Encoder-free permanent magnet synchronous motor control method based on online sparse Bayesian learning and improved reaching law

Abstract

The invention discloses an encoder-free permanent magnet synchronous motor control method based on online sparse Bayesian learning and an improved reaching law, and the method comprises the steps: calculating an RBF kernel function value through the collected real values and reference values of voltage and current; constructing a data set according to the RBF kernel function value; the method comprises the following steps: calculating deviation between flux linkages by calculating rated permanent magnet flux linkages and an observer, and processing the deviation by adopting a nonlinear reaching law based on a logarithmic function to obtain a flux linkage deviation item; according to the motor mathematical model, the flux linkage deviation item and the rated parameter value, constructing an accurate flux linkage increment calculation model containing an adjustable weight factor; dynamically updating a weight factor through a sparse Bayesian learning algorithm in combination with the data set, and calculating a rotor angle according to the updated model; and performing coordinate transformation and actual rotating speed estimation based on the rotor angle, and performing position-sensorless control on the motor by using a double-closed-loop vector control algorithm. The method is small in steady-state angle observation error and high in dynamic response speed.

Description

Technical Field

[0001] The present invention belongs to the technical field of motor control, and in particular relates to an encoderless permanent magnet synchronous motor control method based on online sparse Bayesian learning and improved reaching law. Background Art

[0002] Permanent magnet synchronous motors (PMSMs) are widely used in industrial automation, electric vehicles, aerospace, and other fields due to their high efficiency, high performance, and excellent dynamic response. Traditional field-oriented control strategies typically rely on rotor position sensors for precise control. These sensors provide real-time position information of the motor rotor, enabling accurate vector control. However, under high-speed operation or harsh environmental conditions, position sensors may fail or degrade, compromising the stability and accuracy of the control system, potentially leading to system failure.

[0003] Furthermore, in space-constrained or low-cost applications, installing a position sensor may be impractical or cost-prohibitive. Consequently, encoderless control technology has emerged. It uses motor models and algorithms to estimate rotor position, eliminating the need for additional position sensors. Model-based approaches rely on a mathematical model of the motor, calculating the motor's flux or back EMF and determining the rotor angle through an inverse tangent calculation. However, traditional sensorless control algorithms suffer from significant sensitivity to parameter disturbances, high cost, and slow response speed.

[0004] Existing encoderless control methods primarily include position observers based on high-frequency injection and traditional nonlinear flux observers. High-frequency injection position observers require injecting high-frequency signals of varying amplitudes or frequencies into the permanent magnet synchronous motor's stator windings, which is costly. They also require filter design, potentially resulting in poor dynamic response. Furthermore, they are only suitable for low-speed operation. Traditional nonlinear flux observers use flux amplitude for angle identification, are significantly affected by parameter perturbations, and cannot dynamically adjust parameters. They also require a large gain factor, potentially amplifying noise.

[0005] Therefore, there is an urgent need in the prior art for a sensorless control method for a permanent magnet synchronous motor with faster dynamic response and unaffected by parameter disturbances. Summary of the Invention

[0006] The present invention proposes an encoderless permanent magnet synchronous motor control method based on online sparse Bayesian learning and improved reaching law to solve the problems existing in the above-mentioned prior art.

[0007] To achieve the above object, the present invention provides an encoderless permanent magnet synchronous motor control method based on online sparse Bayesian learning and improved reaching law, comprising the following steps:

[0008] Calculate the RBF kernel function value through the collected real values ​​of voltage and current and the reference values;

[0009] Constructing a data set according to the RBF kernel function value, the data set including current, voltage, deviation term and RBF kernel function value;

[0010] By calculating the deviation between the rated permanent magnet flux and the flux calculated by the observer, the deviation is processed by a nonlinear reaching law based on a logarithmic function to obtain a flux deviation term;

[0011] Based on the motor mathematical model, flux deviation term and rated parameter values, an accurate flux increment calculation model with adjustable weight factors is constructed;

[0012] Dynamically updating the weight factors in combination with the data set using a sparse Bayesian learning algorithm, and calculating the rotor angle based on the updated model;

[0013] Based on the rotor angle, coordinate transformation and actual speed estimation are performed, and a double closed-loop vector control algorithm is used to perform position sensorless control of the motor.

[0014] Optionally, the expression for calculating the RBF kernel function value is:

[0015]

[0016] k αβ,2,i =exp(-|Φ i * -Φ i | 2 );

[0017] Where k represents the adjustable weight coefficient in the precise mathematical model, the subscript αβ represents the two-phase stationary coordinate system, and the subscript i represents the i-th control cycle; i α 、i β and Φ represent the deviation between the current of the α-axis, the current of the β-axis and the flux linkage, respectively. The subscript α represents the α-axis of the two-phase stationary coordinate system, the subscript β represents the β-axis of the two-phase stationary coordinate system, and the superscript * represents the reference value.

[0018] Optionally, the expression for constructing the data set according to the RBF kernel function value is:

[0019]

[0020] κ αβ =[1,k αβ,1,i R s ,k αβ,1,i δ,-k αβ,2,i δ*L] T ;

[0021] Where φ represents the data vector, κ represents the parameter vector; u represents the measured voltage, i represents the current, represents the stator flux observation value; R s , L and δ represent the stator resistance, inductance and adjustable coefficient respectively.

[0022] Optionally, the expression for calculating the deviation between flux linkages is:

[0023]

[0024] Where, is the rotor flux rating, ∈ and ε are two constants,

[0025] Optionally, the expression of the precise flux increment calculation model is:

[0026]

[0027] Where w αi , w βi , i = 1, 2, 3, 4 and w αi , w βi , i = 1, 2, 3, 4 represent the unknown weight coefficients of the α-axis and β-axis respectively, which are obtained through Bayesian learning of coefficients.

[0028] Optionally, dynamically updating the weight factor includes: calculating a new magnetic linkage increment according to the data set, calculating the posterior distribution mean and variance of the weight factor based on a sparse Bayesian learning algorithm and the new magnetic linkage increment, and using the mean of the weight factor as the updated weight factor.

[0029] Optionally, the expression for calculating the new flux increment is:

[0030]

[0031] Where, Indicates the new flux increment;

[0032] The expression for calculating the variance of the posterior distribution of the weight factor is:

[0033]

[0034] Where ρ represents the variance of the noise Gaussian distribution;

[0035] The expression for calculating the mean of the weight factors is:

[0036]

[0037] Where S w and τ wThey represent the mean and variance of the unknown weight coefficients after considering parameter perturbations, and α is the covariance matrix of the Gaussian distribution.

[0038] Optionally, the expression for calculating the rotor angle is:

[0039]

[0040] Where, represents the observed rotor angle.

[0041] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0042] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method when executed by a processor.

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

[0044] The present invention proposes an encoderless permanent magnet synchronous motor control method based on online sparse Bayesian learning and improved convergence law. The present invention improves the dynamic response speed by designing a convergence law based on a logarithmic function, and uses a sparse Bayesian algorithm to dynamically update the weights in the precise mathematical model of the motor to adapt to changes in the motor parameters, thereby enhancing the parameter robustness of the system. In addition, the system does not need to rely on expensive rotor position sensors, which reduces costs. The present invention is applicable to a variety of motor types, including DC brushed motors, DC brushless motors, switched reluctance motors and synchronous reluctance motors, and has broad application prospects. By constructing a precise mathematical model for calculating flux increments that includes adjustable weight factors, the influence of parameter disturbances on angle observations is reduced, the control accuracy is improved, and position sensorless control is achieved, thereby improving the flexibility and adaptability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0046] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0047] Figure 2 A comparison diagram of the reaching law of the nonlinear angle observer according to an embodiment of the present invention;

[0048] Figure 3This is a control block diagram of an encoderless permanent magnet synchronous motor control system based on online sparse Bayesian learning and improved reaching law according to an embodiment of the present invention;

[0049] Figure 4 Graph showing experimental results under steady-state conditions according to an embodiment of the present invention;

[0050] Figure 5 Graph showing experimental results under speed reversal conditions according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0052] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0053] Example 1

[0054] like Figure 1 As shown, this embodiment provides an encoderless permanent magnet synchronous motor control method based on online sparse Bayesian learning and improved reaching law, including the following steps:

[0055] The RBF kernel function value is calculated by the collected real values ​​of voltage and current and the reference values; the RBF kernel function value is shown in formulas (8)-(9);

[0056] A data set is constructed based on the RBF kernel function value. The data set includes current, voltage, deviation term, and RBF kernel function value. The data set is shown in formulas (6)-(7).

[0057] By calculating the rated permanent magnet flux and the deviation between fluxes calculated by the observer, the deviation is processed by a nonlinear reaching law based on a logarithmic function to obtain the flux deviation term; the deviation between fluxes is shown in formula (3);

[0058] According to the motor mathematical model, flux deviation term and rated parameter value, an accurate flux increment calculation model including an adjustable weight factor is constructed; the accurate flux increment calculation model is shown in formula (5);

[0059] The weight factors are dynamically updated by combining the data set through the sparse Bayesian learning algorithm. The updated weight factors are implemented by formulas (10)-(12), and the rotor angle is calculated based on the updated model. The rotor angle is calculated by formula (4).

[0060] Based on the rotor angle, coordinate transformation and actual speed estimation are performed, and a dual closed-loop vector control algorithm is used for position sensorless control of the motor.

[0061] The following details are provided:

[0062] pass Figure 2 The reaching law of the traditional nonlinear angle observer and the reaching law of the newly designed nonlinear angle observer can be seen. As can be seen from the figure, the reaching law designed in this scheme will amplify the error when the error is large, ensuring the rapid convergence of the angle observer; when the angle error is small, it will reduce the error and avoid amplifying the sensor noise.

[0063] The key to angle observation is to calculate the stator flux. The traditional nonlinear angle observer calculates the stator flux increment by taking advantage of the constant amplitude of the rotor flux. As shown in the following formula:

[0064]

[0065] where u αβ and i αβ is the voltage and current, R s and L are resistance and inductance, δ and Φ are adjustable coefficient and nonlinear term respectively. In (1), we have

[0066] In the reaching law proposed in this embodiment, the calculation formula of Φ is designed as follows:

[0067]

[0068] in

[0069] After integrating formula (1), the rotor angle can be calculated based on the characteristics of the magnetic flux and the inverse tangent function as shown below:

[0070]

[0071] According to the Lyapunov function, this controller It can remain stable under all working conditions.

[0072] Based on the above process, we can see that the reaching law of the nonlinear observer designed in this solution has good dynamic response capabilities. However, this nonlinear angle observer is highly dependent on the motor parameters. If the motor parameters are disturbed, it will lead to large angle observation errors. Therefore, based on this, an encoderless permanent magnet synchronous motor control system based on online sparse Bayesian learning and an improved reaching law was designed.

[0073] In this embodiment, the precise mathematical model and data set of the permanent magnet synchronous motor are constructed as follows:

[0074] The traditional permanent magnet synchronous motor mathematical model does not include weight coefficients, while the mathematical model constructed in this embodiment includes dynamically adjustable weight coefficients, as shown below:

[0075]

[0076] in, As the weight coefficient term, the weight coefficient matrix w can be constructed αβ =[w αβ1 ,w αβ2 ,w αβ3 ,w αβ4 ];

[0077] In order to obtain the characteristics of the current disturbance to the motor, it is necessary to collect historical data of N control cycles and form a data vector and a parameter vector as shown below:

[0078]

[0079] κ αβ =[1,k αβ,1,i R s ,k αβ,1,i δ,-k αβ,1,i δ*L] T (7)

[0080] where k αβ,1,i and k αβ,2,i is the radial basis kernel function, as shown below:

[0081]

[0082] k αβ,2,i =exp(-|Φ * -Φ| 2 ) (9)

[0083] In this embodiment, the online updating method of weight coefficients based on sparse Bayesian learning is as follows:

[0084] The sparse Bayesian algorithm can use the prior distribution and likelihood function to solve for the posterior distribution. Therefore, by analyzing the disturbances in the historical data, the difference between the motor mathematical model and the true value for the next control cycle can be determined and compensated for by adjusting the weighting factors. The expression for the flux increment recalculated using the data set is shown below:

[0085]

[0086] According to sparse Bayes, the mean and variance of the weight vector can be obtained as follows:

[0087]

[0088] The mean of the weight vector τ w Substituting it into formula (5) can calculate the flux increment, and then substituting it into formula (4) can calculate the rotor angle, thus completing the nonlinear sensorless control based on sparse Bayesian.

[0089] The meanings of the parameters involved in this embodiment are as follows:

[0090]

[0091]

[0092] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for controlling an encoderless permanent magnet synchronous motor based on online sparse Bayesian learning and improved reaching law, characterized in that: The following steps are involved: Calculate the RBF kernel function value through the collected real values ​​of voltage and current and the reference values; Constructing a data set according to the RBF kernel function value, the data set including current, voltage, deviation term and RBF kernel function value; By calculating the deviation between the rated permanent magnet flux and the flux calculated by the observer, the deviation is processed by a nonlinear reaching law based on a logarithmic function to obtain a flux deviation term; Based on the motor mathematical model, flux deviation term and rated parameter values, an accurate flux increment calculation model with adjustable weight factors is constructed; Dynamically updating the weight factors in combination with the data set using a sparse Bayesian learning algorithm, and calculating the rotor angle based on the updated model; Based on the rotor angle, coordinate transformation and actual speed estimation are performed, and a double closed-loop vector control algorithm is used to perform position sensorless control of the motor.

2. The method according to claim 1, characterized in that The expression for calculating the RBF kernel function value is: k αβ,2,i =exp(-|Φ i * -F i | 2 ); Where k represents the adjustable weight coefficient in the precise mathematical model, the subscript αβ represents the two-phase stationary coordinate system, and the subscript i represents the i-th control cycle; i α 、i β and Φ represent the deviation between the current of the α-axis, the current of the β-axis and the flux linkage, respectively. The subscript α represents the α-axis of the two-phase stationary coordinate system, the subscript β represents the β-axis of the two-phase stationary coordinate system, and the superscript * represents the reference value.

3. The method according to claim 2, characterized in that The expression for constructing a data set based on the RBF kernel function value is: k αβ =[1,k αβ,1,i R s ,k αβ,1,i d,-k αβ,2,i [d*L] T ; Where φ represents the data vector and κ represents the parameter vector; u represents the measured voltage, i represents the current, represents the stator flux observation value; R s , L and δ represent the stator resistance, inductance and adjustable coefficient respectively.

4. The method according to claim 3, characterized in that The expression for calculating the deviation between magnetic flux linkages is: Where, is the rotor flux rating, ∈ and ε are two constants, 5. The method according to claim 4, characterized in that The expression of the precise flux increment calculation model is: Where w αi ,w βi ,i=1,2,3,4 and w αi , w βi =i=1, 2, 3, 4 represent the unknown weight coefficients of the α-axis and β-axis respectively, which are obtained through Bayesian learning of coefficients.

6. The method according to claim 5, characterized in that Dynamically updating the weight factor includes: calculating a new magnetic flux increment according to the data set, calculating the posterior distribution mean and variance of the weight factor based on a sparse Bayesian learning algorithm and the new magnetic flux increment, and using the mean of the weight factor as the updated weight factor.

7. The method according to claim 6, characterized in that The expression for calculating the new flux increment is: Where, Indicates the new flux increment; The expression for calculating the variance of the posterior distribution of the weight factor is: Where ρ represents the variance of the noise Gaussian distribution; The expression for calculating the mean of the weight factors is: Where S w and τ w They represent the mean and variance of the unknown weight coefficients after considering parameter perturbations, and α is the covariance matrix of the Gaussian distribution.

8. The method according to claim 1, characterized in that The expression for calculating the rotor angle is: Where, represents the observed rotor angle.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images