Intelligent operation control method based on permanent magnet brushless motor

Abstract

The invention relates to the field of motor control, in particular to an intelligent operation control method based on a permanent magnet brushless motor, and the method comprises the steps: defining the input and output of a recursive least square method; calculating the cosine similarity between the current moment input vector and all historical moment input vectors, selecting the prediction error of the historical moment with the highest similarity to calculate the forgetting factor of the current moment, and updating the coefficient vector of the current moment based on the forgetting factor; and according to the coefficient vector, the working condition feature vector of the motor at the current moment and the constraint region of the control parameter vectors, calculating the predicted output efficiency at the current moment, and selecting a group of control parameter vectors with the maximum predicted output efficiency to control the operation of the motor at the next moment. According to the method, the forgetting factors are dynamically adjusted, so that the model can track the time-varying characteristics of the motor, can keep stability under similar working conditions, improves the adaptability of the model, can reversely solve the optimal control parameters, and controls the operation of the motor in real time.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of motor control. In particular, it relates to an intelligent control method for the operation of a permanent magnet brushless motor. BACKGROUND

[0002] Permanent magnet brushless motors have the advantages of high efficiency, high power density, fast response speed, etc., and are widely used in industrial drives, new energy vehicles, smart homes, etc. Motor operating efficiency is one of the core indicators to measure its performance, and the efficiency is jointly affected by the working conditions (speed, load, temperature, etc.) and control parameters (d, q-axis current).

[0003] Chinese patent CN114337427B discloses a rotational inertia identification method of recursive least squares with forgetting factor, which comprises: writing the motion equation of an asynchronous motor into a recursive least squares form, determining the output variable, the to-be-identified parameter and the observation matrix; defining the observation measure, the forgetting factor, the initialized covariance matrix and the identification parameter; calculating the gain matrix at the current time; calculating the covariance at the current time; updating the parameter estimation value; updating the target function value; comparing the calculated target function value with the preset target function value, constantly updating, and finally calculating the motor rotational inertia information.

[0004] Recursive least squares (RLS) can update model parameters in real time, but traditional RLS uses a fixed forgetting factor, which is not adaptive to time-varying working conditions: a too large forgetting factor will cause model update lag, and a too small forgetting factor will introduce noise interference, resulting in poor adaptability of the motor control mode and insufficient efficiency optimization. SUMMARY

[0005] To solve the above technical problems of poor adaptability of the motor control mode and insufficient efficiency optimization, the present application provides an intelligent control method for the operation of a permanent magnet brushless motor.

[0006] An intelligent control method for the operation of a permanent magnet brushless motor, comprising: The collected working condition feature vector of the motor at the current time and the control parameter vector output at the last time are taken as the input of the recursive least squares method, and the actual output efficiency of the motor at the current time is taken as the output of the recursive least squares method; The cosine similarity of the input vector at the current time and the input vector at all historical times is calculated, the historical time with the highest similarity is selected as the reference time, the error between the actual output efficiency and the predicted output efficiency at the reference time is obtained, the forgetting factor at the current time is calculated based on the error, and the coefficient vector at the current time is updated based on the forgetting factor; According to the updated coefficient vector, the current time operation characteristic vector of the motor and the constraint region of the set control parameter vector, the predicted output efficiency at the current time is calculated and a group of control parameter vectors with the maximum predicted output efficiency is selected to control the operation of the motor at the next time.

[0007] The application updates the coefficient vector of the model in real time through the recursive least square method, dynamically adjusts the forgetting factor by combining the cosine similarity and the historical error, makes the model not only track the time-varying characteristics of the motor, but also maintain stability under similar working conditions, improves the adaptability of the model, and solves the optimal control parameters by maximizing the output efficiency of the motor, which not only has small calculation amount, but also meets the real-time control demand.

[0008] Preferably, before recursion is performed by the recursive least square method, an initial coefficient vector and an initial covariance matrix of the recursive least square method are obtained based on the basic least square method, and an initial forgetting factor is set, and the process of obtaining the initial coefficient vector and the initial covariance matrix comprises: The historical operation data of the motor are collected to obtain the working condition characteristic vector, the control parameter vector and the actual output efficiency, the working condition characteristic vector and the control parameter vector obtained at each collection time are taken as input vectors, a historical data matrix is constructed by using the input vectors at all collection times, and a combination vector of the actual output efficiency at all collection times is taken as an output vector; the optimal coefficient and the covariance matrix are calculated based on the basic least square method, so that the initial coefficient vector and the initial covariance matrix are obtained.

[0009] The initial coefficient vector of the recursive least square method is obtained by the basic least square method, so that the recursion starting point is as close to the real parameter as possible, the initial prediction error caused by starting from a zero vector or a random value in the online stage is avoided, the convergence is slow, the uncertainty of the offline coefficient estimation is quantified by calculating the initial covariance matrix, the gain matrix of the recursive least square method is ensured to be in a reasonable order of magnitude for the first time, and the gain matrix is neither too aggressive nor too sluggish, and the initial forgetting factor is given as a historical data dependence benchmark when the recursive least square method is started.

[0010] Preferably, the calculation process of the forgetting factor at the current time comprises: An attenuation function with the error at the reference time as the index is constructed; the maximum value and the minimum value of the forgetting factor are set, the difference between the maximum value and the minimum value is multiplied by the attenuation function, the product obtained is summed with the minimum value, and the forgetting factor at the current time is obtained.

[0011] The difference between the maximum and minimum of the forgetting factor is used as the coefficient of the adjustment degree, avoiding the instability of the model caused by the too small forgetting factor; the forgetting factor is used to control the learning degree of the historical data, if the error is large, it means that the model deviation of the similar working condition of the current time is large, and more new data is needed, so the forgetting factor is reduced; if the error is small, it means that the model deviation of the similar working condition of the current time is small, and more historical data can be relied on, so the forgetting factor is increased; thereby improving the adaptability of the model.

[0012] Preferably, the updating process of the coefficient vector of the current time comprises: calculating the error between the actual output efficiency and the predicted output efficiency of the current time to obtain the efficiency error; calculating the gain matrix of the current time; calculating the sum of the product of the gain matrix and the efficiency error and the coefficient vector of the last time to obtain the updated coefficient vector.

[0013] Preferably, the calculation process of the efficiency error comprises: calculating the predicted output efficiency according to the input vector of the current time and the coefficient vector of the last time, and calculating the difference between the actual output efficiency and the predicted output efficiency of the current time to obtain the efficiency error.

[0014] Preferably, the calculation process of the gain matrix comprises: obtaining the covariance matrix of the last time, taking the product of the covariance matrix and the input vector of the current time as the numerator, and taking the sum of the updated forgetting factor and the second form of the product as the denominator to obtain the gain matrix.

[0015] Preferably, after updating the forgetting factor of the current time, the covariance matrix of the current time is updated, and the updating process comprises: calculating a correction term based on the covariance matrix of the last time, the input vector of the current time and the forgetting factor of the current time; multiplying the covariance matrix of the last time minus the correction term by the inverse of the forgetting factor of the current time to obtain the covariance matrix of the current time.

[0016] Preferably, the control parameter vector comprises the d-axis current and the q-axis current, and the setting process of the constraint region of the control parameter vector comprises: determining the range of the d-axis current and the q-axis current based on the rated parameters of the motor, and the combined current amplitude of the d-axis current and the q-axis current cannot exceed the set current upper limit, thereby obtaining the constraint region of the control parameter.

[0017] Preferably, the process of calculating the predicted output efficiency of the current time and selecting the control parameter vector with the maximum predicted output efficiency comprises: According to the updated coefficient vector of the current time, a linear formula of the output efficiency and the input vector is obtained; the working condition feature vector of the current time is a fixed value, and based on the constraint region of the control parameter, the optimal control parameter is solved through linear programming.

[0018] Preferably, the composition of the working condition feature vector includes the rotating speed of the motor, the load torque, the motor temperature and the DC bus voltage.

[0019] The present application has the following effects: 1. The present application updates the coefficient vector of the model in real time through the recursive least square method, and dynamically adjusts the forgetting factor by combining the cosine similarity and the historical error, so that the model can not only track the time-varying characteristics of the motor, but also maintain stability under similar working conditions, and improve the adaptability of the model. 2. The present application solves the optimal control parameter based on linear programming, quickly finds the maximum efficiency by traversing the vertices of the constraint region of the control parameter, has small calculation amount, maximizes the output efficiency of the motor, and meets the real-time control demand. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is a method flowchart of steps S1-S5 in an intelligent control method for operation of a permanent magnet brushless motor. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments.

[0022] The specific embodiments of the present application will be described in detail below with reference to the drawings.

[0023] The conventional recursive least square method usually presets a constant forgetting factor in the control of the permanent magnet brushless motor. When the preset forgetting factor is too large, the model relies too much on the historical data, cannot timely track the working condition drifts such as motor aging, temperature rise and load mutation, and the efficiency optimization is lagged. When the preset forgetting factor is too small, the latest data is excessively emphasized, the sensor noise and transient disturbance are amplified, the parameters are oscillated, and the motor operation is unstable. The present application considers that the constant forgetting factor makes the model unable to judge the similarity degree of the current working condition and the historical working condition, the adaptability is poor, and the efficiency optimization is insufficient. The present application proposes a scheme of adjusting the forgetting factor according to the historical prediction error, maintains the model to track fast in the time-varying working condition, suppresses the noise interference, and realizes the continuous optimization of the motor efficiency.

[0024] REFERENCE Figure 1 An intelligent control method for operation of a permanent magnet brushless motor includes steps S1-S5, and specifically as follows: S1, obtaining an initial coefficient vector and an initial covariance matrix of the recursive least square method based on the basic least square method, and setting an initial forgetting factor.

[0025] Before the recursive least square method proceeds with recursion, initialization parameters need to be set, including the initial coefficient vector, the initial covariance matrix and the initial forgetting factor, to provide a starting point that is close to the truth and robust enough for the recursive least square method, so that the subsequent recursion process can quickly converge and is not easy to diverge.

[0026] The historical operation data of the motor are collected by the motor test platform, and key information is extracted, including working condition data: speed, load torque, motor temperature, DC bus voltage; control parameters: d-axis current, q-axis current; input power and output power of the motor. For the historical operation data corresponding to each collection time, a working condition feature vector is constructed: [speed , load torque , motor temperature , DC bus voltage ]; a control parameter vector is constructed: [d-axis current , q-axis current ], to realize the control of the torque and speed of the motor; the actual output efficiency of the motor history is calculated: output power / input power .

[0027] The problem solved by the basic least square method is a linear regression problem, that is, to find a linear model to best fit the relationship between the input data and the output data, and the linear model of the basic least square method is constructed , where represents the output efficiency of the motor, represents the input vector of the basic least square method, represents the coefficient vector.

[0028] The input vector of the basic least square method is defined to include the working condition feature vector and the control parameter vector, and the input vector of the basic least square method is represented as: In the formula, represents the input vector of the basic least square method, is a constant term, represents the speed, represents the load torque, represents the motor temperature, represents the DC bus voltage, represents the d-axis current, represents the q-axis current.

[0029] The input vectors of all collection times are constructed into a historical data matrix , This represents the number of input vectors, and the historical data matrix is... The matrix.

[0030] The combined vector of the actual output efficiencies of the motor at all acquisition times is used as the output vector of the basic least squares method. The output vector of the basic least squares method is expressed as: By using the basic least squares method, the optimal coefficient estimate under the historical operating data is calculated by minimizing the cumulative error of the historical operating data, which is the initial coefficient vector of the recursive least squares method.

[0031] The objective function is to find a coefficient vector that minimizes the sum of squared errors between the model's predicted and actual output efficiencies. Therefore, the objective function can be expressed as: In the formula, Represents the objective function value. Indicates the first The actual output efficiency corresponding to each input vector. The model represents the first The efficiency of predicting the output of an input vector. This indicates the number of input vectors.

[0032] Based on the above objective function, the optimal coefficient estimate is obtained by solving for the optimal coefficient estimate (initial coefficient vector): In the formula, This indicates the optimal coefficient estimate of the basic least squares method based on historical operating data. Represents a historical data matrix. Represents the transpose of the historical data matrix. The scale used to adjust the coefficient estimates ensures the uniqueness and optimality of the solution. This represents a combined vector of actual output efficiencies at all acquisition times. This represents the product of the transposed input historical data matrix and the output vector, used to calculate the optimal coefficient estimate.

[0033] After obtaining the optimal coefficient estimate, the variance of the error is further calculated. The variance of the error is calculated according to the following formula: In the formula, The variance represents the error; Indicates the number of input vectors; This represents the number of elements (including constant terms) in the input vector. In this embodiment, ; actual output efficiency corresponding to the i th input vector, predicted output efficiency of the model under the optimal coefficient estimation actual output efficiency corresponding to the i th input vector, predicted output efficiency of the model under the optimal coefficient estimation actual output efficiency corresponding to the i th input vector.

[0034] The covariance matrix is calculated according to the variance of the error, that is, the initial covariance matrix of the recursive least squares method. The calculation of the covariance matrix (initial covariance matrix) satisfies the following formula: In the formula, denotes the covariance matrix (initial covariance matrix), denotes the historical data matrix, denotes the transpose matrix of the historical data matrix, denotes the variance of the error.

[0035] The initial coefficient vector of the recursive least squares method is obtained by the basic least squares method, so that the recursive starting point is as close to the true parameter as possible, and the initial prediction error is large and the convergence is slow caused by starting from a zero vector or random value in the online phase. By calculating the initial covariance matrix, the uncertainty of the offline coefficient estimation is quantified to ensure that the gain matrix of the recursive least squares method is in a reasonable order for the first time, neither too aggressive nor too sluggish.

[0036] In this embodiment, the initial forgetting factor is set to 0.95 as the historical data reference when the recursive least squares method is started.

[0037] S2, the working condition feature vector of the motor at the current time and the control parameter vector output at the last time are collected as the input of the recursive least squares method, and the actual output efficiency of the motor at the current time is collected as the output of the recursive least squares method.

[0038] Constructing a linear model of the recursive least squares method , wherein denotes the actual output efficiency of the motor, which is the output of the recursive least squares method; denotes the input vector of the recursive least squares method at the current time, including the working condition feature vector of the motor at the current time and the control parameter vector output at the last time, and can be specifically denoted as , wherein, denotes a constant term, denotes the speed of the motor at the current time, denotes the load torque of the motor at the current time, denotes the motor temperature at the current time, denotes the DC bus voltage at the current time, represents the d-axis current output at the last time, represents the q-axis current output at the last time of the motor; represents the coefficient vector to be estimated.

[0039] It should be noted that when the recursive least squares method is performed for the first time, because there is no recursive data at the last time, an initial control parameter vector needs to be artificially set, so that the recursive least squares method is started based on the initial control parameter vector, the initial coefficient vector in the above step S1, the initial covariance matrix and the initial forgetting factor.

[0040] S3, calculate the cosine similarity of the input vector at the current time and the input vectors at all historical times, select the historical time with the highest similarity as the reference time, obtain the error between the actual output efficiency and the predicted output efficiency at the reference time, and calculate the forgetting factor at the current time based on the error.

[0041] In the recursive process of the recursive least squares method, the cosine similarity of the input vector at the current time and the input vectors at all historical times in the previous recursive process is calculated, the cosine similarity measures the similarity of two input vectors in direction, and the value closer to 1 indicates that the similarity is higher. Select the historical time with the highest similarity as the reference time, and obtain the error between the actual output efficiency and the predicted output efficiency at the reference time , represents the actual output efficiency of the motor at the reference time, represents the input vector at the reference time, represents the coefficient vector updated at the last time of the reference time, represents the predicted output efficiency of the model to the input vector at the reference time.

[0042] The forgetting factor at the current time is adjusted based on the error at the reference time, and a decay function with the error at the reference time as the index is constructed; the maximum value and the minimum value of the forgetting factor are set, the difference between the maximum value and the minimum value is multiplied by the decay function, and the sum of the product obtained and the minimum value is obtained. The forgetting factor at the current time is obtained, which is specifically expressed by the following formula: In the formula, represents the forgetting factor at the current time; and are the maximum value and the minimum value of the forgetting factor set according to the demand of the permanent magnet brushless motor, the minimum value is generally determined according to the noise level of the system to prevent the noise from being amplified when the forgetting factor is too small, and the maximum value is generally set to 1. The difference between the maximum value and the minimum value of the forgetting factor is used as a coefficient of the adjustment degree to avoid instability of the model caused by the forgetting factor being too small; represents the error between the actual output efficiency and the predicted output efficiency at the reference time. This represents a decay function with respect to the error at the reference time as an exponent. The increase, The value will decrease.

[0043] The forgetting factor controls the degree of learning from historical data. A larger forgetting factor indicates that more historical data should be learned, while a smaller factor indicates that learning from new data is needed. By adjusting the forgetting factor as described above, if the error is large, it means that the current operating condition of the motor deviates significantly from the model of similar historical operating conditions, requiring more reliance on new data; therefore, the forgetting factor should be decreased. Conversely, if the error is small, it means that the current operating condition of the motor deviates less significantly from the model of similar historical operating conditions, allowing more reliance on historical data; therefore, the forgetting factor should be increased.

[0044] S4. Update the coefficient vector at the current time based on the forgetting factor at the current time.

[0045] After obtaining the forgetting factor at the current moment, the efficiency error at the current moment is calculated: based on the input vector at the current moment and the coefficient vector at the previous moment, the predicted output efficiency is calculated, and the difference between the actual output efficiency collected at the current moment and the predicted output efficiency is obtained, thus yielding the efficiency error. The specific calculation formula is as follows: In the formula, This represents the efficiency error at the current moment. This indicates the actual output efficiency of the motor at the current moment; This represents the transpose of the input vector at the current time step; This represents the coefficient vector from the previous time step. This indicates the model's prediction output efficiency for the input vector at the current time step.

[0046] Calculate the gain matrix at the current time step: Obtain the covariance matrix from the previous time step. Multiply the covariance matrix from the previous time step by the input vector at the current time step, using this as the numerator. Sum the quadratic form of this product with the updated forgetting factor, using this as the denominator, to obtain the gain matrix. The specific calculation formula is as follows: In the formula, This represents the gain matrix at the current time. Represents the covariance matrix of the previous time step; This represents the input vector at the current moment; This represents the forgetting factor at the current time step. The forgetting factor appears in the denominator and affects the size of the gain matrix. A larger forgetting factor results in a smaller gain matrix, and new data has a smaller impact on the coefficient vector update; a smaller forgetting factor results in a larger gain matrix, and new data has a larger impact on the coefficient vector update.

[0047] The coefficient vector of the current time is calculated: the product of the gain matrix and the efficiency error is summed with the coefficient vector of the previous time to obtain the updated coefficient vector. The specific calculation formula is as follows: In the formula, The coefficient vector of the current time is calculated: the product of the gain matrix and the efficiency error is summed with the coefficient vector of the previous time to obtain the updated coefficient vector. The specific calculation formula is as follows: The coefficient vector of the previous time is represented by The gain matrix of the current time is represented by The efficiency error of the current time is represented by The update amount of the coefficient vector is represented by The greater the value of each element in the gain matrix, the stronger the influence of the error of the current time on the correction of the coefficient vector, and the size of the gain matrix directly reflects the degree of influence of new data on the coefficient vector.

[0048] The covariance matrix of the current time is updated, and the updating process includes: calculating a correction term based on the covariance matrix of the previous time, the input vector of the current time, and the forgetting factor of the current time; subtracting the correction term from the covariance matrix of the previous time and multiplying the result by the reciprocal of the forgetting factor of the current time to obtain the covariance matrix of the current time. The covariance matrix of the current time is updated for the recursive calculation of the next time, and the calculation formula of the covariance matrix of the current time is as follows: In the formula, The covariance matrix of the current time is represented by The covariance matrix of the previous time is represented by The forgetting factor of the current time is represented by The input vector of the current time is represented by The transpose of the input vector of the current time is represented by

[0049] The forgetting factor appears in the denominator and affects the size of the change in the covariance matrix. The greater the forgetting factor, the smaller the change in the covariance matrix, indicating that the influence of new data on the covariance matrix is smaller. Since the forgetting factor is larger, the covariance matrix changes less, so when calculating the gain matrix of the next time, the gain matrix of the next time will be smaller, and the update amount of the coefficient vector of the next time will be smaller, indicating that the model has become relatively stable, or the new data is consistent with the historical data and does not need to be adjusted significantly. The smaller the forgetting factor, the greater the change in the covariance matrix, indicating that the influence of new data on the covariance matrix is greater. Since the forgetting factor is smaller, the covariance matrix changes more, so when calculating the gain matrix of the next time, the gain matrix of the next time will be larger, and the update amount of the coefficient vector of the next time will be larger, indicating that there is a significant difference between the new data and the historical data, or the model needs to quickly adapt to the new situation.

[0050] Therefore, by adjusting the forgetting factor, when the working condition of the motor changes or the load is disturbed, the forgetting factor will be adaptively reduced, making the recursive least squares model more "sensitive", and the coefficient vector can quickly track the new characteristics to avoid model lag; when the working condition of the motor is stable and the error at the reference time is small, the forgetting factor is automatically increased, the model is more "reliable" to historical data, and the measurement noise and sensor drift are suppressed to prevent the coefficient vector from oscillating. The forgetting factor is adjusted in real time without the need for manual repeated parameter adjustment or additional filters, and the entire closed-loop system is self-maintained to achieve self-tuning; moreover, the accurate forgetting factor enables the model to always maintain the minimum prediction error, thereby obtaining the truly optimal control parameter in the subsequent linear programming stage and improving the efficiency of the motor operation.

[0051] S5, according to the updated coefficient vector, the working condition characteristic vector of the motor at the current time, and the constraint region of the set control parameter vector, the predicted output efficiency at the current time is calculated and the control parameter vector with the maximum predicted output efficiency is selected to control the operation of the motor at the next time.

[0052] The control parameter vector includes the d-axis current and the q-axis current, the range of the d-axis current and the q-axis current is determined based on the rated parameters of the motor, and the combined current amplitude of the d-axis current and the q-axis current cannot exceed the set current upper limit, to determine the boundary constraint of the current: ; ; In the formula, Id represents the d-axis current; Idmin represents the minimum value of the d-axis current set based on the rated parameters of the motor; Idmax represents the maximum value of the d-axis current set based on the rated parameters of the motor; Iq represents the q-axis current; Iqmin represents the minimum value of the q-axis current set based on the rated parameters of the motor; Iqmax represents the maximum value of the q-axis current set based on the rated parameters of the motor; Ilim represents the maximum allowable value of the combined current amplitude, which is usually equal to the rated current of the motor or slightly higher than the rated current.

[0053] The constraint region of the current is a polygon region in a two-dimensional space, which is defined by the above boundary constraints. Specifically, the constraint region is defined by the following inequalities: After obtaining the coefficient vector at the current time, the linear formula of the output efficiency and the input vector is obtained: In the formula, P represents the output efficiency of the motor, are the element values in the coefficient vector at the current time.

[0054] The working condition feature vector of the motor at the current time is determined, and in order to maximize the output efficiency of the motor, the optimal control parameter is needed to be solved by linear programming under the given working condition feature vector The effective intersection points of the linear formula in the constraint region are found, including all boundary vertices of the constraint region and the intersection points of the straight line and the region, for each found intersection point (x, y) ), the corresponding predicted output efficiency is calculated by substituting the linear formula, and the point with the maximum predicted output efficiency is selected as the optimal control parameter vector output at the current time , to control the operation of the motor at the next time, so as to maximize the efficiency of the permanent magnet brushless motor.

[0055] The above steps S2-S5 are repeated at the next time to form a real-time control closed loop of the permanent magnet brushless motor, and the control parameter is always adapted to the current working condition.

[0056] The application solves the optimal control parameter based on linear programming, quickly finds the maximum efficiency by traversing the constraint region vertices of the control parameter, not only has small calculation amount, maximizes the output efficiency of the motor, but also meets the real-time control demand.

[0057] It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which all belong to the protection scope of the present application. Therefore, the protection scope of the present application patent should be subject to the appended claims.

Claims

1. A method for intelligent control of operation based on a permanent magnet brushless motor, characterized in that, include: The current operating condition feature vector of the motor and the control parameter vector output at the previous moment are used as inputs to the recursive least squares method, and the actual output efficiency of the motor at the current moment is used as the output of the recursive least squares method. Calculate the cosine similarity between the input vector at the current time and the input vectors at all historical times, select the historical time with the highest similarity as the reference time, obtain the error between the actual output efficiency and the predicted output efficiency at the reference time, calculate the forgetting factor at the current time based on the error, and update the coefficient vector at the current time based on the forgetting factor. Based on the updated coefficient vector, the motor's current operating condition characteristic vector, and the constraint region of the set control parameter vector, the predicted output efficiency at the current moment is calculated, and the set of control parameter vectors with the highest predicted output efficiency is selected to control the motor's operation at the next moment.

2. The intelligent control method for operation based on a permanent magnet brushless motor according to claim 1, characterized in that, Before performing the recursive least squares method, the process includes obtaining the initial coefficient vector and initial covariance matrix of the recursive least squares method based on the basic least squares method, and setting an initial forgetting factor. The process of obtaining the initial coefficient vector and initial covariance matrix includes: Historical operating data of the motor is collected to obtain operating condition feature vectors, control parameter vectors, and actual output efficiency. The operating condition feature vectors and control parameter vectors obtained at each acquisition moment are used as input vectors. A historical data matrix is ​​constructed using the input vectors at all acquisition moments, and the combined vector of actual output efficiency at all acquisition moments is used as the output vector. The optimal coefficients and covariance matrix are calculated based on the basic least squares method to obtain the initial coefficient vector and the initial covariance matrix.

3. The intelligent control method for operation based on a permanent magnet brushless motor according to claim 1, characterized in that, The calculation process for the forgetting factor at the current moment includes: Construct a decay function with the error at the reference time as the exponent; set the maximum and minimum values ​​of the forgetting factor, multiply the difference between the maximum and minimum values ​​by the decay function, and sum the resulting product with the minimum value to obtain the forgetting factor at the current time.

4. The intelligent control method for operation based on a permanent magnet brushless motor according to claim 2, characterized in that, The update process of the coefficient vector at the current moment includes: Calculate the error between the actual output efficiency and the predicted output efficiency at the current moment to obtain the efficiency error; calculate the gain matrix at the current moment; sum the product of the gain matrix and the efficiency error with the coefficient vector at the previous moment to obtain the updated coefficient vector.

5. The intelligent control method for operation based on a permanent magnet brushless motor according to claim 4, characterized in that, The calculation process for the efficiency error includes: The predicted output efficiency is calculated based on the input vector at the current moment and the coefficient vector at the previous moment. The difference between the actual output efficiency at the current moment and the predicted output efficiency is then calculated to obtain the efficiency error.

6. The intelligent control method for operation based on a permanent magnet brushless motor according to claim 4, characterized in that, The calculation process of the gain matrix includes: Obtain the covariance matrix of the previous time step, multiply the covariance matrix by the input vector of the current time step as the numerator, and use the sum of the updated forgetting factor and the quadratic form of the product as the denominator to obtain the gain matrix.

7. The intelligent control method for operation based on a permanent magnet brushless motor according to claim 1, characterized in that, After updating the forgetting factor at the current time step, update the covariance matrix at the current time step. The update process includes: The correction term is calculated based on the covariance matrix of the previous time step, the input vector of the current time step, and the forgetting factor of the current time step. The correction term is then subtracted from the covariance matrix of the previous time step and multiplied by the reciprocal of the forgetting factor of the current time step to obtain the covariance matrix of the current time step.

8. The intelligent control method for operation based on a permanent magnet brushless motor according to claim 7, characterized in that, The control parameter vector includes d-axis current and q-axis current, and the process of setting the constraint region of the control parameter vector includes: The range of the d-axis current and q-axis current is determined based on the rated parameters of the motor. At the same time, the combined current amplitude of the d-axis current and q-axis current cannot exceed the set current upper limit, thereby obtaining the constraint range of the control parameters.

9. The intelligent control method for operation based on a permanent magnet brushless motor according to claim 1, characterized in that, The process of calculating the prediction output efficiency at the current moment and selecting the control parameter vector that maximizes the prediction output efficiency is as follows: Based on the updated coefficient vector at the current moment, a linear formula for the output efficiency and the input vector is obtained; the operating condition feature vector at the current moment is a fixed value, and the optimal control parameters are solved by linear programming based on the constraint region of the control parameters.

10. The intelligent control method for operation based on a permanent magnet brushless motor according to claim 1, characterized in that, The operating condition feature vector consists of motor speed, load torque, motor temperature, and DC bus voltage.

Images