Permanent magnet motor continuous set model-free predictive control method based on Bayesian optimization

Abstract

The invention provides a permanent magnet motor continuous set model-free predictive control method based on Bayesian optimization, and relates to the field of permanent magnet synchronous motor control. The method comprises the following steps: acquiring a hyper-local model and an extended state observer; determining the bandwidth and the input gain as optimization variables, and obtaining a training set and a test set for Bayesian optimization; joint probability distribution of the training set and the test set is calculated through a kernel function, and a prediction mean value and a prediction variance of optimization variables are determined; an optimization variable combination is screened based on the expected value collection function, and the bandwidth and the input gain are updated based on the optimization variable combination; based on the updated extended state observer and the hyper-local model, calculating the reference voltage of the next control period by using a dead-beat control principle; the reference voltage is decomposed into basic voltage vectors through space vector modulation, and the switching sequence of the basic voltage vectors is executed according to the minimum switching times to drive the permanent magnet motor. The method is not affected by motor parameter changes, and the current prediction control steady-state performance can be improved.

Description

Technical Field

[0001] The present invention relates to the field of motor control, and particularly to a continuous set model-free predictive control method for a permanent magnet motor based on Bayesian optimization. Background Art

[0002] In a wind power generation system, due to the characteristics of high power density and high efficiency, a permanent magnet synchronous motor has become an optimal energy conversion device for the wind power generation system, especially suitable for an offshore wind power generation system with harsh environment. In the related art, the proportional-integral control method has a slow dynamic response speed. To overcome this problem, a finite set model predictive control is proposed. However, the finite set model predictive control has a large steady-state error and strong parameter dependence, and its robustness is weak. Summary of the Invention

[0003] To solve the deficiencies of the prior art, the purpose of the present application is to provide a continuous set model-free predictive control method for a permanent magnet motor based on Bayesian optimization. This method is not affected by the change of motor parameters and can improve the steady-state performance and dynamic characteristics of current predictive control.

[0004] In a first aspect, the present application provides a continuous set model-free predictive control method for a permanent magnet motor based on Bayesian optimization, the method comprising: Obtaining a super-local model and a pre-established extended state observer; wherein, the super-local model is determined based on the mathematical model of the permanent magnet synchronous motor in a two-phase rotating coordinate system; Determining the bandwidth of the extended state observer and the input gain of the super-local model as optimization variables, obtaining a training set and a test set for Bayesian optimization, the training set being obtained by sampling historical bandwidth and historical input gain through a sliding window; Calculating the joint probability distribution of the training set and the test set through a kernel function, and determining the predicted mean and predicted variance of the optimization variables; Screening combinations of optimization variables based on an expected value acquisition function, and updating the bandwidth of the extended state observer and the input gain of the super-local model based on the screened combinations of optimization variables; Based on the updated extended state observer and super-local model, and using the deadbeat control principle, calculating a reference voltage for the next control period; Decomposing the reference voltage into basic voltage vectors through space vector modulation, and controlling the inverter to execute the switching sequence of the basic voltage vectors according to the minimum number of switches to drive the permanent magnet motor.

[0005] In one embodiment, determining the super-local model based on the mathematical model of the permanent magnet synchronous motor in a two-phase rotating coordinate system includes: Establish a mathematical model of a permanent magnet synchronous motor in a two-phase rotating coordinate system; wherein, the mathematical model describes the current dynamic characteristics through the voltage equations of the d-axis and q-axis, and the mathematical model includes the following parameters: the d-axis and q-axis components of the stator resistance of the permanent magnet synchronous motor, the d-axis and q-axis components of the stator inductance of the permanent magnet synchronous motor, the mechanical speed of the rotor of the permanent magnet synchronous motor, and the permanent magnet flux linkage of the permanent magnet synchronous motor. When there are errors between the stator resistance in the mathematical model and the actual resistance value and / or between the stator inductance and the actual inductance, the mathematical model is converted into a super-local model through a first-order linear transformation.

[0006] In one embodiment, the super-local model is the product of the input gain and the input voltage of the permanent magnet synchronous motor plus the lumped disturbance; wherein, the input gain is affected by the stator inductance error, and the lumped disturbance includes at least one of the following: stator resistance error, external load change, and unmodeled dynamics.

[0007] In one embodiment, the establishment of the extended state observer includes: Define the current prediction error as the difference between the actual current of the permanent magnet synchronous motor and the estimated current value of the permanent magnet synchronous motor. Design the observer equation; wherein, the observer equation includes the product of the input gain and the input voltage of the permanent magnet synchronous motor, the lumped disturbance estimation, and the feedback compensation based on the current prediction error, and the response speed of the lumped disturbance estimation can be controlled by adjusting the size of the observer bandwidth.

[0008] In one embodiment, the method further includes: discretizing the super-local model and the extended state observer. Wherein, the discretized observer equation is: the current estimated value at the current moment is related to the current estimated value at the previous moment, the product of the input voltage of the permanent magnet synchronous motor and the input gain at the previous moment, the lumped disturbance estimation at the previous moment, and the feedback compensation based on the current prediction error at the previous moment. The discretized super-local model is: the current estimated value at the current moment is related to the current value at the previous moment, the lumped disturbance at the previous moment, and the product of the input voltage of the permanent magnet synchronous motor and the input gain at the previous moment.

[0009] In one embodiment, screening and optimizing variable combinations based on the expected value acquisition function includes: Calculate the expected performance improvement of all candidate optimization variable combinations, where the performance improvement is the difference between the predicted value and the current optimal value. Select the candidate optimization variable combination with the largest expected improvement as the optimal solution for the next control cycle. If the expected improvement of all candidate optimization variable combinations is negative, the current optimization variable combination remains unchanged.

[0010] In one embodiment, the reference voltage is calculated as follows: ; where represents a constant parameter, is the lumped disturbance estimate of the extended state observer, T is the time constant, represents the control period the reference voltage value at time represents the value of the input gain at control period k + P, represents at the control period the current reference value at time represents at the control period the current reference value at time represents at the control period the lumped disturbance estimate at time represents at the control period the current estimate at time

[0011] In one embodiment, the reference voltage is decomposed into basic voltage vectors through space vector modulation, and the inverter is controlled to execute the switching sequence of the basic voltage vectors according to the minimum number of switches to drive the permanent magnet motor, including: Converting the reference voltage from the rotating coordinate system to the stationary coordinate system; Determining the basic voltage vectors based on the position of the target sector; Calculating the action time of the basic voltage vectors; Based on the action time, controlling the inverter to execute the switching sequence of the basic voltage vectors according to the arrangement order of the minimum number of switches to drive the permanent magnet motor.

[0012] In a second aspect, the present application also provides a continuous model-free predictive control device for a permanent magnet motor based on Bayesian optimization. The control device includes: An acquisition module for acquiring a hyperlocal model and a pre-established extended state observer; wherein, the hyperlocal model is determined based on the mathematical model of the permanent magnet synchronous motor in the two-phase rotating coordinate system; An update module for determining the bandwidth of the extended state observer and the input gain of the hyperlocal model as optimization variables, acquiring a training set and a test set for Bayesian optimization, and the training set is obtained by sampling historical bandwidths and historical input gains through a sliding window; A prediction module for calculating the joint probability distribution of the training set and the test set through a kernel function and determining the predicted mean and predicted variance of the optimization variables; The screening module screens and optimizes variable combinations based on the expected value acquisition function, and updates the bandwidth of the extended state observer and the input gain of the hyperlocal model based on the screened optimized variable combinations; the screening module also calculates the reference voltage for the next control period based on the updated extended state observer and hyperlocal model and using the deadbeat control principle. The control module is used to decompose the reference voltage into basic voltage vectors through space vector modulation, and control the inverter to execute the switching sequence of the basic voltage vectors according to the minimum number of switches to drive the permanent magnet motor.

[0013] In a third aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the permanent magnet motor continuous set model-free predictive control method based on Bayesian optimization in the first aspect.

[0014] The above-mentioned permanent magnet synchronous motor continuous set model-free current predictive control method obtains a hyperlocal model and an extended state observer based on the mathematical model of the permanent magnet synchronous motor, determines the observer bandwidth and model input gain as optimization variables, and uses a sliding window to sample historical data to construct a training set required for Bayesian optimization. The joint probability distribution of the data is calculated through a kernel function, and the mean and variance of the optimization variables are predicted. The expected value acquisition function is used to screen the optimal variable combination and update the observer and model parameters. Based on the updated model and observer, the reference voltage for the next period is calculated in combination with the deadbeat control principle, and the reference voltage is decomposed into basic voltage vectors through space vector modulation, and the inverter is controlled to execute the switching sequence of the basic voltage vectors according to the minimum number of switches to drive the permanent magnet motor. This method uses an adaptively changing hyperlocal model and extended state observer to replace the traditional permanent magnet motor model for current predictive control, which is not affected by the change of the motor's own parameters and improves the steady-state performance and dynamic characteristics of current predictive control. Description of the Drawings

[0015] Figure 1 It is a flowchart of the permanent magnet motor continuous set model-free predictive control method based on Bayesian optimization in an embodiment; Figure 2 It is a flowchart of determining the hyperlocal model in an embodiment; Figure 3 It is a flowchart of the establishment of the extended state observer in an embodiment Figure 4 It is a flowchart of screening the optimized variable combination based on the expected value acquisition function in an embodiment; Figure 5 It is a flowchart of calculating the voltage vector in an embodiment; Figure 6 It is a schematic diagram of online updating the training set in an embodiment; Figure 7 It is a framework diagram of model-free current predictive control for a permanent magnet synchronous motor based on Bayesian optimization in an embodiment; Figure 8 It is a device diagram of model-free predictive control for a permanent magnet motor based on Bayesian optimization in an embodiment. Specific implementation manners

[0016] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. Unless otherwise defined, the technical terms or scientific terms involved in the present application shall have the general meanings understood by those with ordinary skills in the technical field to which the present application belongs.

[0017] In one embodiment, as Figure 1 shown, a model-free predictive control method for a permanent magnet motor based on Bayesian optimization is provided, and the method includes the following steps: Step 101: Obtain a hyper-local model and a pre-established extended state observer; wherein, the hyper-local model is determined based on the mathematical model of the permanent magnet synchronous motor in the two-phase rotating coordinate system; Establish the mathematical model of the permanent magnet synchronous motor in the two-phase rotating coordinate system, that is, the mathematical model in the d-q coordinate system. This model can describe the dynamic behavior of the permanent magnet synchronous motor, including the mutual relationships among parameters such as stator voltage, current, inductance, resistance, and permanent magnet flux linkage. Through the mathematical model of the permanent magnet synchronous motor in the two-phase rotating coordinate system, the characteristics of the permanent magnet synchronous motor in different operating states can be understood.

[0018] Furthermore, the hyper-local model is a simplified model. Through mathematical transformation, such as first-order affine transformation, key dynamic characteristics can be extracted from the mathematical model of the permanent magnet synchronous motor, and complex non-linear factors or high-order dynamics are ignored. The hyper-local model only retains the input-output relationship directly related to current control, and introduces parameters such as input gain to represent the dynamic characteristics of the permanent magnet synchronous motor.

[0019] The pre-established extended state observer can estimate the current state and external disturbances of the permanent magnet synchronous motor in real time. Among them, the extended state observer can estimate the internal state and external disturbances of the motor without relying on the motor model parameters. By feeding back the estimation result of the extended state observer to the hyper-local model, real-time monitoring of the state of the permanent magnet synchronous motor and dynamic compensation for disturbances can be achieved.

[0020] Step 102: Determine the bandwidth of the extended state observer and the input gain of the super-local model as optimization variables, and obtain the training set and test set for Bayesian optimization. The training set is obtained by sampling historical bandwidths and historical input gains through a sliding window. The bandwidth of the extended state observer and the input gain of the super-local model are optimization variables. To optimize the bandwidth and input gain, a dynamically updated training set needs to be constructed. Among them, the construction of the training set uses the sliding window technique for dynamic update. The length of the sliding window represents the control step size, ensuring that it always contains the latest state information of the permanent magnet synchronous motor.

[0021] Exemplarily, assume that the length of the sliding window is 5, and the current control period is the 10th period. The training set contains the historical bandwidths and historical input gains from the 6th to the 10th control periods. When the control period advances to the 11th period, the sliding window moves forward one position, and the training set is updated to the bandwidths and input gains from the 7th to the 11th control periods.

[0022] It should be noted that the upper limit, lower limit, and step size of the data test set are determined, and the test set remains unchanged all the time. Before Bayesian optimization, clarify the data range and sampling interval of the test set. Ensure that the test set remains unchanged during the subsequent optimization process, so as to provide a stable performance evaluation benchmark. The upper and lower limits of the test set define the data range, and the step size determines the interval between data points. Keeping the test set unchanged can ensure that evaluation biases introduced by changes in test data are avoided during the optimization process.

[0023] Step 103: Calculate the joint probability distribution of the training set and the test set through a kernel function, and determine the predicted mean and predicted variance of the optimization variables. Select a suitable kernel function, such as the radial basis kernel function (RBF). This kernel function can measure the similarity between data points in the training set and the test set, forming a kernel matrix. Using the kernel function, the data points in the training set can be mapped to a high-dimensional space. In the high-dimensional space, the distance between data points in the training set can reflect the similarity degree between data points.

[0024] In the high-dimensional space, the joint probability distribution of the training set and the test set can be calculated. The joint probability distribution can describe the possible value range of the test set data under the condition of the given training set data. Through the joint probability distribution, the predicted mean and predicted variance of the optimization variables can be further derived. Among them, the predicted mean can represent the expected value of predicting the test set data under the condition of the given training set data, and the predicted variance can represent the uncertainty or variability of the test set data.

[0025] Step 104: Screen and optimize the variable combination based on the expected value acquisition function, and update the bandwidth of the extended state observer and the input gain of the super-local model based on the screened optimized variable combination; The expected value acquisition function ( EI ) is an index that measures the possible improvement degree of a candidate parameter combination relative to the current optimal solution. The expected value acquisition function can comprehensively consider the prediction mean and prediction variance, paying attention to both potential performance improvements and uncertainties.

[0026] During the optimization process, the joint probability distribution of the training set and the test set calculated using the kernel function is used to derive the prediction mean and prediction variance of the optimization variables. Based on the prediction mean and prediction variance, an expected value acquisition function is constructed, and the potential improvement value of each candidate parameter combination is evaluated through the expected value acquisition function. Further, the expected value acquisition function can calculate an expected value for each candidate point, and the expected value can reflect the expected improvement degree of the candidate point on the optimization objective. The optimization algorithm screens out the parameter combination with the highest expected improvement by comparing the expected values.

[0027] The screened optimized variable combination is the optimal value of the bandwidth of the extended state observer and the input gain of the super-local model. The bandwidth of the extended state observer and the input gain of the super-local model can affect the dynamic response and robustness of the permanent magnet synchronous motor. Further, the screened optimal parameter combination is applied to the extended state observer and the super-local model to update the bandwidth of the extended state observer and the input gain of the super-local model. This update process enables the permanent magnet synchronous motor to adaptively adjust the parameters to better adapt to the current operating state and external disturbances.

[0028] Step 105: Based on the updated extended state observer and super-local model, and using the deadbeat control principle, calculate the reference voltage applied to the next control cycle; The updated extended state observer can estimate the current state and external disturbances of the permanent magnet synchronous motor in real time. The estimated values of the current state and external disturbances of the permanent magnet synchronous motor can reflect the actual operating conditions and external interference conditions of the permanent magnet synchronous motor at the current moment. The super-local model can predict the reference voltage of the permanent magnet synchronous motor in the next control cycle based on the estimated values provided by the extended state observer and combined with the dynamic characteristics of the permanent magnet synchronous motor.

[0029] It should be noted that the deadbeat control principle is a prediction-based control method. Its core lies in using the optimized variable combination of the current cycle to accurately predict the state of the permanent magnet synchronous motor in the next control cycle, enabling the control command to quickly and accurately track the expected value, thereby achieving a control without steady-state error. Exemplarily, if the current control cycle is the 10th control cycle, the reference voltage of the 11th control cycle can be predicted using the optimized variable combination of the 10th control cycle.

[0030] During this process, the hyperlocal model uses the current estimation value and the lumped disturbance value provided by the extended state observer, combines with the hyperlocal model, and calculates the reference voltage required for the next control period. Among them, the calculation of the reference voltage takes into account the dynamic characteristics of the motor, the current state, and the desired current trajectory to ensure that the current of the permanent magnet synchronous motor quickly and accurately tracks the desired current value in the next control period.

[0031] Step 106: Decompose the reference voltage into basic voltage vectors through space vector modulation, and control the inverter to execute the switching sequence of the basic voltage vectors according to the minimum number of switches to drive the permanent magnet motor.

[0032] In a three-phase inverter, there are 8 basic voltage vectors, which are composed of different on-off states of each switch tube in the inverter. Among them, 6 are non-zero basic voltage vectors and 2 are zero basic voltage vectors. The reference voltage is converted from the rotating coordinate system to the stationary coordinate system, and the components of the reference voltage in the stationary coordinate system are calculated. The sector where the reference voltage is located is judged according to the position corresponding to the components of the reference voltage in the stationary coordinate system. Among them, each sector corresponds to specific two adjacent non-zero basic voltage vectors and zero basic voltage vectors. According to the volt-second balance principle, the action time of the two adjacent non-zero basic voltage vectors and the zero basic voltage vector in a sampling period is calculated to ensure that the synthesized voltage of the two adjacent non-zero basic voltage vectors and the zero basic voltage vector in this period is equivalent to the reference voltage.

[0033] After determining the two adjacent non-zero basic voltage vectors and the zero basic voltage vector and their action times, in order to reduce the switching loss and electromagnetic interference of the inverter switching devices, the switching sequence of the basic voltage vectors is arranged according to the principle of the minimum number of switches. That is, in two adjacent switches, try to change the state of only one switch tube. For example, when switching from one basic voltage vector to another basic voltage vector, only let one switch tube change from on to off, or from off to on. Further, according to the arranged switching sequence and the corresponding action time, the inverter is driven to output the basic voltage vectors in turn. The voltage waveform output by the inverter will approximate the waveform of the reference voltage, and then generate a suitable three-phase AC voltage to drive the permanent magnet synchronous motor.

[0034] In this embodiment, the method determines the observer bandwidth and the model input gain as optimization variables by obtaining a super-local model and an extended state observer based on the mathematical model of a permanent magnet synchronous motor, and uses sliding window sampling of historical data to construct a training set required for Bayesian optimization. The joint probability distribution of the data is calculated through a kernel function, and the mean and variance of the optimization variables are predicted. The expected value acquisition function is used to screen the optimal variable combination, and the observer and model parameters are updated. Based on the updated model and observer, the reference voltage for the next cycle is calculated in combination with the deadbeat control principle, and the reference voltage is decomposed into basic voltage vectors through space vector modulation, and the inverter is controlled to execute the switching sequence of the basic voltage vectors according to the minimum number of switches to drive the permanent magnet motor. This method uses an adaptively varying super-local model and an extended state observer to replace the traditional permanent magnet motor model for predictive control of current, which is not affected by changes in the motor's own parameters and improves the steady-state performance and dynamic characteristics of the current predictive control.

[0035] In one embodiment, as Figure 2 shown, determining a super-local model based on the mathematical model of a permanent magnet synchronous motor in a two-phase rotating coordinate system includes the following steps: Step 201: Establish the mathematical model of the permanent magnet synchronous motor in a two-phase rotating coordinate system; wherein, the mathematical model describes the current dynamic characteristics through the voltage equations of the d-axis and q-axis, and the mathematical model includes the following parameters: the d-axis and q-axis components of the stator resistance of the permanent magnet synchronous motor, the d-axis and q-axis components of the stator inductance of the permanent magnet synchronous motor, the mechanical speed of the rotor of the permanent magnet synchronous motor, and the permanent magnet flux linkage of the permanent magnet synchronous motor; The operating principle of a permanent magnet synchronous motor is based on the law of electromagnetic induction and the basic laws of circuits. The law of electromagnetic induction and the basic laws of circuits can describe the electromagnetic phenomena and circuit behaviors inside the permanent magnet synchronous motor, so that the operating characteristics of the permanent magnet synchronous motor can be described by mathematical equations. In a two-phase rotating coordinate system (d-q coordinate system), the d-axis, also known as the direct axis, is aligned with the direction of the permanent magnet flux linkage of the rotor of the permanent magnet synchronous motor, representing the flux direction of the permanent magnet synchronous motor. The q-axis, also known as the quadrature axis, is perpendicular to the d-axis and represents the torque generation direction of the permanent magnet synchronous motor. In the d-q coordinate system, physical quantities such as voltage, current, and flux linkage of the permanent magnet synchronous motor can be decomposed into components on the d-axis and q-axis. By controlling the currents on the d-axis and q-axis respectively, the flux and torque of the permanent magnet synchronous motor can be independently controlled, thus achieving precise control of the permanent magnet synchronous motor.

[0036] By applying the law of electromagnetic induction and the basic laws of circuits, the voltage equations of the permanent magnet synchronous motor on the d-axis and q-axis can be derived: .

[0037] The d-axis voltage equation describes the rate of change of the d-axis current, which is mainly affected by the d-axis resistance, the d-axis inductance, and the q-axis current. The q-axis voltage equation describes the rate of change of the q-axis current, which is mainly affected by the q-axis resistance, the q-axis inductance, the d-axis current, and the permanent magnet flux linkage.

[0038] Among them, and respectively represent the d-axis and q-axis components of the voltage at the stator output terminal of the permanent magnet synchronous motor; represents the stator resistance of the permanent magnet synchronous motor; and respectively represent the d-axis and q-axis components of the stator current of the permanent magnet synchronous motor; and respectively represent the d-axis and q-axis components of the stator inductance of the permanent magnet synchronous motor; represents the mechanical speed of the rotor of the permanent magnet synchronous motor; represents the permanent magnet flux linkage of the permanent magnet synchronous motor.

[0039] Step 202: When there is an error between the stator resistance in the mathematical model and the actual resistance value and / or an error between the stator inductance and the actual inductance, the mathematical model is converted into a super-local model through a first-order linear transformation. The super-local model is the product of the input gain and the input voltage of the permanent magnet synchronous motor plus the lumped disturbance; among them, the input gain is affected by the stator inductance error, and the lumped disturbance includes at least one of the following: stator resistance error, external load change, and unmodeled dynamics.

[0040] When the stator inductance and resistance parameters of the permanent magnet motor do not match the actual values, the mathematical model of the voltage on the d-axis and q-axis of the permanent magnet synchronous motor is: .

[0041] Among them, and respectively represent the parameter errors of the stator resistance and the stator inductance.

[0042] Furthermore, using a first-order affine transformation, that is, rearranging the mathematical model, the time derivative of the current is obtained as: .

[0043] Simplify the coefficients in the above equation, introduce new input gains α d , α q , and the lumped disturbance F d , F q . Among them, , , = , = . Further, by using a first-order affine transformation and simplifying the coefficients, the obtained hyper-local model is as follows: ; where and represent the model input gain that changes due to the perturbation of the inductance parameter; and represent the lumped perturbation.

[0044] In one embodiment, as shown in Figure 3 , the establishment of the extended state observer includes the following steps: Step 301: Define the current prediction error as the difference between the actual current of the permanent magnet synchronous motor and the estimated value of the current of the permanent magnet synchronous motor; ; where is the estimated value of the current of the permanent magnet synchronous motor, is the actual current of the permanent magnet synchronous motor, is the current prediction error of the extended state observer. Further, = , T , is the d-axis current error. is the q-axis current error. T is the transpose operator, which is used to transpose the matrix, = , T means is a column vector composed of and .

[0045] Step 302: Design the observer equation; where the observer equation includes the product of the input gain and the input voltage of the permanent magnet synchronous motor, the lumped perturbation estimation, and the feedback compensation based on the current prediction error, and the response speed of the lumped perturbation estimation can be controlled by adjusting the size of the observer bandwidth.

[0046] The design of the extended state observer equation based on the lumped perturbation is as follows: .

[0047] where is the voltage reference value, is the current prediction error of the extended state observer; is the estimated value of the current; is the estimated value of the lumped perturbation; and ​​are the feedback gains of the observer respectively; represents the bandwidth of the extended state observer, is the input gain.

[0048] In this embodiment, through this design, the observer can estimate the external disturbance in real time and perform feedback compensation, thereby improving the dynamic performance of current control.

[0049] In one embodiment, the method further includes: discretizing the super-local model and the extended state observer; wherein, the observer equation after discretization is: the current estimated value at the current moment is related to the current estimated value at the previous moment, the product of the permanent magnet synchronous motor input voltage at the previous moment and the input gain, the lumped disturbance estimate at the previous moment, and the feedback compensation based on the current prediction error at the previous moment; the super-local model after discretization is: the current estimated value at the current moment is related to the current value at the previous moment, the lumped disturbance at the previous moment, and the product of the permanent magnet synchronous motor input voltage at the previous moment and the input gain.

[0050] After discretizing the extended observer equation, we can obtain: .

[0051] Wherein, is the current prediction error at the current moment k is the current estimated value at the current moment k, is the current estimated value at the moment k + 1. is the actual current at the current moment k. is the lumped disturbance estimate at the current moment k, is the lumped disturbance estimate at the moment k + 1. is the input gain at the current moment k. is the permanent magnet synchronous motor voltage reference value at the current moment k. is the bandwidth of the extended state observer at the current moment k, is a single control period.

[0052] After discretizing the first-order super-local model, we get: .

[0053] Wherein, is the actual current at the current moment k, is the actual current at the moment k + 1. is the input gain at the current moment k, is the permanent magnet synchronous motor voltage reference value at the current moment k. is the lumped disturbance at the current moment k, is a single control period.

[0054] It should be noted that both the observer equation and the super-local model after discretization consider the current estimation value at the previous moment, which can maintain the continuity and stability of current estimation. The product of the input gain and the input voltage of the permanent magnet synchronous motor at the previous moment can reflect the influence of the motor input voltage on the current and is used to predict the current change at the current moment. The lumped disturbance estimation can reflect various disturbance factors, such as stator resistance error, external load change, and unmodeled dynamics, etc. By considering the lumped disturbance estimation at the previous moment, the accuracy of current estimation can be improved. The current prediction error is used for feedback compensation to adjust the output of the observer and improve the estimation accuracy. The current estimation error can be reduced through feedback compensation.

[0055] In one embodiment, the method for online updating the test set and training set of Bayesian optimization is as follows: Determine the upper limit, lower limit, and step size of the test set, and keep the test set unchanged all the time. Whenever the control timing moves backward once, the data in the training set moves forward one position in sequence. At this time, the data with the initial subscript of 1 will be discarded and covered by the data with the subscript of 2. And so on until the number of data shifts is equal to the control step size; then store the optimization parameters equal to the number of control step sizes collected during the shifting period into the corresponding time-sequence data bits of the training set one by one. The expression is as follows: .

[0056] Where, = , represents the training set, and its size is t ×3; represents the acquisition data storage matrix, and its size is t ×3; P is the control step size.

[0057] Furthermore, use the radial basis kernel function (RBF) to calculate the similarity between each data point in the updated training set and test set. The kernel function formula is: .

[0058] Where and respectively represent two associated vectors, is a constant parameter. The higher the similarity, the closer the value of the kernel function is to 1; the lower the similarity, the closer the value of the kernel function is to 0. It should be noted that, and can be expressed as the training set and the data set. Using the kernel function, the similarity between the training set and the data set can be calculated, which provides a basis for the formation of the subsequent joint probability.

[0059] The control objective of the super-local model is to minimize the current tracking error. The randomly selected training set data can be considered to form a high-dimensional normal distribution with the tracking error: .

[0060] Among them, represents the control objective corresponding to the training set data, the reciprocal of the cumulative error of steady-state current tracking; and respectively represent the predicted mean and predicted variance corresponding to the training set data; is the set number of integral periods of the cumulative error; is the reference current value; is the actual current value. represents the input gain and the observer bandwidth is a function of, following a Gaussian distribution with mean and covariance matrix , reflecting the predicted distribution of performance under different and combinations. is a function of the input gain , representing the predicted value of performance under different input gains. is a function of the observer bandwidth ω 0, representing the predicted value of performance under different observer bandwidths. represents the current control period. represents the step size. represents the offset of the time window. represents the input gain from control period to value. represents the observer bandwidth from control period to value.

[0061] It should be noted that assuming there is a high-dimensional normal distribution relationship between the training set data and the control objective, that is, it means that under different control periods, the combination of the input gain and the observer bandwidth has a certain statistical law with the corresponding control objective f . The predicted mean and the predicted variance can describe the central tendency and dispersion degree of the training set data, providing a basis for subsequent Bayesian optimization. The parameter r can control the integral period of the cumulative error and can affect the calculation of the control objective.

[0062] Furthermore, as the control time increases, the steady-state error under a fixed control instruction will gradually decrease and approach 0. Therefore, the high-dimensional normal distribution of the training set can be approximated as a zero-mean normal distribution. Similarly, the test set data also satisfies the zero-mean normal distribution.

[0063] Furthermore, through the kernel function, the joint probability distribution of the training set and the test set can be constructed. The joint probability distribution can describe the possible value range of the test set data given the training set data. Among them, both the training set and the test set are constructed according to the operating law of the normal distribution, and their joint probability distribution also conforms to the law of the normal distribution. According to the definition of the normal joint distribution, the joint probability distribution of the training set and the test set can be obtained. The mean value in this joint probability distribution is 0 (the target expectation of optimization); the variance is a matrix composed of the kernel function K. The mathematical expression of the joint probability distribution is: 。

[0064] Where represents the control target corresponding to the test set data. denotes the training data set X and the kernel function matrix between itself, which is used to measure the similarity between the training set data points. denotes the training data set and the test data set between the kernel function matrix, which is used to measure the similarity between the training set data points and the test set data points. denotes the test data set and the kernel matrix between itself, which is used to measure the similarity between the test set data points.

[0065] Furthermore, by using the calculation formula of the standard zero-mean normal distribution for the joint probability distribution of the training set and the test set, the target optimization parameters, that is, the prior Gaussian distributions of the input gain and bandwidth, can be obtained: 。

[0066] Where represents the predicted mean value of the prior Gaussian distribution, and

[0067] is the standard deviation of the prior Gaussian distribution. and value judgment criterion: 。

[0068] Where represents all possible Xto find the one that maximizes the objective function (such as the expected value acquisition function or the performance metric reaches the maximum value X . , represents the training set data set; represents the maximum value in the training set objective function vector at the current moment; represents the newly added data in the training data set after the start of the next step when the optimization iteration loop has not terminated; then represents the optimal parameter combination obtained through Bayesian optimization at the current iteration. Calculate the values of all candidate optimization variable combinations, and select the combination with the largest

[0069] value as the optimal solution for the next control cycle. Further, update the bandwidth of the extended state observer and the input gain of the super-local model based on the selected optimization variable combination. The process of screening and optimizing variable combinations by the expected value acquisition function is as follows: the maximum value in the objective function screened out within the training set and its corresponding optimal parameter set and are brought into the calculation formula of the function for calculation. Further, can be introduced to transform the training set, and the formula is simplified to a calculation formula of a normal distribution with respect to . The calculated result is the expected value of the optimization objective at each data point of the test set , and this process requires the participation of the results in the previously calculated joint Gaussian distribution. Take the maximum value of the data vector calculated, and the data result in the test set corresponding to this maximum value is the selected optimization result. The specific calculation formula of the expected acquisition function EI function is as follows: .

[0070] Among them, is the standard normal probability density function; represents the distribution function of the standard normal distribution. Apply the selected optimal optimization variable combination to the extended state observer and the super-local model, and update their bandwidth and input gain.

[0071] In one embodiment, such as Figure 4As shown, the optimization variable combination is screened based on the expected value acquisition function, including the following steps: Step 401: Calculate the expected performance improvement of all candidate optimization variable combinations. The performance improvement is the difference between the predicted value and the current optimal value. For each candidate combination, use the expected value acquisition function to predict its performance value; compare this predicted value with the currently known optimal performance value, and calculate the difference between each candidate group. This difference is the expected performance improvement of the candidate combination.

[0072] Step 402: Select the candidate optimization variable combination with the largest expected improvement as the optimal solution for the next control cycle. Select the candidate optimization variable combination with the largest expected improvement as the optimal solution for the next control cycle, which means that at the end of each control cycle, from all possible optimization variable combinations, select the combination that is most likely to bring the greatest performance improvement under the predicted situation. This combination can be applied to the next control cycle to achieve a significant improvement in control performance.

[0073] Step 403: If the expected improvement of all candidate optimization variable combinations is negative, keep the current optimization variable combination unchanged.

[0074] If the expected performance improvement of all calculated candidate combinations is negative, it means that under the predicted situation, none of the candidate combinations can bring performance improvement, but may even lead to performance degradation. To ensure the stability and control performance of the system, the current optimization variable combination can be selected to remain unchanged, and the current parameters can continue to be used for control until a new candidate combination that can bring performance improvement is found in the subsequent optimization process.

[0075] In this embodiment, this method can effectively improve the optimization efficiency and effect, ensure the stability and optimality of performance, and avoid performance degradation caused by blind attempts.

[0076] In one embodiment, the calculation method of the reference voltage is: ; where is the voltage reference value, is the lumped disturbance estimation value of the extended state observer, represents the control period, represents the step size, is the time constant, used to adjust the integral part of the control input.

[0077] represents the reference voltage value at the control period ; represents the value of the input gain at the control period ; represents the current reference value at the control period represents the current reference value at the control period represents the lumped disturbance estimation value at the control period represents the current estimation value at the control period

[0078] Calculate the current reference value for the next control period and the current estimation value for the current control period to find the difference. Adjust the current error using the proportional control gain to quickly respond to current changes. Subtract the disturbance estimation value for the current period to compensate for disturbance interference. Calculate the rate of change of the current error by dividing by the control period . Divide the above items by the input gain for the current period to obtain the final reference voltage value .

[0079] In this embodiment, by combining proportional control, disturbance compensation, and the rate of change of current error, this formula can accurately calculate the reference voltage value required for a permanent magnet synchronous motor to achieve precise current control.

[0080] In one embodiment, as Figure 5 shown, the reference voltage is decomposed into the action times of the basic voltage vectors through space vector modulation, and the inverter output is controlled according to the principle of the minimum number of switches, including the following steps: Step 501: Convert the reference voltage from the rotating coordinate system to the stationary coordinate system; The reference voltage is represented as and in the rotating coordinate system (d-q coordinate system). Through the inverse Clark transformation, it can be converted to and in the stationary coordinate system (α-β coordinate system). The formula for the inverse Clark transformation is as follows: .

[0081] where is the rotor position angle, representing the current position of the rotor of the permanent magnet synchronous motor, Step 502: Determine the basic voltage vectors based on the position of the target sector; Calculate the angle of the reference voltage in the stationary coordinate system (α-β coordinate system) using the arctangent function arctan φ , the formula is: ; Since the value range of the function is (−2 π , 2 π ), to obtain the angle within the complete [0, 2 π ) range, it is necessary to correct it according to the signs of and . The specific rules are as follows: When > 0 and ≥ 0, ; When < 0, ; When > 0 and < 0, .

[0082] Determine the sector where the reference voltage is located according to the angle of the reference voltage in the stationary coordinate system. Among them, the coordinate system is divided into six sectors, each sector corresponding to a certain angle range, and each sector corresponding to specific two adjacent non-zero basic voltage vectors and zero basic voltage vectors. Among them, the basic voltage vectors include V 0(000), V 1(100), V 2(110), V 3(010), V 4(011), V 5(001), V 6(101) and V 7(111), V 0 and V 7 are zero basic voltage vectors, and the remaining 6 basic voltage vectors are non-zero basic voltage vectors. The entire voltage space is divided into 6 sectors by 6 non-zero basic voltage vectors. Based on the sector where the reference voltage is located, determine the two adjacent non-zero basic voltage vectors and zero basic voltage vectors in this sector as the required basic voltage vectors.

[0083] Step 503: Calculate the action time of the basic voltage vector; Exemplarily, taking u s in the first sector as an example: .

[0084] Among them, is the amplitude of the reference voltage vector in the two-phase stationary coordinate system; is the DC-side voltage of the frequency converter. Among them, and are two adjacent non-zero basic voltage vectors.

[0085] Furthermore, the action time of each basic voltage vector in the next control period is: .

[0086] Among them, the modulation index represents the ratio of the reference voltage amplitude to the DC-side voltage , and is used to control the magnitude of the output voltage. The modulation index usually ranges from 0 to 1, indicating the ratio of the output voltage to the DC voltage. represents the action time of the non-zero basic voltage vector V1, represents the action time of the non-zero basic voltage vector V2. represents the action time of the zero basic voltage vector V0 or V7.

[0087] It should be noted that according to the volt-second balance principle, within a sampling period T s , the volt-second product of the reference voltage vector is equal to the sum of the volt-second products of each basic voltage vector, that is: VT s = V x T x + V y T y + V 0 T 0.

[0088] Among them, and are two adjacent non-zero basic voltage vectors, and are the action times corresponding to the two adjacent non-zero basic voltage vectors respectively. Here, the two adjacent non-zero basic voltage vectors can be V 1 and V 2; V 0 is the zero vector ( V 0 or V 7), is the action time of the zero basic voltage vector, and satisfies + + = .

[0089] Step 504: Based on the action time, control the inverter to execute the switching sequence of the basic voltage vectors in the arrangement order of the minimum number of switchings to drive the permanent magnet motor.

[0090] To reduce the loss of the switching devices in the inverter, the switching sequence of the basic voltage vectors is arranged according to the principle of the minimum number of switchings. In the switching between two adjacent basic voltage vectors, try to change the state of only one switching tube. For example, when originally in the switching state corresponding to a certain basic voltage vector and switching to the next basic voltage vector, only let one switching tube change from conduction to cutoff, or from cutoff to conduction.

[0091] The inverter outputs the basic voltage vectors in sequence according to the arranged minimum number of switchings and the corresponding action time. The voltage waveform output by the inverter will approximate the waveform of the reference voltage vector, thereby generating appropriate three-phase AC voltage.

[0092] Exemplarily, assuming that the reference voltage vector is located in sector I, the corresponding non-zero basic voltage vectors are V 1 and V 2, and the zero basic voltage vectors are V 0 and V 7. After calculating the action times of V 1, V 2, V 0, and V 7, the switching sequence is determined according to the principle of the minimum number of switchings as: V 0 - V 1 - V 2 - V 7 - V 7 - V 1 - V 0. The inverter first outputs the action time corresponding to V 0, then outputs the action times corresponding to V 1, V 2, V 7, etc., and continuously loops this process to drive the permanent magnet synchronous motor to operate.

[0093] In one embodiment, the structure diagrams of the test set and training set for online updating of Bayesian optimization are as shown in Figure 6 shown. The training set (storing historical input gains and bandwidths), the test set (with a fixed value range for verification), and the data acquisition and storage module (real-time recording of operating data such as current and speed) implement data management.

[0094] The training set data is updated using a sliding window mechanism. Taking the current control moment as the benchmark on the time axis, the control step size To optimize the interval, the window size of the training set is (the length of stored historical data), and the prediction period is , where the data at the moment will be discarded, the data from the to moment is used as training data. The data at the latest k moment is added to the storage, and the

[0095] data flow moves in the direction of discarded data (old data beyond the window) → training data (valid historical data) → measurement data (latest real-time data) → prediction data (future state prediction). Through the calculation iteration loop, the closed-loop process of "model update → test verification → state prediction → parameter optimization → control implementation → new data acquisition" is continuously executed. This incremental update mechanism not only ensures data timeliness but also improves computational efficiency, and can dynamically track parameter changes. Among them, the ultra-local model provides a dynamic simplified representation, and the extended state observer estimates the unmeasurable state and disturbance in real time. The two work together to ensure excellent control performance and robustness under various working conditions.

[0096] In one embodiment, the framework diagram of the continuous set model-free current prediction control of a permanent magnet synchronous motor based on Bayesian optimization is as shown in Figure 7 , and is used for the current control of a permanent magnet synchronous motor (PMSM). It includes modules such as an ultra-local model, an EI acquisition function, a Gaussian process model, an SVPWM frequency converter, state observation, and a PI controller.

[0097] The current measurement module acquires the phase currents , , of the permanent magnet synchronous motor, and the encoder measures the rotational speed and the rotor position θ of the permanent magnet synchronous motor. Then, the three-phase currents are converted , , into the current in the coordinate system. The coordinate transformation from αβ to coordinates requires converting the voltage in the αβ coordinate system to the voltage in the coordinate system. The extended state observer aggregates the disturbance estimation value and the current estimation value . The ultra-local model receives the optimization variables and , and outputs the voltage component on the 。The Gaussian process model makes predictions based on historical and current data and outputs the predicted mean and the predicted standard deviation 。The EI acquisition function receives the predicted mean and the predicted standard deviation ,calculates the expected performance improvement for each candidate optimization variable combination, and screens out the optimal combination. The frequency converter converts the reference voltage into an accurate PWM signal through the SVPWM control module, controls the switching state of the inverter, and thus adjusts the supply voltage and frequency of the motor. Control the output of the inverter. The PI controller adjusts the current reference value through the rotational speed of the permanent magnet synchronous motor and the rotational speed reference value to ensure the accuracy of current tracking.

[0098] Based on the same inventive concept, the embodiment of the present application also provides a permanent magnet motor continuous set model-free predictive control device based on Bayesian optimization. The implementation solution provided by this device to solve the problem is similar to the implementation solution described in the above method. Therefore, the specific limitations in one or more embodiments of the permanent magnet motor continuous set model-free predictive control based on Bayesian optimization provided below can refer to the limitations on the permanent magnet motor continuous set model-free predictive control method based on Bayesian optimization in the above text, and will not be repeated here.

[0099] In one embodiment, as Figure 8 shown, the embodiment of the present application also provides a permanent magnet motor continuous set model-free predictive control device based on Bayesian optimization. This control device includes: An acquisition module 801, configured to acquire a hyperlocal model and a pre-established extended state observer; wherein, the hyperlocal model is determined based on the mathematical model of the permanent magnet synchronous motor in the two-phase rotating coordinate system; An update module 802, using the bandwidth of the determined extended state observer and the input gain of the hyperlocal model as optimization variables, acquires a training set and a test set for Bayesian optimization. The training set is obtained by sampling historical bandwidth and historical input gain through a sliding window; A prediction module 803, configured to calculate the joint probability distribution of the training set and the test set through a kernel function, and determine the predicted mean and predicted variance of the optimization variables; A screening module 804, screening the optimization variable combination based on the expected value acquisition function, and updating the bandwidth of the extended state observer and the input gain of the hyperlocal model based on the screened optimization variable combination; the screening module also calculates the reference voltage acting on the next control cycle based on the updated extended state observer and hyperlocal model and using the deadbeat control principle; A control module 805 is configured to decompose a reference voltage into basic voltage vectors through space vector modulation, and control an inverter to execute a switching sequence of the basic voltage vectors according to the minimum number of switches, so as to drive a permanent magnet motor.

[0100] In one embodiment, an acquisition module 801 determines a super-local model based on a mathematical model of a permanent magnet synchronous motor in a two-phase rotating coordinate system. Specifically, the acquisition module 801 is configured to: establish a mathematical model of a permanent magnet synchronous motor in a two-phase rotating coordinate system; wherein, the mathematical model describes the current dynamic characteristics through voltage equations of the d-axis and q-axis, and the mathematical model includes the following parameters: d-axis and q-axis components of the stator resistance of the permanent magnet synchronous motor, d-axis and q-axis components of the stator inductance of the permanent magnet synchronous motor, the mechanical speed of the rotor of the permanent magnet synchronous motor, and the permanent magnet flux linkage of the permanent magnet synchronous motor; when there is an error between the stator resistance in the mathematical model and the actual resistance value and / or an error between the stator inductance and the actual inductance, the mathematical model is converted into a super-local model through a first-order linear transformation.

[0101] In one embodiment, the acquisition module 801 takes the super-local model as the product of an input gain and the input voltage of the permanent magnet synchronous motor plus a lumped disturbance; wherein, the input gain is affected by the stator inductance error, and the lumped disturbance includes at least one of the following: stator resistance error, external load change, and unmodeled dynamics.

[0102] In one embodiment, the acquisition module 801 defines a current prediction error as the difference between the actual current of the permanent magnet synchronous motor and the current estimated value of the permanent magnet synchronous motor; and designs an observer equation; wherein, the observer equation includes the product of the input gain and the input voltage of the permanent magnet synchronous motor, a lumped disturbance estimation, and a feedback compensation based on the current prediction error, and the response speed of the lumped disturbance estimation can be controlled by adjusting the size of the observer bandwidth.

[0103] In one embodiment, the acquisition module 801 discretizes the super-local model and the extended state observer; wherein, the discretized observer equation is: the current estimated value at the current moment is related to the current estimated value at the previous moment, the product of the input voltage of the permanent magnet synchronous motor and the input gain at the previous moment, the lumped disturbance estimation at the previous moment, and a feedback compensation based on the current prediction error at the previous moment; the discretized super-local model is: the current estimated value at the current moment is related to the current value at the previous moment, the lumped disturbance at the previous moment, and the product of the input voltage of the permanent magnet synchronous motor and the input gain at the previous moment.

[0104] In one embodiment, the screening module 804 screens the optimized variable combinations based on the expected value acquisition function, specifically for: calculating the expected performance improvement of all candidate optimized variable combinations, where the performance improvement is the difference between the predicted value and the current optimal value; selecting the candidate optimized variable combination with the largest expected improvement as the optimal solution for the next control period; if the expected improvement of all candidate optimized variable combinations is negative, maintaining the current optimized variable combination unchanged.

[0105] In one embodiment, the calculation formula for the reference voltage calculated by the screening module 804 is: ; where represents a constant parameter, is the lumped disturbance estimation value of the extended state observer, T is the time constant, represents the reference voltage value at the control period , represents the value of the input gain at the control period , represents the current reference value at the control period , represents the current reference value at the control period , represents the lumped disturbance estimation value at the control period , represents the current estimation value at the control period .

[0106] In one embodiment, the control module 805 decomposes the reference voltage into basic voltage vectors through space vector modulation and controls the inverter to execute the switching sequence of the basic voltage vectors according to the minimum number of switches to drive the permanent magnet motor, specifically for: converting the reference voltage from the rotating coordinate system to the stationary coordinate system; determining the basic voltage vectors based on the position of the target sector; calculating the action time of the basic voltage vectors; controlling the inverter to execute the switching sequence of the basic voltage vectors according to the arrangement order of the minimum number of switches based on the action time to drive the permanent magnet motor.

[0107] Based on the same concept, the present application also provides a computer-readable storage medium, on which a computer program is stored, and the computer program is executed by a processor to implement the model-free predictive control method for continuous sets of permanent magnet motors based on Bayesian optimization.

[0108] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0109] The above embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A model - free predictive control method for a permanent - magnet motor continuous set based on Bayesian optimization, characterized in that, The method includes: Obtaining a hyper-local model and a pre-established extended state observer; wherein, the hyper-local model is determined based on the mathematical model of the permanent magnet synchronous motor in the two-phase rotating coordinate system; Determining the bandwidth of the extended state observer and the input gain of the hyper-local model as optimization variables, obtaining a training set and a test set for Bayesian optimization, and the training set is obtained by sampling historical bandwidths and historical input gains through a sliding window; Calculating the joint probability distribution of the training set and the test set through a kernel function, and determining the predicted mean and predicted variance of the optimization variables; Screening the optimization variable combinations based on the expected value acquisition function, and updating the bandwidth of the extended state observer and the input gain of the hyper-local model based on the screened optimization variable combinations; Based on the updated extended state observer and the hyper-local model, and using the deadbeat control principle, calculating the reference voltage applied to the next control cycle; Decomposing the reference voltage into basic voltage vectors through space vector modulation, and controlling the inverter to execute the switching sequence of the basic voltage vectors according to the minimum number of switches to drive the permanent magnet motor.

2. The model - free predictive control method for a permanent - magnet motor continuous set based on Bayesian optimization according to claim 1, characterized in that, Determining the hyper-local model based on the mathematical model of the permanent magnet synchronous motor in the two-phase rotating coordinate system, including: Establishing the mathematical model of the permanent magnet synchronous motor in the two-phase rotating coordinate system; wherein, the mathematical model describes the current dynamic characteristics through the voltage equations of the d-axis and q-axis, and the mathematical model includes the following parameters: the d-axis and q-axis components of the stator resistance of the permanent magnet synchronous motor, the d-axis and q-axis components of the stator inductance of the permanent magnet synchronous motor, the mechanical speed of the rotor of the permanent magnet synchronous motor, and the permanent magnet flux linkage of the permanent magnet synchronous motor; When there is an error between the stator resistance in the mathematical model and the actual resistance value and / or an error between the stator inductance and the actual inductance, converting the mathematical model into the hyper-local model through a first-order linear transformation.

3. The model - free predictive control method for a permanent - magnet motor continuous set based on Bayesian optimization according to claim 2, characterized in that, The hyper-local model is the product of the input gain and the input voltage of the permanent magnet synchronous motor plus a lumped disturbance; wherein, the input gain is affected by the stator inductance error, and the lumped disturbance includes at least one of the following: stator resistance error, external load change, and unmodeled dynamics.

4. The model - free predictive control method for a permanent - magnet motor continuous set based on Bayesian optimization according to claim 1, characterized in that, The establishment of the extended state observer includes: Defining the current prediction error as the difference between the actual current of the permanent magnet synchronous motor and the estimated current value of the permanent magnet synchronous motor; Designing an observer equation; wherein, the observer equation includes the product of the input gain and the input voltage of the permanent magnet synchronous motor, the lumped disturbance estimation, and the feedback compensation based on the current prediction error, and the response speed of the lumped disturbance estimation can be controlled by adjusting the size of the observer bandwidth.

5. The model - free predictive control method for a permanent - magnet motor continuous set based on Bayesian optimization according to any one of claims 1 to 4, characterized in that, The method further includes: discretizing the hyper-local model and the extended state observer; Wherein, the discretized observer equation is: the current estimated value at the current moment is related to the current estimated value at the previous moment, the product of the input voltage of the permanent magnet synchronous motor and the input gain at the previous moment, the lumped disturbance estimation at the previous moment, and the feedback compensation based on the current prediction error at the previous moment; The discretized hyperlocal model is as follows: the current estimated current value is related to the previous current value, the previous lumped disturbance, and the product of the input voltage and the input gain of the permanent magnet synchronous motor at the previous moment.

6. The model - free predictive control method for a permanent - magnet motor continuous set based on Bayesian optimization according to claim 1, characterized in that, Based on the expected value acquisition function, optimize the variable combination, including: Calculate the expected performance improvement of all candidate optimized variable combinations, where the performance improvement is the difference between the predicted value and the current optimal value; Select the candidate optimized variable combination with the largest expected improvement as the optimal solution for the next control cycle; If the expected improvement of all candidate optimized variable combinations is negative, keep the current optimized variable combination unchanged.

7. The model - free predictive control method for a permanent - magnet motor continuous set based on Bayesian optimization according to claim 1, characterized in that, The calculation method of the reference voltage is as follows: ； Among them, represents a constant parameter, is the lumped disturbance estimation value of the extended state observer, T is the time constant, represents the reference voltage value at the control period . represents the value of the input gain at the control period . represents the current reference value at the control period . represents the current reference value at the control period . represents the lumped disturbance estimation value at the control period . represents the current estimation value at the control period .

8. The model - free predictive control method for permanent - magnet motors with continuous sets based on Bayesian optimization according to claim 1, wherein, Decompose the reference voltage into basic voltage vectors through space vector modulation, and control the inverter to execute the switching sequence of the basic voltage vectors according to the minimum number of switches to drive the permanent magnet motor, including: Convert the reference voltage from the rotating coordinate system to the stationary coordinate system; Determine the target sector based on the voltage in the stationary coordinate system; Determine the basic voltage vectors based on the position of the target sector; Calculate the action time of the basic voltage vectors; Based on the action time, control the inverter to execute the switching sequence of the basic voltage vectors according to the arrangement order of the minimum number of switches to drive the permanent magnet motor.

9. A device for model - free predictive control of permanent - magnet motors with continuous sets based on Bayesian optimization, wherein, The control device includes: An acquisition module, configured to acquire a hyperlocal model and a pre-established extended state observer; wherein, the hyperlocal model is determined based on the mathematical model of the permanent magnet synchronous motor in the two-phase rotating coordinate system; An update module, configured to determine the bandwidth of the extended state observer and the input gain of the hyperlocal model as optimization variables, and acquire a training set and a test set for Bayesian optimization, where both the training set and the test set are obtained by sampling historical bandwidth and historical input gain through a sliding window; A prediction module, configured to calculate the joint probability distribution of the training set and the test set through a kernel function, and determine the predicted mean and predicted variance of the optimization variables; A screening module, which screens the optimized variable combination based on the expected value acquisition function, and updates the bandwidth of the extended state observer and the input gain of the hyperlocal model based on the screened optimized variable combination; the screening module also calculates the reference voltage applied to the next control cycle based on the updated extended state observer and the hyperlocal model, and using the deadbeat control principle; A control module, configured to decompose the reference voltage into basic voltage vectors through space vector modulation, and control the inverter to execute the switching sequence of the basic voltage vectors according to the minimum number of switches to drive the permanent magnet motor.

10. A computer - readable storage medium, on which a computer program is stored, wherein, When the computer program is executed by a processor, it implements the steps of the permanent magnet motor continuous set model-free predictive control method based on Bayesian optimization according to any one of claims 1 to 8.

Images