A method for parameter identification of permanent magnet linear synchronous motor based on fuzzy width learning

Abstract

The present invention discloses a permanent magnet synchronous linear motor parameter identification method based on fuzzy width learning, comprising the following steps: S1, designing a control algorithm based on model prediction according to control requirements, and completing PMLSM double closed-loop control; S2, giving an excitation signal, and collecting the electrical signal generated during the operation of the PMSM to construct a training sample set and a test sample set, and using a normalization algorithm to pre-process the sample data; S3, using the acquired training sample set to construct a motor parameter identification model based on FBLS, and using the test sample set and the error evaluation index RMSE to determine the optimal hyperparameters of the FBLS model; S4, using the trained FBLS-based parameter identification model to identify the parameters of the PMLSM online. The present invention solves the problem that most of the permanent magnet synchronous motor parameter identification methods of the traditional neural network algorithm use linear models for identification, resulting in poor accuracy of the identification parameters.

Description

Technical Field

[0001] The present invention belongs to the technical field of motor parameter identification, and in particular, relates to a permanent magnet linear synchronous motor parameter identification method based on fuzzy width learning. Background Art

[0002] With the advancement of science and technology and the needs of production and life, motors are widely present in our daily lives. In order to meet different needs, motors have many different structures. Compared with other motors, permanent magnet synchronous motors have the advantages of simple structure, high energy density, simple maintenance and low cost, so they are widely used in various fields. The premise of using motors for work is to use motors with appropriate parameters for target needs. Generally, motor manufacturers will mark the relevant parameters of permanent magnet synchronous motors in the form of nameplates. It is usually believed that these factory parameters remain unchanged. However, since the motor runs in a specific working environment for a long time, the surrounding environment will inevitably have a certain impact on the motor itself, which will cause some factory parameters of the motor to change. Therefore, certain means are needed to identify certain parameters of the motor. Effective parameter identification can not only ensure the stable operation of the system, but also improve the control accuracy and response speed. Therefore, efficient and reliable parameter identification of the motor is of great significance.

[0003] The defect in the prior art is that most of the permanent magnet synchronous motor parameter identification methods of traditional neural network algorithms use linear models for identification, but in practical applications, the parameters of permanent magnet synchronous motors are nonlinear, resulting in the need to improve the accuracy of the identification parameters. Summary of the invention

[0004] Most permanent magnet synchronous motor parameter identification methods based on traditional neural network algorithms use linear models for identification, which leads to the problem of poor accuracy of identification parameters. The present invention provides a permanent magnet linear synchronous motor parameter identification method based on fuzzy width learning.

[0005] In order to achieve the above technical objectives, the technical solution adopted by the present invention is as follows:

[0006] A method for identifying parameters of a permanent magnet linear synchronous motor based on fuzzy width learning comprises the following steps:

[0007] S1. According to the control requirements, design the control algorithm based on model prediction and complete the PMLSM double closed-loop control;

[0008] S2, given an excitation signal, and collecting the electrical signal generated during the operation of the PMSM to construct a training sample set and a test sample set, and use a normalization algorithm to preprocess the sample data;

[0009] S3, using the acquired training sample set to build a motor parameter identification model based on fuzzy width learning (FBLS), and using the test sample set and the error evaluation index RMSE to determine the optimal hyperparameters of the FBLS model;

[0010] S4. Use the trained FBLS-based parameter identification model to online identify the parameters of the PMLSM.

[0011] Furthermore, the control algorithm based on model prediction is designed in step S1 to predict the state of the next step according to the state at the current operating moment, and determine the switching state of the inverter at the next moment according to the cost function.

[0012] Furthermore, in step S2, a training sample set X and a test sample set X′ are constructed, and the expression of the training sample set X is:

[0013] X=[i d ,i q ,u d ,u q ,ω e |R s ,L d ,L q ,ψ f ]

[0014] Among them, i q is the q-axis current, is the q-axis reference current, i d is the d-axis current, u d is the d-axis reference voltage, u q is the d-axis reference voltage, ω e is the electrical angle obtained by the sensor, R s is the internal resistance of PMLSM, L d is the d-axis component of the motor inductance, L q is the q-axis component of the motor inductance, ψ f Permanent magnet flux. d ,i q ,u d ,u q ,ω e ] is the input of the training sample set X, [R s ,L d ,L q ,ψ f ]Output of training sample set X.

[0015] Furthermore, in step S2, a normalization algorithm is used to preprocess the training data, and its expression is:

[0016]

[0017] Among them, xmean is the average value of the data in the training sample set, x max is the maximum value of the data in the training sample set, x min is the minimum value of the data in the training sample set.

[0018] Furthermore, the fuzzy width learning system model in step S3 uses a set of TS fuzzy subsystems to replace the characteristic nodes of the width learning system, and finally forms a hybrid fuzzy neural network.

[0019] Further, in step S3, a fuzzy width learning system model is constructed, which includes a fuzzy subsystem and an enhancement node;

[0020] S301, for the fuzzy subsystem, let the sth sample x in the training sample set X s Generate the i-th fuzzy subsystem Z si , construct the i-th fuzzy subsystem set Z of all training samples X i , and construct n fuzzy subsystem set matrix Z n :

[0021]

[0022]

[0023]

[0024] in, is the weight of each fuzzy rule, that is, is the firepower intensity of the kth fuzzy rule in the i-th fuzzy subsystem, corresponds to the fuzzy set The Gaussian membership function of Among them, they are and The width and center of the Gaussian membership function are randomly generated respectively; is a first-order TS fuzzy system, that is

[0025]

[0026] S302: for the enhanced node, according to the fuzzy subsystem set matrix Z n Generate enhanced node H j , and build enhanced node group H m :

[0027] H j =ξ j (Z n W hj +β hj),j=1,2,3,…m

[0028] H m =[H1,H2,…,H m ]

[0029] Among them, ξ j is a nonlinear activation function, W hj and β hj are random weights and biases, respectively.

[0030] S303: For the fuzzy width learning system model, the appropriate output weight W is learned mainly through input data. m The fuzzy width learning system quickly solves the output weight W by using the pseudo-inverse algorithm m :

[0031] W=(DΩ,H m ) + Y

[0032] Among them, (DΩ,H m ) + =((DΩ,H m ) T (DΩ,H m )) -1 (DΩ,H m ) T , I is the identity matrix and Y is the output of the training set.

[0033] S304: The fuzzy width learning system includes feature nodes and enhancement nodes, and constitutes a hidden layer of the width learning system, and outputs weights W m Get the final output result

[0034]

[0035] (DΩ,H m )W

[0036] Among them, F n is the aggregate output of n fuzzy subsystems,

[0037] Ω is the blur intensity The matrix composed of Δ is composed of the parameters to be calculated later

[0038] S305. For the fuzzy width learning system, use MAPE (mean absolute percentage error) to evaluate the accuracy:

[0039] MAPE can not only measure the model bias, but also measure the quality of a model's prediction results. The calculation formula is as follows:

[0040]

[0041] In the formula, represents the true value, y i represents the predicted value, and n represents the number of samples. When the MAPE value is 0%, it means that the prediction model is a perfect model, and when the MAPE value is greater than 100%, it means that the prediction model is a poor model.

[0042] Furthermore, the step S4 described in which the trained fuzzy width learning model is substituted into the permanent magnet synchronous linear motor system parameter identification to perform a simulation test specifically includes:

[0043] The trained fuzzy width learning model is added to the permanent magnet synchronous linear motor control system to output various parameters of the permanent magnet synchronous motor with unknown parameters. The specific parameters that can be output include: permanent magnet synchronous motor line resistance R s , d-axis inductance L d , q-axis inductance L q and the magnetic flux ψ f .

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The fuzzy width learning algorithm is used to complete the permanent magnet synchronous motor parameter identification. Compared with the traditional parameter identification algorithm, its design does not require the mathematical model of the control system, nor does it require the necessary professional knowledge, and the model building process is simple;

[0046] Introducing fuzzy width learning into the field of motor parameter identification has less computational complexity and higher accuracy than the least squares method.

[0047] The fuzzy width learning algorithm is used to train the parameter identifier. Compared with the traditional neural network algorithm, it has the advantages of high model accuracy and short training time, which greatly improves the accuracy of parameter identification.

[0048] The adopted method can complete the identification of multiple parameters at the same time, with less calculation amount and less calculation burden in the online identification of motor parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a flow chart of a method for parameter identification of a permanent magnet linear synchronous motor based on fuzzy width learning according to the present invention;

[0050] Figure 2 This is a structural diagram of the fuzzy width learning system of the present invention;

[0051] Figure 3 This is a structural block diagram of permanent magnet synchronous motor parameter identification using a fuzzy width learning algorithm according to the present invention. DETAILED DESCRIPTION

[0052] In order to facilitate the understanding of those skilled in the art, the present invention is further described below in conjunction with embodiments and drawings. The contents mentioned in the implementation modes are not intended to limit the present invention.

[0053] like Figure 1 As shown, this embodiment provides a permanent magnet linear synchronous motor parameter identification method based on fuzzy width learning, including the following steps: S1. According to the control requirements, a control algorithm based on model prediction is designed, and the PMLSM dual closed-loop control is completed; S2. An excitation signal is given, and the electrical signal generated during the operation of the PMSM is collected to construct a training sample set and a test sample set, and the sample data is preprocessed using a normalization algorithm; S3. A motor parameter identification model based on fuzzy width learning (FBLS) is constructed using the acquired training sample set, and the test sample set and the error evaluation index RMSE are used to determine the optimal hyperparameters of the FBLS model; S4. The parameters of the PMLSM are identified online using the trained FBLS-based parameter identification model.

[0054] like Figure 3 As shown, the control algorithm based on model prediction is designed in step S1 to predict the state of the next step according to the state at the current running moment, and determine the switching state of the inverter at the next moment according to the cost function.

[0055] The next state of the above predicted current is expressed as:

[0056]

[0057] in,

[0058] Among them, i d ,i q They are the dq axis stator current, T s is the sampling time, L is the motor inductance, R s is the motor inductance, ψ s is the permanent magnet flux, ω e is the motor rotor speed, U dc is the DC side voltage of the system, and s(k) represents the switch state at time k.

[0059] The switching state of the inverter at the next moment is determined according to the cost function, and its expression is:

[0060]

[0061] Among them, x dq (k+1|k)=A(k|k)x dq (k∣k)+B(k∣k)s(k∣k)+F(k∣k), Δs(k∣k)=s(k∣k)-s(k-1∣k).

[0062] In step S2, a training sample set X and a test sample set X′ are constructed, and the expression of the training sample set X is:

[0063] X=[i d ,i q ,u d ,u q ,ω e |R s ,L d ,L q ,ψ f ]

[0064] Among them, i q is the q-axis current, is the q-axis reference current, i d is the d-axis current, u d is the d-axis reference voltage, u q is the d-axis reference voltage, ω e is the electrical angle obtained by the sensor, R s is the internal resistance of PMLSM, L d is the d-axis component of the motor inductance, L q is the q-axis component of the motor inductance, ψ f Permanent magnet flux. d ,i q ,u d ,u q ,ω e ] is the input of the training sample set X, [R s ,L d ,L q ,ψ f ]Output of training sample set X.

[0065] like Figure 2 As shown, in step S2, a normalization algorithm is used to preprocess the training data, and its expression is:

[0066]

[0067] Among them, x mean is the average value of the data in the training sample set, x max is the maximum value of the data in the training sample set, x min is the minimum value of the data in the training sample set.

[0068] The fuzzy width learning system model in step S3 uses a set of TS fuzzy subsystems to replace the characteristic nodes of the width learning system, and finally forms a hybrid fuzzy neural network.

[0069] Step S3 constructs a fuzzy width learning system model, which includes a fuzzy subsystem and an enhanced node;

[0070] S301, for the fuzzy subsystem, let the sth sample x in the training sample set X s Generate the i-th fuzzy subsystem Z si , construct the i-th fuzzy subsystem set Z of all training samples X i , and construct n fuzzy subsystem set matrix Z n :

[0071]

[0072]

[0073]

[0074] in, is the weight of each fuzzy rule, that is, is the firepower intensity of the kth fuzzy rule in the i-th fuzzy subsystem, corresponds to the fuzzy set The Gaussian membership function of Among them, they are and The width and center of the Gaussian membership function are randomly generated respectively; is a first-order TS fuzzy system, that is

[0075] S302: for the enhanced node, according to the fuzzy subsystem set matrix Z n Generate enhanced node H j , and build enhanced node group H m :

[0076] H j =ξ j (Z n W hj +β hj ),j=1,2,3,…m

[0077] H m =[H1,H2,…,H m ]

[0078] Among them, ξ j is a nonlinear activation function, W hjand β hj are random weights and biases, respectively.

[0079] S303: For the fuzzy width learning system model, the appropriate output weight W is learned mainly through input data. m The fuzzy width learning system quickly solves the output weight W by using the pseudo-inverse algorithm m :

[0080] W=(DΩ,H m ) + Y

[0081] Among them, (DΩ,H m ) + =((DΩ,H m ) T (DΩ,H m )) -1 (DΩ,H m ) T ,

[0082] I is the identity matrix and Y is the output of the training set.

[0083] S304: The fuzzy width learning system includes feature nodes and enhancement nodes, and constitutes a hidden layer of the width learning system, and outputs weights W m Get the final output result

[0084]

[0085] (DΩ,H m )W

[0086] Among them, F n is the aggregate output of n fuzzy subsystems,

[0087] Ω is the blur intensity The matrix composed of Δ is composed of the parameters to be calculated later

[0088] S305. For the fuzzy width learning system, use MAPE (mean absolute percentage error) to evaluate the accuracy:

[0089] MAPE can not only measure the model bias, but also measure the quality of a model's prediction results. The calculation formula is as follows:

[0090]

[0091] In the formula, represents the true value, yi represents the predicted value, and n represents the number of samples. When the MAPE value is 0%, it means that the prediction model is a perfect model, and when the MAPE value is greater than 100%, it means that the prediction model is a poor model.

[0092] Furthermore, the step S4 described in which the trained fuzzy width learning model is substituted into the permanent magnet synchronous linear motor system parameter identification to perform a simulation test specifically includes:

[0093] The trained fuzzy width learning model is added to the permanent magnet synchronous linear motor control system to output various parameters of the permanent magnet synchronous motor with unknown parameters. The specific parameters that can be output include: permanent magnet synchronous motor line resistance R s , d-axis inductance L d , q-axis inductance L q and the magnetic flux ψ f .

[0094] Compared with the prior art, the present invention has the following beneficial effects:

[0095] The fuzzy width learning algorithm is used to complete the permanent magnet synchronous motor parameter identification. Compared with the traditional parameter identification algorithm, its design does not require the mathematical model of the control system, nor does it require the necessary professional knowledge, and the model building process is simple;

[0096] Introducing fuzzy width learning into the field of motor parameter identification has less computational complexity and higher accuracy than the least squares method.

[0097] The fuzzy width learning algorithm is used to train the parameter identifier. Compared with the traditional neural network algorithm, it has the advantages of high model accuracy and short training time, which greatly improves the accuracy of parameter identification.

[0098] The adopted method can complete the identification of multiple parameters at the same time, with less calculation amount and less calculation burden in the online identification of motor parameters.

[0099] The above is a detailed introduction to a permanent magnet linear synchronous motor parameter identification method based on fuzzy width learning provided by the present application. The description of the specific embodiment is only used to help understand the method and its core idea of ​​the present application. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the scope of protection of the claims of the present application.

Claims

1. A permanent magnet synchronous linear motor parameter identification method based on fuzzy width learning, characterized in that: Includes steps: S1. According to the control requirements, design the control algorithm based on model prediction and complete the PMLSM double closed-loop control; S2, given an excitation signal, and collecting the electrical signal generated during the operation of the PMSM to construct a training sample set and a test sample set, and use a normalization algorithm to preprocess the sample data; S3, using the acquired training sample set to build a motor parameter identification model based on fuzzy width learning, and using the test sample set and the error evaluation index RMSE to determine the optimal hyperparameters of the fuzzy width learning model; S4, using the trained parameter identification model based on fuzzy width learning to identify the parameters of PMLSM online; The fuzzy width learning system model in step S3 uses a set of TS fuzzy subsystems to replace the characteristic nodes of the width learning system, and finally forms a hybrid fuzzy neural network; Step S3 constructs a fuzzy width learning system model, which includes a fuzzy subsystem and an enhanced node; S301, for the fuzzy subsystem, let the sth sample x in the training sample set X s Generate the i-th fuzzy subsystem Z si , construct the i-th fuzzy subsystem set Z of all training samples X i , and construct n fuzzy subsystem set matrix Z n : in, is the weight of each fuzzy rule, that is, is the firepower intensity of the kth fuzzy rule in the i-th fuzzy subsystem, is the fuzzy set The Gaussian membership function of Among them, they are and The width and center of the Gaussian membership function are randomly generated respectively; is a first-order TS fuzzy system, that is S302: for the enhanced node, according to the fuzzy subsystem set matrix Z n Generate enhanced node H j , and build enhanced node group H m : H j =ξ j (Z n W hj +β hj ),j=1,2,3,…m H m =[H1,H2,…,H m ] Among them, ξ j is a nonlinear activation function, W hj and β hj are random weights and biases respectively; S303: For the fuzzy width learning system model, learn a suitable output weight W through input data, and the fuzzy width learning system can quickly solve the output weight W by using a pseudo-inverse algorithm: W=(DΩ,H m ) + Y Among them, (DΩ, H m ) + =((DΩ,H m ) T (DΩ,H m )) -1 (DΩ,H m ) T , I is the identity matrix, and Y is the output of the training set; S304: The fuzzy width learning system includes feature nodes and enhancement nodes, and constitutes a hidden layer of the width learning system, and obtains the final output result by outputting weight W. Among them, F n is the aggregate output of n fuzzy subsystems, Ω is the blur intensity The matrix composed of Δ is composed of the parameters to be calculated later S305: For the fuzzy width learning system, the accuracy is evaluated using MAPE (mean absolute percentage error), and the calculation formula is as follows: In the formula, represents the true value, y i represents the predicted value, n represents the number of samples, When the MAPE value is 0%, it means that the prediction model is a perfect model, and when the MAPE value is greater than 100%, it means that the prediction model is a poor model.

2. According to claim 1, a permanent magnet linear synchronous motor parameter identification method based on fuzzy width learning is characterized in that: In step S1, the control algorithm based on model prediction is designed to predict the state of the next step according to the state at the current operating moment, and determine the switching state of the inverter at the next moment according to the cost function.

3. The method for parameter identification of a permanent magnet linear synchronous motor based on fuzzy width learning according to claim 2 is characterized in that: In step S2, a training sample set X and a test sample set X′ are constructed, and the expression of the training sample set X is: X=[i d ,i q ,u d ,u q ,ω e |R s ,L d ,L q ,ψ f ] Among them, i q is the q-axis current, is the q-axis reference current, i d is the d-axis current, u d is the d-axis reference voltage, u q is the d-axis reference voltage, ω e is the electrical angle obtained by the sensor, R s is the internal resistance of PMLSM, L d is the d-axis component of the motor inductance, L q is the q-axis component of the motor inductance, ψ f The permanent magnet flux [id,iq,ud,uq,ωe] is the input of the training sample set X, and [Rs,Ld,Lq,ψf] is the output of the training sample set X.

4. The method for parameter identification of a permanent magnet linear synchronous motor based on fuzzy width learning according to claim 3 is characterized in that: In step S2, a normalization algorithm is used to preprocess the training data, and its expression is: Among them, x mean is the average value of the data in the training sample set, x max is the maximum value of the data in the training sample set, x min is the minimum value of the data in the training sample set.

5. The method for parameter identification of a permanent magnet linear synchronous motor based on fuzzy width learning according to claim 4 is characterized in that: Step S4 substitutes the trained fuzzy width learning model into the permanent magnet synchronous linear motor system parameter identification to perform a simulation test, which specifically includes: The trained fuzzy width learning model is added to the permanent magnet synchronous linear motor control system to output various parameters of the permanent magnet synchronous motor with unknown parameters. The specific parameters that can be output include: permanent magnet synchronous motor line resistance R s , d-axis inductance L d , q-axis inductance L q and the flux linkage ψ f .

Images