Permanent magnet synchronous motor parameter identification method based on PINN

Abstract

The invention relates to a PINN-based permanent magnet synchronous motor parameter identification method. The invention relates to the technical field of permanent magnet synchronous motor driving control, and the method comprises the steps: collecting motor operation parameters, and obtaining the quadrature-direct axis voltage, current and electrical angular velocity of a motor; a PMSM model is built; a PINN model is built; and performing finite element simulation according to the established model to obtain motor operation data, and identifying motor parameters. Compared with a traditional permanent magnet synchronous motor model, the method is higher in precision, and solves the problems that the control performance of the motor is reduced, the dynamic response is not accurate and the efficiency is low due to the influence of a traditional motor equation under a high-load or complex working condition, so as to guarantee the efficient and stable operation of the motor.

Description

Technical Field

[0001] The invention relates to the technical field of permanent magnet synchronous motor drive control, and is a permanent magnet synchronous motor parameter identification method based on PINN. Background Art

[0002] As electric vehicles become more and more popular, their safety and reliability technologies are receiving increasing attention and becoming an important research direction. Permanent magnet synchronous motors (PMSMs) have become the first choice for electric vehicle power systems due to their advantages such as high power density and efficiency. However, in complex and changing application environments, the accuracy and reliability of motor parameter identification face huge challenges. It not only affects the control performance of the motor, but also has a significant impact on the motor's state monitoring system. Therefore, a highly reliable and accurate motor parameter identification method becomes crucial.

[0003] In actual built-in permanent magnet synchronous motors, there are phenomena such as magnetic saturation and cross-coupling, but traditional motor models do not take these into account. This will affect the fidelity of the motor model and the control strategy. At present, in the processing of motor nonlinear parameters, the motor is mainly modeled by high-dimensional tables and fitting functions. In general, although the high-dimensional table method has high modeling accuracy, with the increase of table dimensions and the number of tables, the computing time and table storage space consumed by the interpolation operation increase significantly. Although it still has obvious advantages over the finite element model, it still cannot be used directly in embedded systems. In comparison, the fitting function has unique advantages in establishing motor models and has broader prospects in practical applications.

[0004] At present, the parameter identification of permanent magnet synchronous motor is mainly divided into offline identification and online identification. Online identification mainly includes recursive least squares method, extended Kalmar filter method, model reference adaptive method and some intelligent optimization algorithms. The main problem of online identification is how to solve the problem of lack of rank in the equation. The most commonly used solutions include current injection method or fixing the two electrical parameters in the equation at their nominal values. Offline identification methods include particle swarm algorithm, genetic algorithm, ant colony algorithm and neural network. Compared with traditional intelligent optimization algorithms, neural networks have unique advantages in the field of parameter identification due to their powerful data fitting ability and model plasticity. So far, the concept of physical constrained neural network (PINN) is gaining widespread attention. Due to the addition of physical information to the neural network, the physically constrained neural network has great advantages in accuracy and robustness. At present, many scholars have applied PINN to the field of electrical engineering and revealed the great potential of this method in parameter estimation. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a permanent magnet synchronous motor parameter identification method based on PINN, which has strong accuracy and robustness. Firstly, a high-precision permanent magnet synchronous motor model is established by a polynomial fitting method, and then a PINN motor parameter identification model driven by mechanism and data hybrid is built, and the applicability of the present invention is verified by simulation.

[0006] The present invention provides the following technical solutions:

[0007] A permanent magnet synchronous motor parameter identification method based on PINN, the method comprising the following steps:

[0008] Step 1: Collect the motor operating parameters to obtain the motor AC and DC axis voltage, current and electrical angular velocity;

[0009] Step 2: Build a PMSM model;

[0010] Step 3: Build the PINN model;

[0011] Step 4: Perform finite element simulation based on the constructed model to obtain the motor operation data and identify the motor parameters.

[0012] Preferably, the step 2 is specifically:

[0013] The motor voltage equation for motor nonlinearity is expressed as:

[0014]

[0015] The relationship between inductance and flux in the motor voltage equation is:

[0016]

[0017] Among them, L dd , L qq is the incremental inductance of the d-axis and q-axis, L dq , L qd is the incremental mutual inductance of the d-axis and q-axis;

[0018]

[0019] Polynomial fitting is performed on the AC and DC axis flux and current of the motor to determine the flux parameter equation form and weight coefficient, and the PMSM model is established as follows:

[0020]

[0021] Among them, a0, a1, a2, b0, c0, c1 are polynomial coefficients.

[0022] Preferably, the PINN model includes a data layer and a physical layer:

[0023] The first half of the model is the data layer of the PINN model, which converts the i d 、i q 、u d 、u q ,ω e The sampling interval of the data is used as the input of the data layer, and the number of hidden layers in the neural network and the number of neurons in each layer are configured according to the complexity of the data;

[0024] The input of the model is obtained by fitting the AC and DC axis flux through the neural network. The motor flux fitted by the neural network is also used as the input of the second half of the model. This layer is also the connection layer of the model.

[0025] The AC and DC axis flux of the motor is used as the input of the subsequent physical layer, and the formula constructed in the physical layer is as follows:

[0026]

[0027] After the output of the second half of the physical layer, the predicted value of the motor AC and DC axis voltage is obtained. The mean square error between the predicted value and the true value is used as the learning error of the PINN model as shown in the following formula:

[0028]

[0029] The learning error is passed step by step through the physical layer and the data layer to update the learning gradient and the internal parameters of the model.

[0030] Preferably, for the processing of differential terms, the backward Euler method is used to solve the problem and then substituted into the formula. In the configuration of the PINN model, the network structure is [6, 100, 100, 100, 100, 2], where the numbers represent the number of neurons in each layer of the neural network.

[0031] Preferably, the step 4 is specifically:

[0032] 200 consecutive sampling points are selected as the input of the PINN parameter identification model, most of which are the dynamically changing working points of the motor, and a small number of them are selected when the motor reaches a steady state or all are selected as the dynamic working points of the motor;

[0033] Preferably, in the simulation model, the sampling frequency is set to 10 kHz, the number of iterations is set to 200,000 generations, and the learning rate is set to 0.1.

[0034] Preferably, the method is tested to obtain parameter identification results. By comparing with the simulation data, the parameter identification results of larger magnitude are within 8%, and the parameters a0 and R s The error is within 3 percent.

[0035] A permanent magnet synchronous motor parameter identification system based on PINN, the system comprising:

[0036] A data acquisition module, wherein the data acquisition module collects motor operating parameters to obtain the motor AC and DC axis voltage, current and electrical angular velocity;

[0037] A PMSM model building module, wherein the PMSM model building module builds a PMSM model;

[0038] A PINN model building module, wherein the PINN model building module builds a PINN model;

[0039] An identification module performs finite element simulation based on the constructed model to obtain motor operation data and identify motor parameters.

[0040] A computer-readable storage medium stores a computer program, which is executed by a processor to implement a PINN-based permanent magnet synchronous motor parameter identification method.

[0041] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements a PINN-based permanent magnet synchronous motor parameter identification method when executing the computer program.

[0042] The present invention has the following beneficial effects:

[0043] Compared with the prior art, the present invention has the following advantages:

[0044] The present invention first considers the influence of nonlinear factors such as motor magnetic saturation and cross-coupling, and establishes a new permanent magnet synchronous motor physical model by polynomial fitting. Compared with the traditional permanent magnet synchronous motor model, it has higher accuracy. Under high load or complex working conditions, it solves the problems of traditional motor equations affecting motor control performance, inaccurate dynamic response and low efficiency, so as to ensure efficient and stable operation of the motor.

[0045] In view of the complex working conditions of electric vehicles and the difficulty in obtaining accurate data, this patent establishes a PINN motor parameter identification model by integrating the motor physical model into the neural network. Compared with the traditional parameter identification method, the advantage of the PINN parameter identification model is that it can learn a more generalized model with fewer data samples. At the same time, when used for parameter identification, the PINN model will obtain parameter identification results that meet physical constraints, and the parameter identification process is more stable and reliable.

[0046] Finally, the practicality of the motor model and PINN parameter identification model proposed in this patent was verified through simulation, and the parameter identification accuracy also achieved the expected effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 It shows a schematic diagram of the data layer and the physical layer of the present invention;

[0049] Figure 2 Shown is a schematic diagram of the PINN model driven by the mechanism and data hybrid of the present invention. DETAILED DESCRIPTION

[0050] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments 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 creative work are within the scope of protection of the present invention.

[0051] The present invention is described in detail below in conjunction with specific embodiments. Specific embodiment one:

[0053] according to Figure 1 to Figure 2 As shown, the specific optimization technical solution adopted by the present invention to solve the above technical problems is: the present invention relates to a permanent magnet synchronous motor parameter identification method based on PINN.

[0054] A permanent magnet synchronous motor parameter identification method based on PINN, the method comprising the following steps:

[0055] Step 1: Collect the motor operating parameters to obtain the motor AC and DC axis voltage, current and electrical angular velocity;

[0056] Step 2: Build a PMSM model;

[0057] Step 3: Build the PINN model;

[0058] Step 4: Perform finite element simulation based on the constructed model to obtain the motor operation data and identify the motor parameters. Specific embodiment 2:

[0060] The difference between the second embodiment of the present application and the first embodiment is that:

[0061] The step 2 is specifically as follows:

[0062] The motor voltage equation for motor nonlinearity is expressed as:

[0063]

[0064] The relationship between inductance and flux in the motor voltage equation is:

[0065]

[0066] Among them, L dd , L qq is the incremental inductance of the d-axis and q-axis, L dq , L qd is the incremental mutual inductance of the d-axis and q-axis;

[0067]

[0068] Polynomial fitting is performed on the AC and DC axis flux and current of the motor to determine the flux parameter equation form and weight coefficient, and the PMSM model is established as follows:

[0069]

[0070] Among them, a0, a1, a2, b0, c0, c1 are polynomial coefficients. Specific embodiment three:

[0072] The difference between the third embodiment of the present application and the second embodiment is that:

[0073] The PINN model includes the data layer and the physical layer:

[0074] The first half of the model is the data layer of the PINN model, which converts the i d 、i q 、u d 、u q ,ω e The sampling interval of the data is used as the input of the data layer, and the number of hidden layers in the neural network and the number of neurons in each layer are configured according to the complexity of the data;

[0075] The input of the model is obtained by fitting the AC and DC axis flux through the neural network. The motor flux fitted by the neural network is also used as the input of the second half of the model. This layer is also the connection layer of the model.

[0076] The AC and DC axis flux of the motor is used as the input of the subsequent physical layer, and the formula constructed in the physical layer is as follows:

[0077]

[0078] After the output of the second half of the physical layer, the predicted value of the motor AC and DC axis voltage is obtained. The mean square error between the predicted value and the true value is used as the learning error of the PINN model as shown in the following formula:

[0079]

[0080] The learning error is passed step by step through the physical layer and the data layer to update the learning gradient and the internal parameters of the model. Specific embodiment four:

[0082] The difference between the fourth embodiment of the present application and the third embodiment is that:

[0083] For the processing of differential terms, the backward Euler method is used to solve it and then substituted into the formula. In the configuration of the PINN model, the network structure is [6, 100, 100, 100, 100, 2], where the number represents the number of neurons in each layer of the neural network. Specific embodiment five:

[0085] The difference between the fifth embodiment of the present invention and the fourth embodiment is that:

[0086] The step 4 is specifically as follows:

[0087] 200 consecutive sampling points are selected as the input of the PINN parameter identification model, most of which are the dynamically changing working points of the motor, and a small number of them are selected when the motor reaches a steady state or all are selected as the dynamic working points of the motor; Specific embodiment six:

[0089] The difference between the sixth embodiment of the present invention and the fifth embodiment is that:

[0090] In the simulation model, the sampling frequency is set to 10 kHz, the number of iterations is set to 200,000, and the learning rate is set to 0.1. Specific embodiment seven:

[0092] The difference between the seventh embodiment of the present invention and the sixth embodiment is that:

[0093] The method was tested and the parameter identification results were obtained. By comparing with the simulation data, the identification results of the parameters with larger magnitude were within 8%, and the parameters a0 and R s The error is within 3 percent. Specific embodiment eight:

[0095] The difference between the eighth embodiment of the present invention and the seventh embodiment is that:

[0096] The present invention provides a permanent magnet synchronous motor parameter identification system based on PINN, the system comprising:

[0097] A data acquisition module, wherein the data acquisition module collects motor operating parameters to obtain the motor AC and DC axis voltage, current and electrical angular velocity;

[0098] A PMSM model building module, wherein the PMSM model building module builds a PMSM model;

[0099] A PINN model building module, wherein the PINN model building module builds a PINN model;

[0100] An identification module performs finite element simulation based on the constructed model to obtain motor operation data and identify motor parameters. Specific embodiment nine:

[0102] The difference between the ninth embodiment of the present invention and the eighth embodiment is that:

[0103] The present invention provides a computer-readable storage medium on which a computer program is stored. The program is executed by a processor to implement a permanent magnet synchronous motor parameter identification method based on PINN. Specific embodiment ten:

[0105] The difference between the tenth embodiment of the present invention and the ninth embodiment is that:

[0106] The present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements a PINN-based permanent magnet synchronous motor parameter identification method when executing the computer program. Specific embodiment eleven:

[0108] The difference between the eleventh embodiment of the present invention and the tenth embodiment is that:

[0109] In order to solve the above technical problems, the present invention provides a high-precision motor parameter identification method with strong accuracy and robustness. Firstly, a high-precision permanent magnet synchronous motor model is established by a polynomial fitting method, and then a mechanism and data hybrid driven PINN motor parameter identification model is built, and the applicability of the present invention is verified by simulation.

[0110] Step 1: Data collection: During the simulation and experiment, the dynamic working condition of the motor is first constructed to ensure that the subsequent PINN model has sufficient data support in the parameter identification process. During the simulation process, the motor operating parameters are collected to obtain the motor AC and DC axis voltage, current, and electrical angular velocity.

[0111] Step 2: Build a PMSM model: The traditional permanent magnet synchronous motor model does not take into account the influence of nonlinear factors such as magnetic saturation and cross-coupling. When the motor runs under high load, the control effect and state monitoring of the motor will be seriously affected. Therefore, the present invention proposes a new permanent magnet synchronous motor model established by polynomial fitting.

[0112] The traditional permanent magnet synchronous motor voltage equation is:

[0113]

[0114] Where u d 、u q 、i d 、i q is the voltage and current of the AC and DC axes of the motor; R s , L d , L q , f is the stator resistance, AC and DC axis inductance and permanent magnet flux of the motor; ω e is the electrical angular velocity of the motor.

[0115] The motor voltage equation considering the motor nonlinearity can be expressed as:

[0116]

[0117] The relationship between inductance and flux in the motor voltage equation is:

[0118]

[0119] Where L dd , L qq is the incremental inductance of the d-axis and q-axis, L dq , L qd is the incremental mutual inductance of the d-axis and q-axis. The direct-axis flux of the motor is an even function of the q-axis current, and the quadrature-axis flux is an odd function of the q-axis current. dq , L qd is the incremental mutual inductance of the d-axis and the q-axis, which are equal. So the equation is established as follows:

[0120]

[0121] Polynomial fitting is performed on the AC and DC axis flux and current of the motor to determine the flux parameter equation form and weight coefficient, and the motor flux parameter model is established as follows:

[0122]

[0123] Among them, a0, a1, a2, b0, c0, and c1 are polynomial coefficients. In the design of this patent, there is a constant term in the equation of the motor's direct-axis flux. According to the physical knowledge of the motor, the motor's direct-axis flux includes the motor's permanent magnet flux, which is a value close to a constant. Therefore, in the constructed high-precision motor model, the constant term a0 can also represent the motor's permanent magnet flux.

[0124] Step 3: Build a PINN model: Integrate the motor mechanism knowledge into the neural network construction mechanism and data hybrid driven PINN parameter identification model.

[0125] The PINN parameter identification model mainly consists of two parts: data layer and physical layer. Figure 1 shown.

[0126] (1) The first half of the model is the data layer of the PINN model. First, the i d 、i q 、u d 、u q ,ω e And the sampling interval of the data is used as the input of the data layer, and then the number of hidden layers in the neural network and the number of neurons in each layer are configured according to the complexity of the data.

[0127] (2) The input of the model is obtained by fitting the AC and DC axis magnetic flux through the neural network. At the same time, the motor magnetic flux fitted by this part of the neural network is also used as the input of the second half of the model. Therefore, this layer is also the connection layer of the model.

[0128] (3) The AC and DC axis flux of the motor is used as the input of the subsequent physical layer. The formula constructed in the physical layer is derived from formulas (2), (3), and (5), as shown below:

[0129]

[0130] (4) After the output of the second half of the physical layer, the predicted value of the motor AC and DC axis voltage is obtained. The mean square error between the predicted value and the true value is used as the learning error of the PINN model as shown in the following formula:

[0131]

[0132] The final learning error can be passed through the physical layer and the data layer step by step to update the learning gradient and the internal parameters of the model. This is a typical back-propagation learning method to train the parameters in the model. The parameters updated in the model mainly include the weights and biases between the neural networks in the data layer and the parameters to be identified in the physical layer.

[0133] (5) Regarding the treatment of differential terms in the formula, this patent adopts the backward Euler method to solve the differential terms and then brings them into the formula.

[0134] (6) In terms of the configuration of the PINN model, after multiple tests, its network structure was determined to be [6, 100, 100, 100, 100, 2], where the numbers represent the number of neurons in each layer of the neural network.

[0135] Step 4: Simulation and experimental verification:

[0136] (1) This patent first performs finite element simulation on the motor to obtain the motor operation data. The motor parameters are shown in Table 1

[0137] The parameters of the motor high-precision model are obtained by fitting the AC and DC axis flux and current according to formula (5) as shown in Table 2:

[0138] Table 1 Permanent magnet synchronous motor parameters

[0139]

[0140] Table 2 Parameters of high-precision motor model

[0141]

[0142] (2) Based on the motor parameters obtained, a dual closed-loop vector control model of the motor is built. The schematic diagram of the model is shown in the figure. Figure 2 shown.

[0143] In the setting of the simulation model, the motor operating conditions are set under various dynamic conditions for testing.

[0144] (3) 200 consecutive sampling points are selected as the input of the PINN parameter identification model, most of which are the dynamically changing operating points of the motor, and a small number of them are selected when the motor reaches a steady state or all are selected as the dynamic operating points of the motor.

[0145] (4) This method is to ensure that there is enough data to support the PINN model to update during the parameter identification process, while ensuring that the data demand is smaller than other models. The sampling frequency is set to 10 kHz in the simulation model. In the configuration of the PINN model, the network structure is [6, 100, 100, 100, 100, 2], where the number represents the number of neurons in each layer of the neural network. After multiple tests, the number of iterations was determined to be 200,000 generations, and the learning rate was 0.1.

[0146] (5) Based on the data obtained from the simulation, the results obtained after parameter identification using the PINN model are shown in Table III.

[0147] Table 3 Permanent magnet synchronous motor parameter identification results

[0148]

[0149] After multiple tests, the parameter identification results were obtained. By comparing with the simulation data, the parameter identification results of larger magnitudes were within 8%, and the parameters a0 and R s The error is within 3%, which is a very good result. However, there is still a lot of room for improvement in the recognition results of smaller magnitude.

[0150] The above is only a preferred implementation of a method for identifying parameters of a permanent magnet synchronous motor based on PINN. The protection scope of a method for identifying parameters of a permanent magnet synchronous motor based on PINN is not limited to the above embodiment. All technical solutions under this idea belong to the protection scope of the present invention. It should be pointed out that for those skilled in the art, several improvements and changes without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A permanent magnet synchronous motor parameter identification method based on PINN, characterized in that: The method comprises the following steps: Step 1: Collect the motor operating parameters to obtain the motor AC and DC axis voltage, current and electrical angular velocity; Step 2: Build a PMSM model; Step 3: Build the PINN model; Step 4: Perform finite element simulation based on the constructed model to obtain the motor operation data and identify the motor parameters.

2. The method according to claim 1, characterized in that: The step 2 is specifically as follows: The motor voltage equation for motor nonlinearity is expressed as: The relationship between inductance and flux in the motor voltage equation is: Among them, L dd , L qq is the incremental inductance of the d-axis and q-axis, L dq , L qd is the incremental mutual inductance of the d-axis and q-axis; Polynomial fitting is performed on the AC and DC axis flux and current of the motor to determine the flux parameter equation form and weight coefficient, and the PMSM model is established as follows: Among them, a0, a1, a2, b0, c0, c1 are polynomial coefficients.

3. The method according to claim 2, characterized in that: PI The NN model consists of a data layer and a physical layer: The first half of the model is the data layer of the PINN model, which converts the i d 、i q 、u d 、u q ,ω e The sampling interval of the data is used as the input of the data layer, and the number of hidden layers in the neural network and the number of neurons in each layer are configured according to the complexity of the data; The input of the model is obtained by fitting the AC and DC axis flux through the neural network. The motor flux fitted by the neural network is also used as the input of the second half of the model. This layer is also the connection layer of the model. The AC and DC axis flux of the motor is used as the input of the subsequent physical layer, and the formula constructed in the physical layer is as follows: After the output of the second half of the physical layer, the predicted value of the motor AC and DC axis voltage is obtained. The mean square error between the predicted value and the true value is used as the learning error of the PINN model as shown in the following formula: The learning error is passed step by step through the physical layer and the data layer to update the learning gradient and the internal parameters of the model.

4. The method according to claim 1, characterized in that: For the processing of differential terms, the backward Euler method is used to solve it and then substituted into the formula. In the configuration of the PINN model, the network structure is [6, 100, 100, 100, 100, 2], where the number represents the number of neurons in each layer of the neural network.

5. The method according to claim 4, characterized in that: The step 4 is specifically as follows: 200 consecutive sampling points are selected as the input of the PINN parameter identification model, most of which are the dynamically changing working points of the motor, and a small number of them are selected when the motor reaches a steady state or all are selected as the dynamic working points of the motor.

6. The method according to claim 1, characterized in that: In the simulation model, the sampling frequency is set to 10 kHz, the number of iterations is set to 200,000, and the learning rate is set to 0.

1.

7. The method according to claim 6, characterized in that: The method was tested and the parameter identification results were obtained. By comparing with the simulation data, the identification results of the parameters with larger magnitude were within 8%, and the parameters a0 and R s The error is within 3 percent.

8. A permanent magnet synchronous motor parameter identification system based on PINN, characterized by: The system comprises: A data acquisition module, wherein the data acquisition module collects motor operating parameters to obtain the motor AC and DC axis voltage, current and electrical angular velocity; A PMSM model building module, wherein the PMSM model building module builds a PMSM model; A PINN model building module, wherein the PINN model building module builds a PINN model; An identification module performs finite element simulation based on the constructed model to obtain motor operation data and identify motor parameters.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the method according to claims 1-7.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method of claims 1-7 is implemented.

Images