Permanent magnet synchronous motor full flux linkage model parameter identification method based on full-connection neural network

Abstract

The invention relates to a permanent magnet synchronous motor full flux linkage model parameter identification method based on a full-connection neural network, and belongs to the technical field of permanent magnet synchronous motor control. The method comprises the following steps: deducing a discrete voltage equation of a motor voltage equation under the action of a digital controller according to a forward Euler method; obtaining a simplified discrete voltage equation under the condition that the given dq-axis current and load of the motor are maintained at stable values; according to the simplified discrete voltage equation, calculating flux linkage data corresponding to the current by changing the load torque and scanning the dq axis current; and obtaining a flux linkage value corresponding to the current through a mapping relation between the current and the flux linkage in the full-connection deep neural network flux linkage mapping model, and further obtaining the electromagnetic torque under the action of the current. According to the method, the full-connection neural network is constructed to carry out parameter learning and identification on the flux linkage, and the flux linkage-current mapping relation of the permanent magnet synchronous motor is accurately fitted.

Description

Technical Field

[0001] The invention belongs to the technical field of permanent magnet synchronous motor control, and relates to a method for identifying parameters of a full magnetic linkage model of a permanent magnet synchronous motor based on a fully connected neural network. Background Art

[0002] Permanent magnet synchronous motor (PMSM) has the advantages of high efficiency, high power density, and good control performance, and has been widely used in the field of servo motors. Accurate model parameters are often required to achieve high-precision control and excellent dynamic response, especially for model predictive control, whose control performance is extremely sensitive to model parameters. Under short-term overload conditions, the motor inductance can no longer be characterized by a single constant value due to the limitation of flux linkage, but instead shows obvious nonlinear characteristics as the overload degree increases, and a certain cross-saturation effect occurs. Therefore, it is necessary to adopt a mathematical model based on direct mapping of flux linkage and current. This model can more intuitively and accurately reflect the input and output characteristics of the motor under full servo conditions, and can also provide a theoretical basis for servo overload performance evaluation and high-performance servo control algorithm design.

[0003] However, under servo conditions, the motor needs to perform acceleration and deceleration control frequently, and the current vector spans the second and third quadrants, making the change trend of the flux surface more complicated and increasing the difficulty of fitting. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a method for identifying the parameters of the full magnetic linkage model of a permanent magnet synchronous motor based on a fully connected neural network. Deep neural networks can be divided into recurrent neural networks, convolutional neural networks, fully connected neural networks, etc. from the perspective of structure. Among them, the fully connected neural network can theoretically achieve arbitrary precision approximation for any continuous function, and has wider practicality and reliability in fitting problems than commonly used methods such as polynomial functions and trigonometric functions. And with the continuous improvement of the computing power of digital processing chips, appropriately optimized and compressed deep neural network strategies can also be transplanted into the driver program while ensuring a higher control frequency.

[0005] In view of this, a parameter identification method of the full flux linkage model of permanent magnet synchronous motor based on fully connected neural network is proposed.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A method for identifying parameters of a full magnetic linkage model of a permanent magnet synchronous motor based on a fully connected neural network, the method comprising:

[0008] S1. Derive the discrete voltage equation of the motor voltage equation under the action of the digital controller according to the forward Euler method;

[0009] S2, when the motor's given dq axis current and load size are maintained at a stable value, a simplified discrete voltage equation is obtained;

[0010] S3, according to the simplified discrete voltage equation, by changing the load torque and scanning the dq axis current, calculating the flux data corresponding to the current;

[0011] S4. The magnetic flux value corresponding to the current is obtained through the mapping relationship between current and magnetic flux in the fully connected deep neural network magnetic flux mapping model, and then the electromagnetic torque under the action of the current is obtained.

[0012] Further, in step S1, the discrete voltage equation of the motor voltage equation under the action of the digital controller is derived according to the forward Euler method:

[0013]

[0014] In the formula, u d 、u q They represent the d-axis voltage and q-axis voltage in the dq coordinate system respectively; i d 、i q They represent the d-axis current and q-axis current in the dq coordinate system respectively; ψ d , q They represent the d-axis flux and q-axis flux in the dq coordinate system respectively; R represents the phase resistance of the stator winding; ω e Indicates the electrical angular velocity of the motor; T S represents the discrete control cycle of the digital controller; k represents the time instant.

[0015] Furthermore, in step S2, given that the dq axis current and the load size are both maintained at stable values, both sides of the discrete voltage equation are integrated simultaneously, and we have:

[0016]

[0017] Where, T s is the discrete control period of the digital controller, and m is the number of integral periods;

[0018] The dq axis voltage, current and flux are regarded as constants, and the integral of the velocity is expressed as the position difference between the beginning and the end of the integral, thus further obtaining the simplified discrete voltage equation:

[0019]

[0020] In the formula, Δθ em is the position difference between the start time and the end time of the integration.

[0021] Further, in step S3, the following steps are included:

[0022] The current closed loop is used to control the dq axis current at the data point current value i ds and i qs , a certain load torque T is applied by the load motor load After the motor speed reaches a stable state, collect multiple mT s Δθ in the time period e with u d 、u q The value of and average value are calculated to obtain Δθ under this load. ei with u di 、u qi ;

[0023] Then, change T load Repeat the process of collecting voltage and position difference to obtain a new set of Δθ ei with u di 、u qi ;

[0024] According to n groups of dq axis voltage and position difference, mT is calculated by the least square method. s U d With mT s U q With Δθ em The slope of change is the current data point (i ds ,i qs ) under the dq-axis flux value ψ q (i ds ,i qs ) and ψ d (i ds ,i qs );

[0025] Data collection and processing are performed for each sampling point in the above manner to obtain an original data set including the dq axis current and the dq axis flux linkage value under the current.

[0026] Further, in step S4, a deep neural network flux mapping model for identifying the mapping relationship between current and flux includes: a dq axis current input layer, a hidden fully connected layer, a Dropout layer and an output layer, wherein each neuron uses a Relu function as an activation function.

[0027] Further, in step S4, the relationship between the input and output of the deep neural network magnetic linkage mapping model is expressed by the following formula:

[0028] ψ dq =f θ (n) (f θ (n-1) (...f θ (1) (idq )))

[0029] In the formula, f θ (i) is the feedforward transfer function of the i-th layer neuron, i=1,2,…,n, n is the total number of neural network layers; for the i-th neuron, its transfer function satisfies:

[0030] f θ (i) (Y i )=R(W i Y i +b i )

[0031] Where Y i is the output of the upper layer neurons; W i is the weight coefficient matrix between the neurons in this layer and the neurons in the upper layer; b i is the offset coefficient vector between the neurons in this layer and the neurons in the upper layer; R is the activation function;

[0032] The three-phase current is converted into the current value under the dq axis through the dq axis coordinate change, and then the magnetic flux value under the action of the current is obtained through the deep neural network magnetic flux mapping model. At the same time, the electromagnetic torque under the action of the current can be obtained according to the following formula:

[0033]

[0034] In the formula, p n Indicates the number of motor pole pairs, w' m represents the magnetic function of each winding, and θ represents the angle between the rotor N pole and the A-phase axis.

[0035] The beneficial effects of the present invention are:

[0036] Based on the full flux linkage model of permanent magnet synchronous motor with direct flux linkage-current mapping, this invention proposes a method for flux linkage parameter identification using a fully connected neural network, which realizes the direct characterization of dq axis flux linkage saturation and cross saturation in the full working range of the motor, and can provide an accurate electromagnetic model for the motor control strategy. The experimental results show that the constructed neural network can fit the flux linkage-current mapping more accurately, and realize relatively accurate flux linkage parameter identification within the error range of 3% of finite element data and 7% of experimental data.

[0037] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0039] Figure 1 It is a schematic diagram of the surface extraction process of the fully connected neural network magnetic flux model mapping of the present invention;

[0040] Figure 2 It is a structural schematic diagram of the fully connected neural network magnetic linkage model of the present invention;

[0041] Figure 3 It is a schematic diagram of the training results of the fully connected neural network magnetic flux model of the present invention;

[0042] Figure 4 is the flux fitting result and error of the fully connected neural network flux model of the present invention, wherein, Figure 4 (a) and Figure 4 (b) are the magnetic flux fitting results of the q and d axes, Figure 4 (c) and Figure 4 (d) The magnetic flux fitting errors of the q and d axes respectively;

[0043] Figure 5 Schematic diagram of flux-current grid points of flux data obtained in the embodiment of the present invention, wherein: Figure 5 (a) and Figure 5 (b) The flux-current grid points of the q-axis and d-axis flux data, respectively;

[0044] Figure 6 Schematic diagram of model parameter identification error in an embodiment of the present invention, wherein: Figure 6 (a) and Figure 6 (b) Schematic diagrams of the q-axis and d-axis flux identification error results respectively. DETAILED DESCRIPTION

[0045] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0046] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0047] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0048] See also Figure 1 to Figure 6 , which is a parameter identification method for the full flux linkage model of a permanent magnet synchronous motor based on a fully connected neural network.

[0049] In the method for identifying the parameters of the full flux linkage model of a permanent magnet synchronous motor based on a fully connected neural network of the present invention, in order to better characterize the electromagnetic characteristics of the motor in the servo electric drive system, the present invention proposes a data set extraction and identification method for the flux linkage-current mapping surface. Through the variable torque method, the flux linkage data corresponding to the current can be calculated in an offline experiment by changing the load torque and scanning the dq axis current, and a fully connected neural network is constructed to perform parameter learning and identification on the flux linkage, so as to achieve a more accurate fitting of the flux linkage-current mapping relationship of the permanent magnet synchronous motor.

[0050] In order to realize the feature extraction and analysis of the motor mathematical model, it is necessary to observe the flux under a given current and obtain an accurate and sufficient flux data set. The motor voltage equation is derived by the forward Euler method to obtain the discrete voltage equation under the action of the digital controller:

[0051]

[0052] In the formula, u d 、u q They represent the d-axis voltage and q-axis voltage in the dq coordinate system respectively; i d 、i q They represent the d-axis current and q-axis current in the dq coordinate system respectively; ψ d , q They represent the d-axis flux and q-axis flux in the dq coordinate system respectively; R represents the phase resistance of the stator winding; ω eIndicates the electrical angular velocity of the motor; T S Represents the discrete control cycle of the digital controller; k represents the time. The current and speed variables that can be directly observed from the formula contain a variety of state information including flux, resistance, etc. However, due to the existence of differential terms and motor speed variables, directly solving the formula easily leads to the state fluctuation of the motor operation process being amplified in the differential part, resulting in errors in the flux observation value. And due to the influence of SVPWM and inverter dead zone, there is often a certain voltage loss between the given voltage and the actual voltage. Therefore, in order to avoid the influence of voltage loss and differential error and improve the accuracy of the flux data set, it is assumed that during each set of data acquisition, the motor's given dq axis current and load size are maintained at a stable value, and both sides of the above formula are integrated at the same time, resulting in:

[0053]

[0054] Where, T s is the discrete control period of the digital controller, and m is the number of integral periods.

[0055] When the current and speed remain constant during the integration period, the dq axis voltage, current and flux can be regarded as constants. At the same time, the integral of the speed can also be expressed as the position difference between the beginning and the end of the integration. Therefore, the above formula can be simplified to:

[0056]

[0057] In the formula, Δθ em is the position difference between the start time and the end time of the integration.

[0058] Figure 1 This is a schematic diagram of the surface extraction process of the fully connected neural network magnetic flux model mapping of the present invention. Figure 1 The off-line schematic diagram of the extraction process is given. To measure a data point in the fitting data set, a current closed loop is used to control the dq axis current at the data point current value i ds and i qs , a certain load torque T is applied by the load motor load After the motor speed reaches a stable state, collect multiple mT s Δθ in the time period e with u d 、u q The value of and average value are calculated to obtain Δθ under this load. ei with u di 、u qi Then, change T load Repeat the process of collecting voltage and position difference to obtain a new set of Δθ ei with u di 、u qiThrough n sets of dq axis voltage and position difference, mT can be calculated by the least square method. s U d With mT s U q With Δθ em The slope of change is the current data point (i ds ,i qs ) under the dq-axis flux value ψ q (i ds ,i qs ) and ψ d (i ds ,i qs ). For each sampling point, data collection and processing are performed in the above manner to obtain an original data set including the dq axis current and the dq axis flux value under the current. The original data set can be used for fitting the motor flux surface and constructing the flux model.

[0059] In order to establish a magnetic flux model, the present invention identifies the mapping relationship between current and magnetic flux through a fully connected deep neural network based on a data set measured offline. Figure 2 The schematic diagram of the structure of the fully connected neural network flux model of the present invention is shown. The model consists of a dq axis current input layer, a hidden fully connected layer, a Dropout layer, and an output layer. Each neuron uses the Relu function as an activation function. Through this model, the flux-current surface can be more accurately characterized. The relationship between the model input and output can be expressed by the following formula:

[0060] ψ dq =f θ (n) (f θ (n-1) (...f θ (1) (i dq )))

[0061] In the formula, f θ (i) is the feedforward transfer function of the i-th layer neuron, i=1,2,…,n, n is the total number of neural network layers. For the i-th neuron, its transfer function satisfies:

[0062] f θ (i) (Y i )=R(W i Y i +b i )

[0063] Where Y i is the output of the upper layer neurons; W iis the weight coefficient matrix between the neurons in this layer and the neurons in the upper layer; b i is the offset coefficient vector between the neurons in this layer and the neurons in the upper layer; R is the activation function, and the Relu function is used as the system activation function in this embodiment.

[0064] Therefore, the three-phase current is converted into the current value under the dq axis through the change of the dq axis coordinates, and then the magnetic flux value under the action of the current is obtained through the deep neural network magnetic flux mapping model. At the same time, the electromagnetic torque under the action of the current can be obtained according to the following formula:

[0065]

[0066] This embodiment uses a finite element model to perform parameter scanning and construct a current-flux data set for neural network training and verification. At each current point, a three-phase excitation current source is established through inverse Park transformation, finite element simulation calculations are performed and the flux simulation results are recorded to construct a data set of the full flux model. Then, a neural network is built based on the Python Keras framework to learn the flux mapping surface. The training results of the model are shown in Figure 1. Figure 3 As shown. Finally, the comparison between the current flux data under the training model and the actual finite element data is obtained. Figure 4 As shown, Figure 4 is the flux fitting result and error of the fully connected neural network flux model of the present invention, wherein, Figure 4 (a) and Figure 4 (b) are the magnetic flux fitting results of the q and d axes, Figure 4 (c) and Figure 4 (d) are the magnetic flux fitting errors of the q and d axes respectively. The model of the present invention achieves a better fitting effect.

[0067] In this embodiment, the dq axis current is traversed on the experimental servo motor, and the flux-current grid points obtained are as follows: Figure 5 As shown, Figure 5 (a) and Figure 5 (b) The flux-current grid points of the q-axis and d-axis flux data, respectively. Compared with the data obtained by scanning the finite element model, the data obtained by the offline experiment has certain fluctuations, which is related to non-ideal factors such as disturbances in the experimental process and AD sampling errors. However, its overall change law and numerical range are basically consistent with the finite element data. Based on this data, the full flux parameter identification is also performed using the method in the previous section. The error between the final identification result and the actual measured flux is shown in Figure 6 As shown, Figure 6 (a) and Figure 6(b) are schematic diagrams of the q-axis and d-axis flux identification error results, respectively. It can be seen that compared with the identification method based on finite element data, the fully connected neural network trained by offline experimental data has a larger error in flux identification. The d-axis flux error reaches the range of [-6.67%, 5.27%], and the q-axis flux error reaches the range of [-5.7%, 6.5%]. The main reason for the increase in error is that the fluctuation of the experimental data set makes it difficult for the fully connected network to approach more complex data changes, making it difficult to achieve an ideal matching effect between the identified flux and the actual flux. However, even with the existence of this error, compared with the fixed dual-axis inductance model, the flux identification can still be closer to the saturation state of the actual motor under high overload conditions. Its error can also be corrected to a certain extent by adding disturbance compensation and other methods.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for identifying parameters of a full flux linkage model of a permanent magnet synchronous motor based on a fully connected neural network, characterized in that: The method comprises: S1. Derive the discrete voltage equation of the motor voltage equation under the action of the digital controller according to the forward Euler method; S2, when the motor's given dq axis current and load size are maintained at a stable value, a simplified discrete voltage equation is obtained; S3, according to the simplified discrete voltage equation, by changing the load torque and scanning the dq axis current, calculating the flux data corresponding to the current; S4. Based on the measured flux data, the flux value corresponding to the current is obtained through the mapping relationship between current and flux in the fully connected deep neural network flux mapping model, and then the electromagnetic torque under the action of the current is obtained.

2. According to claim 1, a method for identifying parameters of a permanent magnet synchronous motor full flux linkage model based on a fully connected neural network is characterized in that: In step S1, the discrete voltage equation of the motor voltage equation under the action of the digital controller is derived according to the forward Euler method: In the formula, u d 、u q They represent the d-axis voltage and q-axis voltage in the dq coordinate system respectively; i d 、i q They represent the d-axis current and q-axis current in the dq coordinate system respectively; ψ d , q They represent the d-axis flux and q-axis flux in the dq coordinate system respectively; R represents the phase resistance of the stator winding; ω e Indicates the electrical angular velocity of the motor; T S represents the discrete control cycle of the digital controller; k represents the time instant.

3. The method for identifying parameters of a permanent magnet synchronous motor full flux linkage model based on a fully connected neural network according to claim 2 is characterized in that: In step S2, given that the dq axis current and the load size are both maintained at stable values, both sides of the discrete voltage equation are integrated simultaneously, and we have: Where, T s is the discrete control period of the digital controller, and m is the number of integral periods; The dq axis voltage, current and flux are regarded as constants, and the integral of the velocity is expressed as the position difference between the beginning and the end of the integral, thus further obtaining the simplified discrete voltage equation: In the formula, Δθ em is the position difference between the start time and the end time of the integration.

4. The method for identifying parameters of a full magnetic flux linkage model of a permanent magnet synchronous motor based on a fully connected neural network according to claim 3 is characterized in that: In step S3, the following steps are included: The current closed loop is used to control the dq axis current at the data point current value i ds and i qs , a certain load torque T is applied by the load motor load After the motor speed reaches a stable state, collect multiple mT s Δθ in the time period e with u d 、u q The value of and average value are calculated to obtain Δθ under this load. ei with u di 、u qi , where i (i=1, 2, 3...) represents the data group number; Then, change T load Repeat the process of collecting voltage and position difference to obtain a new set of Δθ ei with u di 、u qi ; According to n groups of dq axis voltage and position difference, mT is calculated by the least square method. s U d With mT s U q With Δθ em The slope of change is the current data point (i ds ,i qs ) under the dq-axis flux value ψ q (i ds ,i qs ) and ψ d (i ds ,i qs ); Data collection and processing are performed for each sampling point in the above manner to obtain an original data set including the dq axis current and the dq axis flux linkage value under the current.

5. The method for identifying parameters of a full magnetic flux linkage model of a permanent magnet synchronous motor based on a fully connected neural network according to claim 4 is characterized in that: In step S4, a deep neural network flux mapping model for identifying the mapping relationship between current and flux includes: a dq axis current input layer, a hidden fully connected layer, a Dropout layer and an output layer, wherein each neuron uses a Relu function as an activation function.

6. The method for identifying parameters of a permanent magnet synchronous motor full flux linkage model based on a fully connected neural network according to claim 5 is characterized in that: In step S4, The relationship between the input and output of the deep neural network magnetic link mapping model is expressed by the following formula: ψ dq =f θ (n) (f θ (n-1) (...f θ (1) (i dq ))) In the formula, f θ (i) is the feedforward transfer function of the i-th layer neuron, i=1,2,…,n, n is the total number of neural network layers; for the i-th neuron, its transfer function satisfies: f θ (i) (Y i )=R(W i Y i +b i ) Where Y i is the output of the upper layer neurons; W i is the weight coefficient matrix between the neurons in this layer and the neurons in the upper layer; b i is the offset coefficient vector between the neurons in this layer and the neurons in the upper layer; R is the activation function; The three-phase current is converted into the current value under the dq axis through the dq axis coordinate change, and then the magnetic flux value under the action of the current is obtained through the deep neural network magnetic flux mapping model. At the same time, the electromagnetic torque under the action of the current can be obtained according to the following formula: In the formula, p n Indicates the number of motor pole pairs, w' m represents the magnetic function of each winding, and θ represents the angle between the rotor N pole and the A-phase axis.

Images