and the motor dynamics model is updated according to the target parameter model.

[0012] Furthermore, Gaussian process regression training is performed on the training data to obtain the prior distribution of the training data, including:

[0013] Forming a training input matrix according to the training input data of the parameter model to be identified;

[0014] Forming a training output matrix according to the training output data of the parameter model to be identified;

[0015] Define a Gaussian process according to a preset mean function and a preset kernel function;

[0016] According to the Gaussian process, Gaussian process regression training is performed on the training input matrix and the training output matrix, and the covariance matrix between the data points in the training input matrix is ​​calculated.

[0017] Furthermore, the preset mean function is a zero mean function.

[0018] Furthermore, Gaussian process regression training is performed on the training input matrix and the training output matrix according to the Gaussian process, and the covariance matrix between the data points in the training input matrix is ​​calculated, including:

[0019] Parameterize the preset kernel function and select preset hyperparameters;

[0020] By maximizing the marginal likelihood function, the probability of each data in the training output matrix appearing under the preset hyperparameters is maximized to obtain the target hyperparameters;

[0021] Computes the covariance matrix between data points in the training input matrix according to the preset kernel function and target hyperparameters.

[0022] Furthermore, the expression of the preset kernel function is:

[0023]

[0024] Among them, {σ,σ f} represents the preset hyperparameter, l represents the first input of the preset kernel function, l' represents the second input of the preset kernel function, and the first input and the second input can be the same or different.

[0025] Further, the posterior distribution of the test data is obtained based on the prior distribution of the training data, and the parameter identification is performed on the parameter model to be identified based on the posterior distribution of the test data to obtain the parameter identification result, including:

[0026] Obtaining the joint Gaussian distribution of training input data, training output data, and test input data according to the prior distribution of training data;

[0027] Compute the posterior mean function of the test output data according to the joint Gaussian distribution.

[0028] Furthermore, the expression of the motor dynamics model is:

[0029]

[0030] Among them, M0 represents the true value of the mover mass, B0 represents the true value of the viscous friction coefficient, x represents the displacement, Indicates speed, represents acceleration, u represents the control input of the motor, VM represents the deviation between the actual value and the nominal value of the mover mass, M1-M0=VM, M1 represents the nominal value of the mover mass, VB represents the deviation between the actual value and the nominal value of the viscous friction coefficient, B-B0=VB, B represents the nominal value of the viscous friction coefficient, It represents the sum of nonlinear friction and nonlinear disturbance during motor motion.

[0031] Furthermore, the expression of the parameter model to be identified is:

[0032] y=f(x1)+ε,

[0033] Among them, x1 represents the input of the parameter model to be identified, y represents the output of the parameter model to be identified, ε represents noise.

[0034] Furthermore, the training data and test data of the parameter model to be identified are obtained, including:

[0035] Acquire original input data of the motor dynamics model and original output data of the motor dynamics model, wherein the original input data of the motor dynamics model includes the control input of the motor, and the original output data of the motor dynamics model includes the displacement, velocity and acceleration of the motor;

[0036] Obtaining original input data of a parameter model to be identified and original output data of the parameter model to be identified according to original input data of the motor dynamics model and original output data of the motor dynamics model;

[0037] Using a preset number of original input data of the parameter model to be identified as training input data, and using a preset number of original output data of the parameter model to be identified as training output data, to obtain training data of the parameter model to be identified, wherein the preset number is less than the number of original input data of the parameter model to be identified, and less than the number of original output data of the parameter model to be identified;

[0038] The remaining original input data of the parameter model to be identified are used as test input data, and the remaining original output data of the parameter model to be identified are used as test output data to obtain test data of the parameter model to be identified.

[0039] As another aspect of the present invention, a motor parameter identification system is provided, which is used to implement the above motor parameter identification method, including:

[0040] A motor dynamics model building module is used to build a motor dynamics model that takes into account external disturbance factors and deviations from the nominal values ​​of motor parameters;

[0041] A module for constructing a model of parameters to be identified, used for obtaining a model of parameters to be identified according to a motor dynamics model;

[0042] A data acquisition and processing module, used to obtain training data and test data of the parameter model to be identified, wherein the training data and the test data both include displacement, velocity, acceleration, control input, true value of viscous friction coefficient and true value of mover mass of the motor, the training data includes training input data and training output data, and the test data includes test input data and test output data;

[0043] Gaussian process regression training module, used to perform Gaussian process regression training on training data to obtain the prior distribution of training data;

[0044] A parameter identification module is used to obtain the posterior distribution of the test data based on the prior distribution of the training data, and to perform parameter identification on the parameter model to be identified based on the posterior distribution of the test data to obtain a parameter identification result;

[0045] The updating module is used to obtain a target parameter model according to the parameter identification result, and to update the motor dynamics model according to the target parameter model.

[0046] The present invention constructs a parameter model to be identified based on a motor dynamics model that takes external disturbances and nominal deviations into consideration, establishes a priori distribution of training data of the parameter model to be identified through Gaussian process regression training, and infers a posterior distribution of test output data under given input conditions in combination with the prior distribution. Then, parameter identification is performed on the parameter model to be identified based on the posterior distribution. The obtained target parameter model has statistical significance and strong generalization ability, and can realize motor parameter identification of high-dimensional, small-sample and nonlinear terms with high accuracy, thereby obtaining a motor dynamics model with high accuracy, thereby improving the efficiency and stability of the motor system operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present invention, but do not constitute a limitation of the present invention.

[0048] Figure 1 This is a flow chart of the motor parameter identification method provided by the present invention.

[0049] Figure 2A flowchart for obtaining training data and test data for a parameter model to be identified provided by the present invention.

[0050] Figure 3 A flow chart of obtaining a priori distribution of training data provided by the present invention.

[0051] Figure 4 A flowchart of Gaussian process regression training provided by the present invention.

[0052] Figure 5 A flow chart of obtaining a target parameter model provided by the present invention.

[0053] Figure 6 This is a structural block diagram of the motor parameter identification system provided by the present invention. DETAILED DESCRIPTION

[0054] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0055] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present invention.

[0056] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so as to describe the embodiments of the present invention described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0057] The embodiment of the present invention provides a method for identifying motor parameters. Figure 1 As shown, including:

[0058] S100, constructing a motor dynamics model that takes into account external disturbance factors and deviations from nominal values ​​of motor parameters;

[0059] Specifically, the motor dynamics model constructed in the embodiment of the present invention is expressed as shown in the following formula (1):

[0060]

[0061] Among them, M0 represents the true value of the mover mass, B0 represents the true value of the viscous friction coefficient, x represents the displacement, Indicates speed, represents acceleration, u represents the control input of the motor, VM represents the deviation between the actual value and the nominal value of the mover mass, M1-M0=VM, M1 represents the nominal value of the mover mass, VB represents the deviation between the actual value and the nominal value of the viscous friction coefficient, B-B0=VB, B represents the nominal value of the viscous friction coefficient, It represents the sum of nonlinear friction and nonlinear disturbance during motor motion.

[0062] Among them, D is a nonlinear term, and its parameters need to be identified.

[0063] S200, constructing a parameter model to be identified according to the motor dynamics model;

[0064] Specifically, the embodiment of the present invention constructs a parameter model to be identified according to the item D to be identified of the motor dynamics model shown in the above formula (1). The expression of the parameter model to be identified is shown in the following formula (2):

[0065] y=f(x1)+ε(2),

[0066] Among them, x1 represents the input of the parameter model to be identified, y represents the output of the parameter model to be identified, ε represents the noise, That is, the noise has a mean of zero and a variance of The normal distribution of .

[0067] Compared with the ideal second-order dynamics model of the motor, the motor dynamics model constructed in the embodiment of the present invention takes into account the nonlinear friction and nonlinear disturbance factors in the movement of the motor, as well as the deviation between the nominal value and the true value of the motor parameters. Therefore, the parameter model to be identified based on this construction is more practical, accurate and reasonable.

[0068] In addition, the expression of the parameter model to be identified provided in the embodiment of the present invention takes noise factors into consideration and can inherently handle measurement noise without pre-defining specific structures or parameters. Instead, it can adaptively learn to extract noise from data, thereby reducing noise interference, improving the generalization ability of the parameter model to be identified, quantifying the uncertainty of prediction results, reducing the dependence of the parameter model to be identified on prior assumptions and enhancing the robustness of the parameter model to be identified in a dynamic environment, thereby optimizing the performance of the motor dynamics model.

[0069] S300, obtaining training data and test data of the parameter model to be identified, wherein the training data and the test data both include displacement, velocity, acceleration, control input, true value of viscous friction coefficient and true value of mover mass of the motor, the training data includes training input data and training output data, and the test data includes test input data and test output data;

[0070] Specifically, in the embodiment of the present invention, N groups of training data are collected And M group test data in, represents the i-th data in the training input data, Represents the i-th data in the training output data, Represents the i-th data in the test input data, Represents the jth data in the test output data.

[0071] S400, performing Gaussian process regression training on the training data to obtain a priori distribution of the training data;

[0072] The embodiment of the present invention performs training data Perform Gaussian process regression training to obtain the prior distribution between data points in the training input data.

[0073] S500, obtaining a posterior distribution of the test data based on the prior distribution of the training data, and performing parameter identification on the parameter model to be identified based on the posterior distribution of the test data to obtain a parameter identification result;

[0074] Specifically, the embodiment of the present invention obtains the posterior distribution of the test data based on the prior distribution of the training data, and performs parameter identification on the parameter model to be identified based on the posterior distribution of the test data. The obtained parameter identification result can effectively correct the error between the true value and the nominal value of the parameter.

[0075] It should be understood that since the Gaussian process is a probabilistic model, it assumes that each point of the data is sampled from a multivariate Gaussian distribution. This assumption enables the Gaussian process to calculate the posterior distribution of any unobserved point based on the prior distribution of the training data, rather than just a point estimate. This means that when making predictions, not only can we get the predicted value, but also the uncertainty of the prediction, that is, the variance, which provides a natural statistical interpretation for the model. Therefore, the Gaussian process constructs a prior distribution of the training data, combined with the input conditions, to obtain the posterior distribution of the test data, which itself has statistical significance.

[0076] S600: Obtain a target parameter model according to the parameter identification result, and update the motor dynamics model according to the target parameter model.

[0077] Specifically, the embodiment of the present invention obtains the target parameter model according to the parameter identification result. Based on the target parameter model The nonlinear term D in the motor dynamics model is updated to obtain an updated motor dynamics model, thereby effectively resolving the error between the true value and the nominal value of the parameters in the motor dynamics model.

[0078] The embodiment of the present invention constructs a parameter model to be identified based on a motor dynamics model that takes external disturbances and nominal deviations into consideration, establishes a priori distribution of training data of the parameter model to be identified through Gaussian process regression training, and infers a posterior distribution of test output data under given input conditions in combination with the prior distribution. On this basis, the test output data itself has statistical significance, and parameter identification is performed on the parameter model to be identified based on the posterior distribution. The obtained target parameter model also has statistical significance, and the generalization ability of the parameter model to be identified can be improved by setting the kernel function of the Gaussian process core, and the motor parameter identification of high dimension, small sample and nonlinear terms can be realized with high accuracy, thereby obtaining a motor dynamics model with high accuracy and improving the efficiency and stability of the motor system operation.

[0079] Furthermore, if Figure 2 As shown, the training data and test data of the parameter model to be identified are obtained, including:

[0080] S310, acquiring original input data of the motor dynamics model and original output data of the motor dynamics model, wherein the original input data of the motor dynamics model includes the control input of the motor, and the original output data of the motor dynamics model includes the displacement, velocity and acceleration of the motor;

[0081] Specifically, this embodiment adopts proportional plus derivative control (PD control for short), and collects the control input u of the motor, the displacement x of the motor, and the speed x of the motor through the controller and the sensor. and acceleration in, e represents the tracking error, e = x d -x,x d represents the expected displacement, K p Represents proportional gain, K d Indicates the integral gain.

[0082] S320, obtaining original input data of the parameter model to be identified and original output data of the parameter model to be identified according to the original input data of the motor dynamics model and the original output data of the motor dynamics model;

[0083] Specifically, the expression for obtaining the original input data of the parameter model to be identified based on the original input data of the motor dynamics model and the original output data of the motor dynamics model is: The expression of the original output data of the parameter model to be identified is:

[0084] S330, using a preset number of original input data of the parameter model to be identified as training input data, and using a preset number of original output data of the parameter model to be identified as training output data, to obtain training data of the parameter model to be identified, wherein the preset number is less than the number of original input data of the parameter model to be identified, and less than the number of original output data of the parameter model to be identified;

[0085] Specifically, the embodiment of the present invention collects N+M groups of original input data of the parameter model to be identified, and N+M groups of original output data of the parameter model to be identified, and N groups are used as training data. in, represents the i-th data in the training input data, Represents the i-th data in the training output data.

[0086] S340: Use the remaining original input data of the parameter model to be identified as test input data, and use the remaining original output data of the parameter model to be identified as test output data, to obtain test data of the parameter model to be identified.

[0087] Group M is used as test data Represents the i-th data in the test input data, Represents the jth data in the test output data.

[0088] Furthermore, if Figure 3 As shown, Gaussian process regression training is performed on the training data to obtain the prior distribution of the training data, including:

[0089] S410, forming a training input matrix according to the training input data of the parameter model to be identified;

[0090] Specifically, the embodiment of the present invention forms a training input matrix according to the training input data of the parameter model to be identified:

[0091] S420, forming a training output matrix according to the training output data of the parameter model to be identified;

[0092] Specifically, the embodiment of the present invention forms a training output matrix y according to the training output data of the parameter model to be identified. f =[y f ,···,y f (N)].

[0093] S430, defining a Gaussian process according to a preset mean function and a preset kernel function;

[0094] Specifically, the Gaussian process is shown in the following formula (3):

[0095] f(x):GP(m(x),k(l,l'))(3),

[0096] Among them, m(x) represents the preset mean function, k(l,l') represents the preset kernel function, l represents the first input of the preset kernel function, and l' represents the second input of the preset kernel function. The first input and the second input can be the same or different. It should be understood that the kernel function, also known as the covariance function, is the core of the Gaussian process and can flexibly adapt to the characteristics of the data, such as linearity, nonlinearity, and periodicity. It is mainly used to describe the correlation between different data points and determines the smoothness of f(x) and signal variance. By properly selecting the kernel function, the Gaussian process can capture the complex patterns of data from small samples, thereby improving the generalization ability of the model, especially when facing nonlinear and high-dimensional data.

[0097] In addition, Gaussian process regression maps points in the input space to a high-dimensional feature space through a kernel function, and automatically learns the correlation between points in the input space based on the training data. Therefore, even if the input dimension is high, the Gaussian process can still be processed through effective kernel function design, and there is no need to explicitly specify the complex structure of the parameter space, so that high-dimensional problems can be effectively handled.

[0098] And because a common problem in motor parameter identification is limited data samples, especially when the cost of obtaining experimental data is high, in this case, traditional deep learning or complex statistical models are prone to overfitting. Gaussian process regression can effectively model the data and infer parameters through the reasonable design of prior distribution and kernel function in the case of small samples. The obtained target parameter model can avoid the overfitting problem.

[0099] On the other hand, the operating mechanism of the motor is usually highly nonlinear. The Gaussian process can cope with this complex nonlinear relationship through nonlinear kernel functions. With the help of kernel functions, Gaussian process regression can automatically capture the nonlinear structure in the data without explicitly specifying the nonlinear model form. It is adaptive and flexible.

[0100] S440, performing Gaussian process regression training on the training input matrix and the training output matrix according to the Gaussian process, and calculating the covariance matrix between each data point in the training input matrix.

[0101] Specifically, the embodiment of the present invention performs Gaussian process regression training on the training input matrix and the training output matrix according to the Gaussian process to obtain the covariance matrix between the data points in the training input matrix, and the covariance matrix can reflect the similarity and correlation between the data points in the training input data.

[0102] Furthermore, the preset mean function is a zero mean function.

[0103] Mean functions mainly include constant mean function, linear mean function, polynomial mean function, etc. The core part of Gaussian process is covariance function (kernel function), which determines the correlation between data points, while mean function is usually only used to express the initial guess of the target function. In the embodiment of the present invention, the sampling selection of zero mean function here can simplify the model and entrust the main modeling task to kernel function and observation data. Through the zero mean assumption, kernel function can flexibly capture the trend and change in data without being affected by mean function.

[0104] Furthermore, if Figure 4 As shown, according to the Gaussian process, the training input matrix and the training output matrix are subjected to Gaussian process regression training, and the covariance matrix between the data points in the training input matrix is ​​calculated, including:

[0105] S441, parameterizing the preset kernel function and selecting preset hyperparameters;

[0106] Specifically, the preset kernel function provided by the embodiment of the present invention is a radial basis kernel function, as shown in the following formula (4):

[0107]

[0108] Among them, {σ,σ f} represents the preset hyperparameters, σ and σ f Both are parameters that control the kernel function. Specifically, σ is the scale parameter that controls the kernel function, which is used to measure the extent to which the similarity between data points extends in the input space. The scale parameter determines the model's sensitivity to features of different scales. A larger scale factor indicates that the model assumes that the input features change smoothly, while a smaller scale factor indicates that the model is more sensitive to rapid changes in the input features. If the selected scale factor is too large, the model may be too smooth and unable to capture subtle features; if it is too small, the model may overfit, that is, be too sensitive to noise in the data.

[0109] It is possible to control the amplitude variance of the kernel function, thereby controlling the overall amplitude variation of the model output. Specifically, if the amplitude variance is large, the model allows greater fluctuations, but too large a variance may lead to overfitting; if the amplitude variance is small, the fluctuations allowed by the model are more conservative, but too small a variance may lead to underfitting.

[0110] S442, maximizing the probability of each data in the training output matrix appearing under the preset hyperparameters by maximizing the marginal likelihood function, and obtaining the target hyperparameters;

[0111] Specifically, the expression for maximizing the marginal likelihood function is defined as shown in the following formula (5):

[0112]

[0113] Among them, y f represents the training output matrix, X represents the training input matrix, ψ=[σ,σ f ,σ no ], T represents the transpose of the matrix, K represents the covariance matrix, I represents the identity matrix, represents the noise variance;

[0114] It should be understood that the noise variance Mainly used to control the noise level in training data.

[0115] S443. Calculate the covariance matrix between each data point in the training input matrix according to the preset kernel function and the target hyperparameter.

[0116] Specifically, the expression of the covariance matrix between each data point in the training input matrix is ​​shown in the following formula (6):

[0117]

[0118] Furthermore, if Figure 5 As shown, the posterior distribution of the test data is obtained based on the prior distribution of the training data, and the parameter identification is performed on the parameter model to be identified based on the posterior distribution of the test data to obtain the parameter identification result, including:

[0119] S510, obtaining a joint Gaussian distribution of training input data, training output data, and test input data according to a priori distribution of training data;

[0120] Specifically, the joint Gaussian distribution of the training input data, the training output data, and the test input data is as shown in the following formula (7):

[0121]

[0122] in, Represents the test input data, y f represents the training output matrix, K represents the covariance matrix between the data points in the training input matrix, K * represents the covariance matrix between the training input data and the test input data, T represents the transpose of the matrix, K **represents the covariance matrix between the data points of the test input data, represents the noise variance and I represents the identity matrix.

[0123] k * The specific expression of is shown in the following formula (8):

[0124]

[0125] k ** The specific expression of is shown in the following formula (9):

[0126]

[0127] S520, calculating the posterior mean function of the test output data according to the joint Gaussian distribution;

[0128] Specifically, the expression of the posterior mean function is shown in the following formula (10):

[0129]

[0130] The posterior mean function is the parameter identification result in this embodiment. In this embodiment, the posterior mean function is used as the target parameter model. And according to the updated motor dynamics model expression, D is

[0131] Another embodiment of the present invention provides a motor parameter identification system 10, such as Figure 6 As shown, including:

[0132] A motor dynamics model building module 11 is used to build a motor dynamics model that takes into account external disturbance factors and deviations from nominal values ​​of motor parameters;

[0133] Specifically, in the embodiment of the present invention, the expression of the motor dynamics model is shown in formula (1):

[0134]

[0135] Among them, M0 represents the true value of the mover mass, B0 represents the true value of the viscous friction coefficient, x represents the displacement, Indicates speed, represents acceleration, u represents the control input of the motor, VM represents the deviation between the actual value and the nominal value of the mover mass, M1-M0=VM, M1 represents the nominal value of the mover mass, VB represents the deviation between the actual value and the nominal value of the viscous friction coefficient, B-B0=VB, B represents the nominal value of the viscous friction coefficient, It represents the sum of nonlinear friction and nonlinear disturbance during motor motion.

[0136] Among them, D is a nonlinear term, and its parameters need to be identified.

[0137] The to-be-identified parameter model construction module 12 is used to obtain the to-be-identified parameter model according to the motor dynamics model;

[0138] Specifically, in the embodiment of the present invention, the expression of the parameter model to be identified is shown in formula (2):

[0139] y=f(x1)+ε(2),

[0140] Among them, x1 represents the input of the parameter model to be identified, y represents the output of the parameter model to be identified, ε represents noise.

[0141] Compared with the ideal second-order dynamics model of the motor, the motor dynamics model constructed in the embodiment of the present invention takes into account the nonlinear friction and nonlinear disturbance factors in the motor movement process, as well as the deviation between the system nominal parameters and the true values. Therefore, the parameter model to be identified based on this construction is more accurate.

[0142] In addition, the expression of the parameter model to be identified provided by the embodiment of the present invention takes the noise factor into consideration and can construct an inherent processing measurement noise without pre-defining a specific structure or parameter, but can adaptively learn to extract noise from the data.

[0143] The data acquisition and processing module 13 is used to obtain training data and test data of the parameter model to be identified, wherein the training data and the test data both include the displacement, velocity, acceleration, control input, true value of the viscous friction coefficient and true value of the mover mass of the motor, the training data includes training input data and training output data, and the test data includes test input data and test output data;

[0144] Specifically, in the embodiment of the present invention, N groups of training data are collected And M group test data in, represents the i-th data in the training input data, Represents the i-th data in the training output data, Represents the i-th data in the test input data, Represents the jth data in the test output data.

[0145] A Gaussian process regression training module 14 is used to perform Gaussian process regression training on the training data to obtain a priori distribution of the training data;

[0146] Specifically, the embodiment of the present invention performs training data Perform Gaussian process regression training to obtain the prior distribution between data points in the training input data.

[0147] The parameter identification module 15 is used to obtain the posterior distribution of the test data based on the prior distribution of the training data, and to perform parameter identification on the parameter model to be identified based on the posterior distribution of the test data to obtain the parameter identification result;

[0148] Specifically, the embodiment of the present invention obtains the posterior distribution of the test data based on the prior distribution of the training data, and performs parameter identification on the parameter model to be identified based on the posterior distribution of the test data. The obtained parameter identification result can effectively correct the error between the true value and the nominal value of the parameter.

[0149] The updating module 16 is used to obtain a target parameter model according to the parameter identification result, and update the motor dynamics model according to the target parameter model.

[0150] Specifically, the embodiment of the present invention obtains the target parameter model according to the parameter identification result. Based on the target parameter model The nonlinear term D in the motor dynamics model is updated to obtain an updated motor dynamics model, thereby effectively resolving the error between the true value and the nominal value of the parameters in the motor dynamics model.

[0151] The embodiment of the present invention constructs a parameter model to be identified based on a motor dynamics model that takes external disturbances and nominal deviations into consideration, establishes a priori distribution of training data of the parameter model to be identified through Gaussian process regression training, and infers the posterior distribution of test output data under given input conditions in combination with the prior distribution. Then, parameter identification is performed on the parameter model to be identified based on the posterior distribution. The obtained target parameter model has statistical significance and strong generalization ability, and can realize motor parameter identification of high dimensions, small samples and nonlinear terms with high accuracy, and obtain a motor dynamics model with high accuracy, thereby improving the efficiency and stability of the motor system operation.

[0152] The specific working principle of the motor parameter identification system in the embodiment of the present invention can be referred to the above description of the motor parameter identification method, which will not be repeated here.

[0153] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for identifying motor parameters, characterized in that: include: Construct a motor dynamics model that takes into account external disturbance factors and deviations from the nominal values ​​of motor parameters; Constructing a parameter model to be identified according to the motor dynamics model; Acquire training data and test data of the parameter model to be identified, wherein the training data and the test data both include displacement, velocity, acceleration, control input, true value of viscous friction coefficient and true value of mover mass of the motor, the training data include training input data and training output data, and the test data include test input data and test output data; Performing Gaussian process regression training on the training data to obtain a prior distribution of the training data; Obtaining the posterior distribution of the test data based on the prior distribution of the training data, and performing parameter identification on the parameter model to be identified based on the posterior distribution of the test data to obtain a parameter identification result; A target parameter model is obtained according to the parameter identification result, and the motor dynamics model is updated according to the target parameter model.

2. The method according to claim 1, characterized in that The performing Gaussian process regression training on the training data to obtain a prior distribution of the training data includes: Forming a training input matrix according to the training input data of the parameter model to be identified; Forming a training output matrix according to the training output data of the parameter model to be identified; Define a Gaussian process according to a preset mean function and a preset kernel function; Gaussian process regression training is performed on the training input matrix and the training output matrix according to a Gaussian process, and a covariance matrix between data points in the training input matrix is ​​calculated.

3. The method according to claim 2, characterized in that The preset mean function is a zero mean function.

4. The method according to claim 2, characterized in that: The step of performing Gaussian process regression training on the training input matrix and the training output matrix according to the Gaussian process, and calculating the covariance matrix between each data point in the training input matrix, comprises: Parameterizing the preset kernel function and selecting preset hyperparameters; Maximizing the probability of each data in the training output matrix appearing under the preset hyperparameters by maximizing the marginal likelihood function, thereby obtaining the target hyperparameters; The covariance matrix between the data points in the training input matrix is ​​calculated according to the preset kernel function and the target hyperparameters.

5. The method according to claim 4, characterized in that The expression of the preset kernel function is: Among them, {σ,σ f } represents a preset hyperparameter, l represents a first input of the preset kernel function, l' represents a second input of the preset kernel function, and the first input and the second input may be the same or different.

6. The method according to claim 1, characterized in that The obtaining of the posterior distribution of the test data based on the prior distribution of the training data, and performing parameter identification on the parameter model to be identified based on the posterior distribution of the test data to obtain a parameter identification result includes: Obtaining a joint Gaussian distribution of the training input data, the training output data, and the test input data according to a prior distribution of the training data; The posterior mean function of the test output data is calculated according to the joint Gaussian distribution.

7. The method according to claim 1, characterized in that The expression of the motor dynamics model is: Among them, M0 represents the true value of the mover mass, B0 represents the true value of the viscous friction coefficient, x represents the displacement, Indicates speed, represents acceleration, u represents the control input of the motor, VM represents the deviation between the actual value and the nominal value of the mover mass, M1-M0=VM, M1 represents the nominal value of the mover mass, VB represents the deviation between the actual value and the nominal value of the viscous friction coefficient, B-B0=VB, B represents the nominal value of the viscous friction coefficient, It represents the sum of nonlinear friction and nonlinear disturbance during motor motion.

8. The method according to claim 7, characterized in that The expression of the parameter model to be identified is: y=f(x1)+ε, Among them, x1 represents the input of the parameter model to be identified, y represents the output of the parameter model to be identified, ε represents noise.

9. The method according to claim 8, characterized in that The step of obtaining the training data and the test data of the parameter model to be identified includes: Acquire original input data of the motor dynamics model and original output data of the motor dynamics model, wherein the original input data of the motor dynamics model includes the control input of the motor, and the original output data of the motor dynamics model includes the displacement, velocity and acceleration of the motor; Obtaining original input data of a parameter model to be identified and original output data of the parameter model to be identified according to original input data of the motor dynamics model and original output data of the motor dynamics model; Using a preset number of original input data of the parameter model to be identified as training input data, and using a preset number of original output data of the parameter model to be identified as training output data, to obtain training data of the parameter model to be identified, wherein the preset number is less than the number of original input data of the parameter model to be identified, and less than the number of original output data of the parameter model to be identified; The remaining original input data of the parameter model to be identified is used as test input data, and the remaining original output data of the parameter model to be identified is used as test output data to obtain test data of the parameter model to be identified.

10. A motor parameter identification system, used to implement the motor parameter identification method according to any one of claims 1 to 9, characterized in that: include: A motor dynamics model building module is used to build a motor dynamics model that takes into account external disturbance factors and deviations from the nominal values ​​of motor parameters; A module for constructing a parameter model to be identified, used for obtaining a parameter model to be identified according to the motor dynamics model; A data acquisition and processing module, used to obtain training data and test data of the parameter model to be identified, wherein the training data and the test data both include displacement, velocity, acceleration, control input, true value of viscous friction coefficient and true value of mover mass of the motor, the training data includes training input data and training output data, and the test data includes test input data and test output data; A Gaussian process regression training module, used to perform Gaussian process regression training on the training data to obtain a prior distribution of the training data; A parameter identification module, used to obtain the posterior distribution of the test data based on the prior distribution of the training data, and perform parameter identification on the parameter model to be identified based on the posterior distribution of the test data to obtain a parameter identification result; An updating module is used to obtain a target parameter model according to the parameter identification result, and to update the motor dynamics model according to the target parameter model.

Images