State estimation and parameter identification method of linear synchronous motor and electronic equipment

Abstract

The invention relates to a state estimation and parameter identification method of a linear synchronous motor and electronic equipment. The method comprises the following steps: acquiring the current density of the linear synchronous motor and abstracted coordinate data; constructing a linear synchronous motor model based on physical information machine learning based on the partial differential equation and boundary conditions of the electromagnetic characteristics of the linear synchronous motor; and inputting the obtained data into a trained linear synchronous motor model, and performing state estimation and parameter identification in real time. Compared with the prior art, the method has the advantages that state estimation and parameter identification are efficiently achieved, and the parameter identification precision is high.

Description

Technical Field

[0001] The invention relates to the field of long-stator linear synchronous motors, and in particular to a state estimation and parameter identification method of a long-stator linear synchronous motor for a maglev train. Background Art

[0002] With the development of modern industry, the capacity of power systems is increasing, and the analysis of their operational stability is becoming increasingly important. Accurate and reasonable synchronous motor models and parameters are of great significance for accurately analyzing and calculating the dynamic behavior of power systems.

[0003] The transportation system is developing rapidly. Maglev technology has the characteristics of no mechanical contact between the vehicle and the rail, and no wheel-rail adhesion. Maglev trains have high running speed, high stability, good comfort, strong climbing ability, and are environmentally friendly, and have broad development prospects. Linear motors are the core equipment of high-speed maglev trains due to their non-adhesion drive, simple structure and reliable performance. Linear synchronous motors are bilaterally excited, have high efficiency and power factor, and are usually suitable for high-speed occasions. In the high-speed maglev train solution, a long primary is usually adopted, that is, the primary is installed on the ground along the line, and is powered and speed-regulated by a high-power converter on the ground; the secondary is installed on the vehicle and is contactlessly excited by a linear harmonic generator.

[0004] The long-stator linear synchronous motor is the core component of the high-speed maglev train traction system and a key component of the suspension system. Establishing a motor model can lay the foundation for the control and steady-state performance analysis of the long-stator linear synchronous motor; the motor parameter identification is the basis for establishing and improving the motor mathematical model. For the linear synchronous motor simulation calculation and control system design, the identification of motor parameters is particularly important.

[0005] The main methods for building linear synchronous motor models are traditional methods such as finite element analysis. This grid-based method faces the "curse of dimensionality", that is, the complexity increases exponentially with the increase of dimension and computing time. Purely data-driven deep learning methods rely on the scale and quality of data, and the generalization ability of the model is not strong.

[0006] How to efficiently realize the state estimation and accurate parameter identification of the long-stator linear synchronous motor of the maglev train has become a technical problem that needs to be solved. Summary of the invention

[0007] The purpose of the present invention is to provide a state estimation and parameter identification method for a linear synchronous motor in order to overcome the defects of the prior art.

[0008] The purpose of the present invention can be achieved by the following technical solutions:

[0009] According to one aspect of the present invention, a method for state estimation and parameter identification of a linear synchronous motor is provided, the method comprising the following steps:

[0010] Step S1, obtaining the current density and abstracted coordinate data of the linear synchronous motor;

[0011] Step S2, constructing a linear synchronous motor model based on physical information machine learning based on the partial differential equations and boundary conditions of the electromagnetic characteristics of the linear synchronous motor;

[0012] Step S3, inputting the data obtained in step S1 into the trained linear synchronous motor model to perform state estimation and parameter identification in real time.

[0013] Preferably, the construction process of the linear synchronous motor model includes:

[0014] The linear synchronous motor magnetic field region Ω satisfies Maxwell's equations, the external boundary is S, L1 and L2 are the boundary lines between the two types of internal media, S, L1, and L2 satisfy the corresponding boundary conditions respectively. The boundary value problem of the linear synchronous motor is equivalent to solving the following set of equations:

[0015] According to Maxwell's equations for static magnetic fields, the following set of equations is established and solved:

[0016]

[0017] Among them, A Z represents the magnetic field z-axis direction vector magnetic potential, x and y represent the geometric coordinates of a point on the motor, n represents the unit normal vector of the boundary, J and J m represents the current density of the medium, μ, μ1 and μ2 represent the magnetic permeability of the medium;

[0018] The solution of the boundary value problem of the linear synchronous motor is approximated by a neural network parameterized by a trainable variable θ. A fully connected neural network with L layers is selected, and the trainable parameters are θ = {W, b}. The fully connected neural network is expressed as:

[0019] x l =σ l (W l ·x l-1 +b l ), l=1,2,…,L

[0020] Among them, x l-1 and x l denote the neural network of the lth layer and the l-1th layer respectively; W and b denote the weight and bias of each layer respectively, W = {W 1 , …W L}, b={b 1 , …b L}; σ l is the activation function.

[0021] Preferably, the step of constructing the linear synchronous motor model comprises:

[0022] Step 1: Define the partial differential equations of the computational domain and its subdomains;

[0023] Step 2: Define the boundaries and the boundary conditions they satisfy, including the outer boundaries, interfaces and the boundary conditions they satisfy;

[0024] Step 3: Define initial conditions;

[0025] Step 4: For parameter identification problems, it is also necessary to define data loss conditions;

[0026] Step 5: Supplement the anchor points of the interface;

[0027] Step 6: Generate a linear synchronous motor model.

[0028] Preferably, the loss function of the linear synchronous motor model for state estimation includes partial differential structure loss, boundary condition loss and initial condition loss, specifically:

[0029] L(θ)=w p L PDE +w b L boundary +w i L initial

[0030] L PDE , L boundary , L initial are partial differential structure loss, boundary condition loss and initial condition loss respectively, w p 、w b and w i is the loss weight.

[0031] Preferably, the loss function of the linear synchronous motor model for parameter identification includes partial differential structure loss, boundary condition loss, initial condition loss and data loss, specifically:

[0032] L(θ)=w p L PDE +w b L boundary +w i L initial +w d L data

[0033] Among them, L PDE , L boundary , L initial and Ldata are partial differential structure loss, boundary condition loss, initial condition loss and data loss respectively; w p 、w b 、w i and w d is the loss weight.

[0034] Preferably, the training set of the linear synchronous motor model includes a residual point set and a boundary point set, wherein the residual point set and the boundary point set are obtained by sampling on the minimum motor unit.

[0035] Preferably, the linear synchronous motor model is a neural network with embedded physical knowledge.

[0036] Preferably, the state estimation includes state estimation of magnetic flux, suspension force and thrust.

[0037] Preferably, the parameter identification includes parameter identification of resistance and coil turns; the input of the parameter identification also includes magnetic flux.

[0038] According to another aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the method described above is implemented when the processor executes the program.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1) The present invention collects process data of the closed-loop operation of the suspension control system of the maglev train, dynamically models the linear synchronous motor of the maglev train based on the physical information machine learning method, realizes fast and efficient estimation of the states that cannot or are difficult to observe during the operation phase, and realizes high-precision parameter identification based on limited data samples; it avoids the problem that the complexity of solving problems using the traditional finite element method grows exponentially with the increase of dimension and time, and overcomes the problems of the poor generalization of the pure data-driven method and the large amount of data required.

[0041] 2) The modeling method of the present invention can be run online in real time and dynamically track the changes in state quantities. The trained model has the ability to predict the state for a period of time in the future, which has a high efficiency advantage in the performance evaluation of the suspension control system of the maglev train.

[0042] 3) The present invention can achieve high-precision parameter estimation based on limited data samples, which helps to establish a reliable dynamic model and improve modeling accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the minimum unit finite element model of the linear synchronous motor of the maglev train in the present invention;

[0044] Figure 2It is a schematic diagram of the dynamic modeling principle of the long stator linear synchronous motor in the present invention;

[0045] Figure 3 It is a schematic diagram of the parameter identification principle of the long stator linear synchronous motor in the present invention;

[0046] Figure 4 It is a schematic flow chart of a method for dynamic modeling of a long-stator linear synchronous motor in the present invention;

[0047] Figure 5 It is a flow chart of the method for identifying parameters of a long-stator linear synchronous motor in the present invention;

[0048] Figure 6 It is a flow chart of the state estimation and parameter identification method of the linear synchronous motor in the present invention;

[0049] In the accompanying drawings, 1 is the primary coil, 2 is the excitation winding, and 3 is the iron core. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are 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 should fall within the scope of protection of the present invention.

[0051] This embodiment relates to a method for state estimation and parameter identification of a linear synchronous motor. Figure 6 , the method comprising:

[0052] Step S1, obtaining data such as current density and abstracted coordinates of the long stator linear synchronous motor;

[0053] Step S2, constructing a linear synchronous motor model based on physical information machine learning;

[0054] Step S3, inputting the data obtained in step S1 into the trained linear synchronous motor model to perform state estimation and parameter identification in real time.

[0055] Based on physical information machine learning, taking the maglev train in the high-speed maglev transportation system as an example, the maglev train suspension control system sensor equipment is used to collect data, and the partial differential equations satisfied by the maglev train linear synchronous motor and boundary conditions and other identities are used to construct the residual. The sum of various residuals is used to construct the loss function, including partial differential structure loss, boundary condition loss, and initial condition loss. A deep learning model is established, and the model is trained to predict the required state quantity or parameter.

[0056] like Figure 1As shown in the figure, the linear motor rotor part on one side of the maglev train is composed of dozens of suspension modules arranged in sequence, and the boundary difference between the suspension module at the end of the train and the suspension module in the middle part is ignored. Under the assumption that the magnetic field is uniformly distributed along the axial direction, considering the symmetry of the motor structure, the smallest motor unit in the linear synchronous motor of the maglev train is analyzed, and two-dimensional analysis is used to reduce the amount of calculation. Among them, the rotor is a standard suspension electromagnet module, and the stator contains three 1.032m iron core modules.

[0057] The magnetic field region Ω satisfies Maxwell's equations, the external boundary is S, L1 and L2 are the boundary lines between the two types of internal media, S, L1, and L2 satisfy the corresponding boundary conditions respectively. The boundary value problem of the linear synchronous motor is equivalent to solving the following set of equations:

[0058] According to Maxwell's equations for static magnetic fields, the following set of equations is established and solved:

[0059]

[0060] Among them, A Z represents the magnetic field z-axis direction vector magnetic potential, x and y represent coordinates, n represents the unit normal vector of the boundary, J and J m represents the current density of the medium, and μ, μ1 and μ2 represent the magnetic permeability of the medium.

[0061] In order to train the neural network, boundary points and other residual points except boundary points are sampled on the minimum motor unit. The training set τ consists of two parts: residual point set τ f and the boundary point set τ b .

[0062]

[0063] in, Represents the sampled residual points, N f Indicates the number of sampled residual points; Represents the sampling boundary points, N b Indicates the number of sampled boundary points.

[0064] Based on the physical mechanism of the long-stator linear synchronous motor, that is, the partial differential equations and boundary conditions that reflect the electromagnetic characteristics of the linear synchronous motor, a linear synchronous motor model based on physical information machine learning is established. The linear synchronous motor model adopts a physical information machine learning model, including but not limited to a physical knowledge embedded neural network (Physics Informed Neural Network, referred to as PINN), a neural network operator (Deep Operator Net, referred to as DeepOnet), and a Fourier neural operator (Fourier Neural Operator, referred to as FNO). Here, PINN is taken as an example.

[0065] Solution of the boundary value problem of linear synchronous motor st (x) can be represented by a neural network parameterized by trainable variables θ Approximately, we choose a fully connected neural network with L layers, and the trainable parameters are θ = {W, b}, where W and b represent the weight and bias of each layer respectively, and W = {W 1 , ...W L}, b={b 1 , …b L},σ l is the activation function, and the neural network can be expressed as:

[0066] x l =σ l (W l ·x l-1 +b l ), l=1,2,…,L (5)

[0067] Among them, x l-1 and x l They represent the lth layer and l-1th layer of the neural network respectively.

[0068] like Figure 2 As shown in the figure, in the process of constructing the linear synchronous motor model, the residual is constructed based on the motor geometry (including boundary points, residual points and interface points), and the sum of various residuals is used to construct the loss function to solve the nonlinear problem. The loss function includes partial differential structure loss, boundary condition loss, initial condition loss, etc.

[0069] like Figure 3 As shown in Figure 2, in the process of linear synchronous motor parameter identification of maglev train, the loss function also includes data loss. The structure of the loss function of the state estimation process is shown in formula (6), and the loss function of the parameter identification process is shown in formula (7):

[0070] L(θ)=w p L PDE +w b L boundary +wi L initial (6)

[0071] L(θ)=w p L PDE +w b L boundary +w i L initial +w d L data (7)

[0072] Among them, L PDE , L boundary , L initial and L data are partial differential structure loss, boundary condition loss, initial condition loss and data loss respectively; w p 、w b 、w i and w d is the loss weight, which is used to balance the loss.

[0073] By minimizing the loss equations (6) and (7), the optimal parameters of the neural network can be obtained:

[0074] θ * =argminL(θ;τ) (8)

[0075] Among them, τ represents the training set, θ represents the neural network parameters, and θ * represents the optimal parameters of the neural network.

[0076] The trained neural network model can infer the predicted solution of the magnetic field region Ω in equation (1) under any spatiotemporal conditions.

[0077] The main steps of modeling using software include the following:

[0078] 1. Define the partial differential equations of the computational domain and its subdomains;

[0079] 2. Define the boundaries and the boundary conditions they meet, including the outer boundaries, interfaces and the boundary conditions they meet;

[0080] 3. Define initial conditions (conditions related to time changes);

[0081] 4. The parameter identification problem also requires the definition of additional data loss conditions;

[0082] 5. Supplement the anchor points of the interface;

[0083] 6. Generate data and network, build the linear synchronous motor model, compile the model, select the optimizer, learning rate and loss weight, and train the model.

[0084] This embodiment provides a state estimation method for a long-stator linear synchronous motor. Figure 4 As shown in the figure, the input of this model is the coordinates and current density of the linear synchronous motor after abstraction. The model solves physical quantities such as magnetic flux, suspension force, and thrust under different working conditions. When the direction of the magnetic induction intensity is parallel to the area of ​​the passing region, the calculation of the magnetic flux Ψ can be done using the following formula:

[0085] Ψ=NΦ=N∫Bds (9)

[0086] Among them, N is the number of coil turns, Φ is the magnetic flux, B is the magnetic induction intensity, and s represents the area of ​​the passing region.

[0087] For solving electromagnetic force, the virtual displacement principle can be used. In electromagnetic system, the concept of magnetic commensurate energy is introduced to calculate electromagnetic force, which is complementary to magnetic commensurate energy on the BH curve. The magnitude of electromagnetic force is equal to the partial derivative of magnetic commensurate energy to displacement under constant current constraint. For linear synchronous motor, thrust F x and the suspension force F z Respectively expressed as:

[0088]

[0089] in is the total magnetic energy in the motor, Ψ is the magnetic flux, x and z represent the vertical and horizontal displacements respectively, H is the magnetic field intensity, V is the volume, and B is the magnetic induction intensity.

[0090] Compared with traditional finite element and other methods, the long-stator linear synchronous motor model established by the present invention can avoid the curse of dimensionality, and after modeling, it can achieve fast and efficient prediction of state quantities that cannot or are difficult to observe during the operation phase of the linear synchronous motor. Compared with pure data-driven deep learning methods, the modeling method of the present invention can overcome the problems of weak model generalization and large amount of data required, so the method has good practical application value.

[0091] This embodiment provides a method for identifying parameters of a long-stator linear synchronous motor. Figure 5 As shown in the figure, compared with the dynamic modeling process, the loss function introduces data loss, and the data samples are part of the observed data. This model identifies parameters such as resistance and coil turns under different working conditions. The data source for parameter identification is not limited to measured data, but also includes simulation data, such as magnetic flux.

[0092] Taking advantage of PINN's ability to naturally integrate data into the solution process, parameter inversion is performed through observed data, which significantly reduces the demand for data and enables the identification of parameters such as the number of coil turns and resistance of the long-stator linear synchronous motor. High-precision parameter identification is achieved based on limited data samples, which helps to establish a reliable dynamic model and thus improve modeling accuracy.

[0093] There is not much difference in the solution form of PINN for motor state estimation and parameter identification. For state estimation problems, physical information is used as constraints and encoded into the neural network loss function for training; for parameter identification problems, that is, some parameters in the equation are unknown and the values ​​of some observation points need to be known, the unknown parameters are optimized as unknown variables, and the loss function includes data loss.

[0094] The electronic device of the present invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0095] Multiple components in the device are connected to the I / O interface, including: input units, such as keyboards, mice, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunication networks.

[0096] The processing unit performs the various methods and processes described above, such as methods S1 to S3. For example, in some embodiments, methods S1 to S3 may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of methods S1 to S3 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1 to S3 in any other appropriate manner (e.g., by means of firmware).

[0097] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0098] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0099] In the context of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0100] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for state estimation and parameter identification of a linear synchronous motor, characterized in that: The method comprises the following steps: Step S1, obtaining the current density and abstracted coordinate data of the linear synchronous motor; Step S2, constructing a linear synchronous motor model based on physical information machine learning based on the partial differential equations and boundary conditions of the electromagnetic characteristics of the linear synchronous motor; Step S3, inputting the data obtained in step S1 into the trained linear synchronous motor model to perform state estimation and parameter identification in real time.

2. A method for state estimation and parameter identification of a linear synchronous motor according to claim 1, characterized in that: The construction process of the linear synchronous motor model includes: The linear synchronous motor magnetic field region Ω satisfies Maxwell's equations, the external boundary is S, L1 and L2 are the boundary lines between the two types of internal media, S, L1, and L2 satisfy the corresponding boundary conditions respectively. The boundary value problem of the linear synchronous motor is equivalent to solving the following set of equations: According to Maxwell's equations for static magnetic fields, the following set of equations is established and solved: Among them, A Z represents the magnetic field z-axis direction vector magnetic potential, x and y represent the geometric coordinates of a point on the motor, n represents the unit normal vector of the boundary, J and J m represents the current density of the medium, μ, μ1 and μ2 represent the magnetic permeability of the medium; The solution of the boundary value problem of the linear synchronous motor is approximated by a neural network parameterized by a trainable variable θ. A fully connected neural network with L layers is selected, and the trainable parameters are θ = {W, b}. The fully connected neural network is expressed as: x l =σ l (W l ·x l-1 +b l ),l=1,2,…,l Among them, x l-1 and x l denote the neural network of the lth layer and the l-1th layer respectively; W and b denote the weight and bias of each layer respectively, W = {W 1 , …W L }, b={b 1 , …b L }; σ l is the activation function.

3. The method for state estimation and parameter identification of a linear synchronous motor according to claim 1, characterized in that: The steps of constructing the linear synchronous motor model include: Step 1: Define the partial differential equations of the computational domain and its subdomains; Step 2: Define the boundaries and the boundary conditions they satisfy, including the outer boundaries, interfaces and the boundary conditions they satisfy; Step 3: Define initial conditions; Step 4: For parameter identification problems, it is also necessary to define data loss conditions; Step 5: Supplement the anchor points of the interface; Step 6: Generate a linear synchronous motor model.

4. The method for state estimation and parameter identification of a linear synchronous motor according to claim 1, characterized in that: The loss function of the linear synchronous motor model for state estimation includes partial differential structure loss, boundary condition loss and initial condition loss, which are specifically: L(θ)=w p L PDE +w b L boundary +w i l initial l PDE , L boundary , L initial are partial differential structure loss, boundary condition loss and initial condition loss respectively, w p 、w b and w i is the loss weight.

5. The method for state estimation and parameter identification of a linear synchronous motor according to claim 1, characterized in that: The loss function of the linear synchronous motor model for parameter identification includes partial differential structure loss, boundary condition loss, initial condition loss and data loss, specifically: l(θ)=w p L PDE +w b L boundary +w i L initial +w d L data Among them, L PDE , L boundary , L initial and L data are partial differential structure loss, boundary condition loss, initial condition loss and data loss respectively; w p 、w b 、w i and w d is the loss weight.

6. A method for state estimation and parameter identification of a linear synchronous motor according to claim 1, characterized in that: The training set of the linear synchronous motor model includes a residual point set and a boundary point set, wherein the residual point set and the boundary point set are obtained by sampling on the minimum motor unit.

7. A method for state estimation and parameter identification of a linear synchronous motor according to claim 1, characterized in that: The linear synchronous motor model is a neural network with embedded physical knowledge.

8. The method for state estimation and parameter identification of a linear synchronous motor according to claim 1, characterized in that: The state estimation includes state estimation of magnetic flux, suspension force and thrust.

9. A method for state estimation and parameter identification of a linear synchronous motor according to claim 1, characterized in that: The parameter identification includes parameter identification of resistance and coil turns; the input of the parameter identification also includes magnetic flux.

10. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 9 is implemented.

Images