A model-free predictive control method and device for permanent magnet synchronous motors

Abstract

This application discloses a model-free predictive control method and apparatus for a permanent magnet synchronous motor, comprising: constructing a hyperlocal model; using an ESO observer to perform disturbance observation on the hyperlocal model to generate a model-free current predictive controller; acquiring current and voltage vectors for a complex number of historical control cycles; constructing a first cost function; calculating a first descent gradient of the first cost function with respect to the input gain value of the hyperlocal model and a second descent gradient with respect to the bandwidth value of the ESO observer; updating parameters based on the first and second descent gradients; calculating the predicted current value at time k+1 using the model-free current predictive controller after parameter updates; constructing a second cost function; and selecting the switching state that minimizes the second cost function to control the motor system based on the predicted current value at time k+1. This application can eliminate the dependence on motor parameters and improve the prediction accuracy and control robustness of the control system.

Description

Technical Field

[0001] This application relates to the field of motor control technology, and in particular to a model-free predictive control method and device for permanent magnet synchronous motors. Background Technology

[0002] With the rapid development of motor manufacturing technology and power electronics technology, various model-free predictive control methods suitable for permanent magnet synchronous motors have been proposed and applied. The control strategy of the control system plays a crucial role in the performance of the motor system. However, existing model-free predictive control methods are dependent on motor parameters and are easily affected by disturbances such as changes in operating conditions and external noise, leading to a decrease in the performance of the control system and making it difficult to meet the high requirements for motor control performance in practical applications. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this application adopts the following technical solution:

[0004] This application provides a model-free predictive control method for a permanent magnet synchronous motor, the predictive control method comprising the following steps:

[0005] Obtain the output current of the motor system at time k, and perform coordinate transformation on the output current to obtain the current value in the dq two-phase rotating coordinate system.

[0006] A hyperlocal model is constructed based on the current values ​​in the dq two-phase rotating coordinate system. An ESO observer is used to perform perturbation observation on the hyperlocal model to generate a model-free current prediction controller. The model-free current prediction controller is then discretized to obtain a discretized model-free current prediction controller.

[0007] Obtain the current vector and voltage vector of a complex number of historical control cycles, construct a first cost function to evaluate the current prediction error of the historical control cycles based on the data recorded in the finite set, and calculate the first descent gradient of the first cost function with respect to the input gain value of the hyperlocal model and the second descent gradient with respect to the bandwidth value of the ESO observer;

[0008] Based on the first and second descent gradients, the input gain value of the hyperlocal model in the discretized modelless current predictive controller and the bandwidth value of the ESO observer are updated. The current prediction value of the motor system at time k+1 under different switching state control is calculated by the modelless current predictive controller with updated parameters.

[0009] A second cost function is constructed to evaluate the current prediction error in the dq two-phase rotating coordinate system. Based on the current prediction value at time k+1, the switching state that minimizes the second cost function is selected to control the motor system.

[0010] In summary, the model-free predictive control method for permanent magnet synchronous motors provided in this application constructs a first cost function based on data from multiple historical control cycles. This first cost function is used to update the parameters of the model-free current predictive controller, dynamically adjusting the input gain of the hyperlocal model and the bandwidth of the ESO observer. This balances high-frequency noise suppression with low-frequency disturbance tracking capability, effectively suppressing the impact of system noise and other disturbances on prediction accuracy. The updated model-free current predictive controller then predicts the current of the motor system for future cycles. Based on the obtained current prediction values, a second cost function is used to effectively control the motor system. This achieves control of the permanent magnet synchronous motor based on historical cycle data, eliminating the control system's dependence on motor parameters and improving the prediction accuracy and control robustness of the control system.

[0011] Furthermore, the method also includes:

[0012] Calculate the partial derivative of the first cost function with respect to the input gain value of the hyperlocal model to obtain the first descent gradient; and calculate the partial derivative of the first cost function with respect to the bandwidth value of the ESO observer to obtain the second descent gradient;

[0013] Using the gradient descent method, the first descent gradient and the second descent gradient are weighted and calculated with the learning rate of the gradient descent method to obtain a first optimization value for the input gain of the hyperlocal model and a second optimization value for the bandwidth of the ESO observer.

[0014] Based on the input gain value of the hyperlocal model and the bandwidth value of the ESO observer at time k, and combined with the first optimization amount of the input gain value of the hyperlocal model and the second optimization amount of the bandwidth value of the ESO observer, the updated input gain value of the hyperlocal model and the bandwidth value of the ESO observer are calculated.

[0015] Furthermore, the method also includes:

[0016] Step size limiting is applied to the first optimization amount and the second optimization amount, and the step size limiting range of the first optimization amount and the second optimization amount is [- , ].

[0017] Furthermore, the method also includes:

[0018] The first descent gradient and the second descent gradient are obtained by calculating using the chain rule used in gradient calculation of artificial neural networks.

[0019] Furthermore, the first cost function is expressed by the following formula:

[0020] ;

[0021] In the formula, k represents time. This represents the d-axis current prediction error value during the historical control cycle. This represents the q-axis current prediction error value during the historical control cycle. This indicates the number of historical control cycles.

[0022] Furthermore, the second cost function is expressed by the following formula:

[0023] ;

[0024] In the formula, This represents the predicted d-axis current value at time k+2. This represents the predicted d-axis current value at time k+2. This indicates the reference value for the q-axis current. This indicates the reference value for the d-axis current.

[0025] Furthermore, the method also includes:

[0026] Multiple candidate switching states of the motor system are predicted by finite set prediction. The candidate switching states are then substituted into the model-free current prediction controller with updated parameters to calculate the current prediction value at time k+1 corresponding to each candidate switching state.

[0027] Furthermore, the method also includes:

[0028] Substitute the predicted current value at time k+1 corresponding to each of the candidate switch states into the discretized model-free current prediction controller, perform delay compensation on the predicted current value at time k+1, and obtain the predicted current value at time k+2.

[0029] Substitute the predicted current value at time k+2 into the second cost function, and select the switching state that minimizes the second cost function to control the motor system.

[0030] Furthermore, the discretized model-free current predictive controller is represented by the following formula:

[0031] ;

[0032] In the formula, This represents the predicted current value at time k. This represents the actual value of the current at time k. This represents the current prediction error value at time k. Indicates the switch status; , This represents the estimated value of the lumped disturbance. Indicates the sampling time of the control cycle; , This represents the input gain value of the hyperlocal model. This represents the input gain value of the hyperlocal model after normalization. , This represents the bandwidth value of the ESO observer. This represents the normalized bandwidth value of the ESO observer.

[0033] Secondly, this application also provides a model-free predictive control device for a permanent magnet synchronous motor, wherein the control device applies the above-mentioned predictive control method. Attached Figure Description

[0034] Figure 1 This is a flowchart of the steps of a model-free predictive control method for a permanent magnet synchronous motor provided in one embodiment of this application;

[0035] Figure 2 This is a flowchart illustrating the steps of online updating controller parameters in a model-free predictive control method for a permanent magnet synchronous motor provided in an embodiment of this application.

[0036] Figure 3 This is a flowchart illustrating the online updating of controller parameters in a model-free predictive control method for a permanent magnet synchronous motor provided in one embodiment of this application.

[0037] Figure 4 This is a schematic diagram of the chain rule for gradient calculation in a model-free predictive control method for a permanent magnet synchronous motor provided in one embodiment of this application;

[0038] Figure 5 This is a flowchart illustrating the steps of obtaining the switching vector to control the motor system in a model-free predictive control method for a permanent magnet synchronous motor provided in an embodiment of this application.

[0039] Figure 6 The control block diagram of the model-free predictive control method for permanent magnet synchronous motors provided in one embodiment of this application is applied to a motor system. Detailed Implementation

[0040] The present application will be described in detail below with reference to the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present application. Any structural, methodological, or functional modifications made by those skilled in the art based on these embodiments are included within the protection scope of the present application.

[0041] To address the shortcomings of existing technologies, embodiments of this application provide a model-free predictive control method for permanent magnet synchronous motors, such as... Figure 1As shown, the predictive control method includes the following steps:

[0042] Step S11: Obtain the output current of the motor system at time k, perform coordinate transformation on the output current, and obtain the current value in the dq two-phase rotating coordinate system.

[0043] Step S12: Construct a hyperlocal model based on the current values ​​in the dq two-phase rotating coordinate system, use an ESO observer to perform perturbation observation on the hyperlocal model, generate a model-free current predictive controller, and discretize the model-free current predictive controller to obtain a discretized model-free current predictive controller.

[0044] Step S13: Obtain the current vector and voltage vector of a complex number of historical control cycles, construct a first cost function to evaluate the current prediction error of the historical control cycles based on the data recorded in the finite set, and calculate the first descent gradient of the first cost function with respect to the input gain value of the hyperlocal model and the second descent gradient with respect to the bandwidth value of the ESO observer.

[0045] Step S14: Based on the first and second descent gradients, update the parameters of the input gain value of the hyperlocal model in the discretized modelless current predictive controller and the bandwidth value of the ESO observer. Calculate the predicted current value of the motor system at time k+1 under different switching state control using the updated modelless current predictive controller.

[0046] Step S15: Construct a second cost function to evaluate the current prediction error in the dq two-phase rotating coordinate system. Based on the current prediction value at time k+1, select the switching state that minimizes the second cost function to control the motor system.

[0047] Specifically, the output current of the motor system at time k is collected. After the data collection is completed, the Clark transformation is used to convert the three-phase current in the abc coordinate system into... αβ The current values ​​in the two-phase stationary coordinate system are then transformed using the Park transformation algorithm. αβ The two-phase stationary coordinate system is transformed into a dq two-phase rotating coordinate system, yielding the corresponding d-axis and q-axis current values. After the transformation, the originally alternating current signal becomes a relatively stable DC signal, which can intuitively reflect the operating state of the motor, thus facilitating the establishment of a control model.

[0048] A hyperlocal model is constructed based on the current values ​​in the dq two-phase rotating coordinate system obtained in step S11. This hyperlocal model integrates various parameter variations and external disturbances in the motor system into a single lumped disturbance term, describing the motor system characteristics through the relationship between input quantities (such as voltage vectors) and output quantities (such as current). The hyperlocal model does not rely on the specific electrical parameters of the motor, thus eliminating the dependence of traditional control methods on motor parameters and improving the robustness of the control strategy.

[0049] In one embodiment, the mathematical model of the permanent magnet synchronous motor in the dq two-phase rotating coordinate system can be expressed by the following formula:

[0050] (1);

[0051] In the formula, Represents the d-axis component of the stator output voltage of a permanent magnet synchronous motor. Represents the q-axis component of the stator output voltage of the permanent magnet synchronous motor; R represents the stator resistance of the permanent magnet synchronous motor. The d-axis component represents the stator current of a permanent magnet synchronous motor. The q-axis component represents the stator current of a permanent magnet synchronous motor; This indicates the electrical angular velocity of a permanent magnet synchronous motor; L represents the flux linkage of the permanent magnet; L represents the inductance of the permanent magnet synchronous motor.

[0052] Considering the relationship between the dq-axis switching state and the dq-axis components of the stator output voltage of the permanent magnet synchronous motor, we have:

[0053] (2);

[0054] In the formula, Indicates the DC bus voltage. The d-axis component represents the switch state. The q-axis component represents the switching state.

[0055] Substituting equation (2) into formula (1), we can obtain the current equation of the permanent magnet synchronous motor with respect to the switching state of the dq axis as follows:

[0056] (3);

[0057] In the formula, This indicates the DC bus voltage.

[0058] By integrating the external disturbances of the motor system into a lumped disturbance term, a hyperlocal model is obtained, which can be expressed by the following formula:

[0059] (4);

[0060] In the formula, This represents the input gain value of the hyperlocal model. It changes due to disturbances in the inductance parameters; The d-axis estimated component represents the lumped disturbance. This represents the q-axis estimate of the lumped disturbance.

[0061] An ESO (Electronic Stability Observer) is used to observe lumped disturbances in the hyperlocal model. ESO can estimate unknown disturbances and unmodeled dynamics in the motor system in real time, including motor parameter changes, load fluctuations, and external disturbances. Accurate observation of these disturbances allows for compensation during control, further enhancing the control system's anti-interference capability. Using the disturbance information observed by ESO and the hyperlocal model, a model-free current predictive controller is generated. This controller can predict future current trends in the motor system based on the current state and observed disturbances, achieving model-free motor control.

[0062] In one embodiment, the model-free current prediction controller can be represented by the following formula:

[0063] (5);

[0064] In the formula, This represents the current prediction error value, and ; This represents the predicted current value, and ; Let represent the estimated value of the lumped disturbance, and ; and The feedback gain of the ESO observer, and the values ​​along the d-axis and q-axis. and same; This represents the bandwidth value of the ESO observer.

[0065] Furthermore, the generated modelless current predictive controller is discretized using the first-order forward Euler discretization method, which transforms the continuous-time domain control model into a discrete-time domain model, resulting in a discretized modelless current predictive controller. The discretized modelless current predictive controller enables the prediction of the dq-axis current of the motor system at time k+1.

[0066] In step S13, current and voltage vectors from a plurality of historical control cycles are obtained. A finite set is used to store the current and voltage vector information from multiple past control cycles. The current vector includes the current value in the dq coordinate system, and the voltage vector includes the voltage value in the dq coordinate system. For further illustration, the following is an example of recording historical control cycles: Recording the... Current vector from control cycle to the kth control cycle and voltage vector ,in, With the first Using the predicted current and estimated lumped disturbance of the control cycle as initial values, the following exists:

[0067] (6);

[0068] In the formula, Indicates the number of historical control cycles. This represents the predicted value of the dq-axis current. This represents the estimated value of the lumped disturbance along the dq axis.

[0069] Optionally, the current vector and voltage vector of a complex number of historical control cycles can be obtained using a finite set approach.

[0070] Furthermore, based on recorded historical control cycle data, a first cost function is constructed by integrating current prediction error information from multiple historical control cycles. This first cost function is used to evaluate the sum of current prediction errors within the historical control cycles, reflecting the predictive performance of the control system under the current parameters. Based on this, a first descent gradient of the first cost function with respect to the input gain of the hyperlocal model and a second descent gradient with respect to the bandwidth of the ESO observer are calculated. The descent gradient reflects the rate of change of the first cost function with respect to parameter variations. The first descent gradient represents the degree of influence of changes in the hyperlocal model input gain on the sum of historical current prediction errors, while the second descent gradient represents the degree of influence of changes in the ESO bandwidth on the sum of historical current prediction errors. By calculating the first and second descent gradients, the direction of parameter adjustment can be clarified, thereby improving prediction accuracy.

[0071] Based on the first and second descent gradients obtained in step S13, the input gain value of the hyperlocal model and the bandwidth value of the ESO observer in the discretized model-free current predictive controller are updated. In this embodiment, the update process is implemented using the gradient descent algorithm to gradually reduce the current prediction error represented by the first cost function. After the parameter update is completed, the current prediction value of the motor system at time k+1 under different switching states is calculated using the updated model-free current predictive controller. For each candidate switching state, the dq-axis current value of the motor system at time k+1 under different switching states is predicted using the updated model-free current predictive controller, combined with the current state and the observed disturbance.

[0072] A second cost function is constructed to evaluate the current prediction error in the dq two-phase rotating coordinate system. This second cost function measures the error between the predicted current value and the reference current. A smaller value indicates a smaller deviation between the predicted and reference current values, resulting in better control performance. Based on the current prediction value at time k+1 calculated for each switching state in step S14, the second cost function value for each switching state is calculated. The switching state with the smallest second cost function value is selected as the control state for the motor system. Within each control cycle, the switching state is selected based on the latest current prediction value to ensure that the actual motor current always tracks the current reference value, achieving precise motor control.

[0073] Based on the above description, the model-free predictive control method for permanent magnet synchronous motors provided in this application constructs a first cost function based on data from multiple historical control cycles. This first cost function is used to update the parameters of the model-free current predictive controller, dynamically adjusting the input gain of the hyperlocal model and the bandwidth of the ESO observer. This balances high-frequency noise suppression with low-frequency disturbance tracking capability, effectively suppressing the impact of system noise and other disturbances on prediction accuracy. The updated model-free current predictive controller then predicts the current of the motor system for future cycles. Based on the obtained current prediction values, a second cost function is used to effectively control the motor system. This achieves control of the permanent magnet synchronous motor based on historical cycle data, eliminating the control system's dependence on motor parameters and improving the prediction accuracy and control robustness of the control system.

[0074] As an optional implementation, in step S12, the discretized model-free current prediction controller can be represented by the following formula:

[0075] (7);

[0076] In the formula, This represents the predicted current value at time k. This represents the actual value of the current at time k. This represents the current prediction error value at time k. Indicates the switch status; , This represents the estimated value of the lumped disturbance. Indicates the sampling time of the control cycle; , This represents the input gain value of the hyperlocal model. This represents the input gain value of the hyperlocal model after normalization. , This represents the bandwidth value of the ESO observer. This represents the normalized bandwidth value of the ESO observer.

[0077] As an optional implementation, in step S13, the first cost function can be expressed by the following formula:

[0078] (8);

[0079] In the formula, This represents the d-axis current prediction error value during the historical control cycle. This represents the q-axis current prediction error value during the historical control cycle.

[0080] Furthermore, as an optional implementation, the steps for online updating of controller parameters are as follows: Figure 2 As shown, based on the first cost function, the predictive control method also includes the following steps to update the input gain value of the hyperlocal model and the bandwidth value of the ESO observer:

[0081] Step S141: Calculate the partial derivative of the first cost function with respect to the input gain value of the hyperlocal model to obtain the first descent gradient; and calculate the partial derivative of the first cost function with respect to the bandwidth value of the ESO observer to obtain the second descent gradient.

[0082] Step S142: Using the gradient descent method, the first and second descent gradients are weighted and calculated with the learning rate of the gradient descent method to obtain the first optimized value of the input gain of the hyperlocal model and the second optimized value of the bandwidth of the ESO observer.

[0083] Step S143: Based on the input gain value of the hyperlocal model and the bandwidth value of the ESO observer at time k, and combining the first optimization amount of the input gain value of the hyperlocal model and the second optimization amount of the bandwidth value of the ESO observer, calculate the updated input gain value of the hyperlocal model and the bandwidth value of the ESO observer.

[0084] Specifically, a first cost function is constructed based on the current prediction errors from multiple historical control cycles. The value of the first cost function reflects the overall impact of the input gain of the hyperlocal model and the bandwidth of the ESO observer on the current prediction accuracy. The calculation results of the partial derivatives reflect the sensitivity and direction of influence of the two parameters on the current prediction error. When the partial derivative is positive, increasing the parameter increases the value of the first cost function, in which case the parameter needs to be decreased; when the partial derivative is negative, increasing the parameter decreases the value of the first cost function, in which case the parameter needs to be increased.

[0085] In one embodiment, the partial derivative of the first cost function with respect to the input gain value of the hyperlocal model, and the partial derivative of the first cost function with respect to the bandwidth value of the ESO observer, can be expressed by the following formula:

[0086] (9);

[0087] In the formula, This represents the input gain value of the hyperlocal model at time k. This represents the bandwidth value of the ESO observer at time k. This represents the partial derivative of the input gain value of the hyperlocal model. This represents the partial derivative of the ESO observer bandwidth value.

[0088] Taking the derivative of the right side of equation (9), we can obtain:

[0089] (10);

[0090] In the formula, and along with It becomes increasingly complex as it grows.

[0091] The first and second descent gradients are weighted by the learning rate of the gradient descent method to obtain the first and second optimization parameters, respectively. The first optimization parameter determines the specific adjustment range of the hyperlocal model input gain, and the second optimization parameter determines the specific adjustment range of the ESO observer bandwidth. By converting the descent gradients into specific parameter adjustment parameters using the gradient descent method, the direction and magnitude of the parameter updates are determined.

[0092] For the input gain value of the hyperlocal model, the input gain value at time k is combined with the first optimized quantity to obtain the updated hyperlocal model input gain value. The updated input gain value better matches the current input-output relationship of the motor system, reducing current prediction errors caused by gain mismatch. For the bandwidth value of the ESO observer, the bandwidth value at time k is combined with the second optimized quantity to obtain the updated ESO observer bandwidth value. The updated bandwidth value can balance the requirements of disturbance tracking and noise suppression, ensuring that the ESO observer can provide accurate disturbance estimation. Through dynamic updating and adjustment of parameters, the control system can continuously adapt to the dynamic changes of the motor system, solving the problem of insufficient robustness caused by fixed parameters in traditional control methods, and ensuring current prediction accuracy and system control performance.

[0093] In one embodiment, the input gain value of the hyperlocal model and the bandwidth value of the ESO observer can be updated using the following formula:

[0094] (11);

[0095] In the formula, Indicates the learning rate. This represents the first optimization quantity. This represents the second optimization quantity. This represents the input gain value of the updated hyperlocal model. This represents the bandwidth value of the updated ESO observer.

[0096] As an optional implementation, in step S142, a step size limit is applied to the first and second optimized quantities, and the step size limit range for the first and second optimized quantities is [- , The step size limit constrains the maximum adjustment range of the first and second optimization quantities by setting a range of [-ξ, ξ]. When the calculated optimization quantity exceeds this limit, the optimization quantity will be restricted to the boundary value ξ or -ξ, ensuring the stability and reliability of the parameter update process. This avoids system oscillations or instability caused by excessive optimization quantities, effectively buffering the impact of gradient anomalies, and making the parameter adjustment process smooth and gradual, avoiding abrupt parameter changes. Through the step size limit operation, while ensuring the adaptive optimization capability of parameters, it provides a safety guarantee for the entire control system, ensuring that the control system maintains a stable and reliable operating state.

[0097] To further illustrate the parameter updates for the input gain of the hyperlocal model and the bandwidth of the ESO observer, the calculation process for the parameter updates is as follows: Figure 3 As shown, the input gain value of the initial hyperlocal model is... and the bandwidth value of the ESO observer Calculate the partial derivatives of the first cost function with respect to the input gain of the hyperlocal model and the bandwidth of the ESO observer. By using the gradient descent method, the optimized values ​​of the input gain for the hyperlocal model and the bandwidth for the ESO observer are calculated. The optimization parameters are stepped and limited to update the input gain value of the hyperlocal model and the bandwidth value of the ESO observer.

[0098] As an optional implementation, in the predictive control method provided in this application embodiment, the first descent gradient and the second descent gradient can also be calculated using the chain rule used in gradient calculation of artificial neural networks. The chain rule decomposes the partial derivatives of a complex function into the product of the partial derivatives of multiple simple functions, simplifying the overall calculation process by passing gradient information layer by layer. Based on the chain rule calculation method, the original need to process data from multiple time points is decomposed into local calculations and recursive propagation at a single time point, reducing the computational complexity from high complexity related to the number of historical time points to low complexity linearly related to the number of historical time points, significantly reducing the computational complexity of the gradient, enabling rapid online updates of controller parameters, and meeting the real-time control requirements of the control system.

[0099] In one embodiment, by simplifying the calculation of formula (10) based on the chain rule, we can obtain:

[0100] (12);

[0101] In the formula, ,and and Since it is a known constant, and The derivative is 0.

[0102] To further illustrate, here is an example of a chain rule: Taking =4 as an example, the chain rule in gradient calculation is as follows: Figure 4 As shown, at this time m=k-3, the time passing through the (m-3)th time... Value and Value derivation for calculating the value at time m-2 The value, and through the (m-3)th time Value and Value derivation for calculating the value at time m-2 The values ​​are then used to obtain the first and second descent gradients.

[0103] As an optional implementation, in step S14, after updating the parameters of the modelless current predictive controller based on the first and second descent gradients, multiple candidate switching states of the motor system are predicted by a finite set. The candidate switching states are then substituted into the modelless current predictive controller after parameter updates, and the current prediction value at time k+1 corresponding to each candidate switching state is calculated respectively.

[0104] Specifically, in the permanent magnet synchronous motor control system, the motor is usually powered by a three-phase inverter. For the three-phase inverter, eight voltage vectors can be generated, including six non-zero voltage vectors and two zero vectors. Each voltage vector corresponds to a different switching state, and the eight switching states together constitute a set of candidate switching states.

[0105] After updating the parameters of the model-free current predictive controller based on the first and second descent gradients, multiple candidate switching states of the motor system are determined using a finite set. Each candidate switching state in the finite set is then substituted into the updated model-free current predictive controller to calculate the predicted current value at time k+1, corresponding to each candidate switching state. If the deviation between the predicted current value and the reference current value for a particular candidate switching state is small, it indicates that the candidate switching state is more conducive to achieving the control objective.

[0106] Furthermore, as an optional implementation method, such as Figure 5 As shown, after obtaining the predicted current value at time k+1 corresponding to each candidate switch state, step S15 further includes the following steps:

[0107] Step S151: Substitute the predicted current value at time k+1 corresponding to each candidate switch state into the discretized model-free current prediction controller, perform delay compensation on the predicted current value at time k+1, and obtain the predicted current value at time k+2 corresponding to each candidate switch state.

[0108] Step S151: Substitute the predicted current value at time k+2 corresponding to each candidate switch state into the second cost function, and select the candidate switch state that minimizes the second cost function to control the motor system.

[0109] Specifically, the predicted current value at time k+1 corresponding to each candidate switch state is substituted into the discretized modelless current predictive controller. The dynamic characteristics of the modelless current predictive controller are used to further predict the current value at time k+2. Through delay compensation, the predicted current value at time k+2 is closer to the predicted current value required by the actual control, so as to offset the impact of the calculation delay in the digital control system on the control accuracy and effectively improve the control accuracy of the control system.

[0110] In one embodiment, the predicted current value at time k+2 corresponding to each candidate switch state can be calculated using the following formula:

[0111] (13);

[0112] In the formula, This represents the predicted current value at time k+2.

[0113] After obtaining the predicted current values ​​at time (k+2) for each candidate switch state, these predicted current values ​​are successively substituted into the second cost function. For each candidate switch state, the second cost function calculates the deviation between the predicted current values ​​on the d-axis and q-axis and the reference current values. The smaller the value of the second cost function, the better the current tracking performance of that switch state, and the better it meets the operating requirements of the motor. The switch state that minimizes the value of the second cost function is selected from all candidate switch states and used as the control signal output to the motor system in the current control cycle, thereby completing the control of the motor system in the current control cycle.

[0114] As an optional implementation, the second cost function can be expressed by the following formula:

[0115] (14);

[0116] In the formula, This represents the predicted d-axis current value at time k+2. This represents the predicted d-axis current value at time k+2. This indicates the reference value for the q-axis current. This represents the d-axis current reference value; where the q-axis current reference value is calculated by a PI controller based on the actual and reference electrical angular velocities of the permanent magnet synchronous motor. d-axis current reference value It is configured to zero.

[0117] To further illustrate the model-free predictive control method for permanent magnet synchronous motors provided in this application, the predictive control method provided in this application is applied to motor systems such as... Figure 6As shown, based on the output current value of the motor system, a model-free current predictive controller is generated through a hyperlocal model and an ESO observer. A first cost function is constructed to evaluate the current prediction error of the historical control cycle. The first descent gradient of the first cost function with respect to the input gain value of the hyperlocal model and the second descent gradient with respect to the bandwidth value of the ESO observer are calculated. The parameters of the model-free current predictive controller are updated based on the first and second descent gradients. The current prediction value of the motor system at time k+1 under different switching states is calculated using the model-free current predictive controller with updated parameters. The current prediction value of the motor system at time k+1 under different switching states is delayed and compensated to obtain the current prediction value of the motor system at time k+2 under different switching states. The second cost function is used to calculate and obtain the switching state that minimizes the second cost function to control the motor system, so that the motor outputs the target voltage and current, thereby completing the control of the permanent magnet synchronous motor.

[0118] Based on the above description, the model-free predictive control method for permanent magnet synchronous motors provided in this application constructs a first cost function based on data from multiple historical control cycles. This first cost function is used to update the parameters of the model-free current predictive controller, dynamically adjusting the input gain of the hyperlocal model and the bandwidth of the ESO observer. This balances high-frequency noise suppression with low-frequency disturbance tracking capability, effectively suppressing the impact of system noise and other disturbances on prediction accuracy. The updated model-free current predictive controller then predicts the current of the motor system for future cycles. Based on the obtained current prediction values, a second cost function is used to effectively control the motor system. This achieves control of the permanent magnet synchronous motor based on historical cycle data, eliminating the control system's dependence on motor parameters and improving the prediction accuracy and control robustness of the control system.

[0119] Secondly, embodiments of this application also provide a model-free predictive control device for a permanent magnet synchronous motor. The predictive control device includes a control system and applies the predictive control method described above. The predictive control device can get rid of dependence on motor parameters, suppress the influence of disturbance factors such as system noise on prediction accuracy, and improve the prediction accuracy and control robustness of the control system.

[0120] It is understood that the term "exemplary" as used herein means "as an example, illustration, or description." Any embodiment described as "exemplary" is not necessarily preferred or superior to other embodiments and / or does not exclude features in combination with other embodiments. It should be understood that certain features of this application described in the context of a single embodiment for clarity may also be provided in combination in a single embodiment. Conversely, various features of this application described in the context of a single embodiment for clarity may also be provided individually or in any suitable combination or as part of any other described embodiment of this application.

[0121] In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0122] The above-disclosed embodiments are merely preferred embodiments of this application, but are not intended to limit the scope of this application. Those skilled in the art will understand that any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and scope of this application and the appended claims are equivalent substitutions and still fall within the scope of the invention.

Claims

1. A model-free predictive control method for a permanent magnet synchronous motor, characterized in that, The predictive control method includes the following steps: Obtain the output current of the motor system at time k, and perform coordinate transformation on the output current to obtain the current value in the dq two-phase rotating coordinate system. A hyperlocal model is constructed based on the current values ​​in the dq two-phase rotating coordinate system. An ESO observer is used to perform perturbation observation on the hyperlocal model to generate a model-free current prediction controller. The model-free current prediction controller is then discretized to obtain a discretized model-free current prediction controller. The current and voltage vectors of a complex number of historical control cycles are obtained using a finite set approach. A first cost function for evaluating the current prediction error of the historical control cycles is constructed based on the data recorded in the finite set. The first descent gradient of the first cost function with respect to the input gain value of the hyperlocal model and the second descent gradient with respect to the bandwidth value of the ESO observer are calculated. Based on the first and second descent gradients, the input gain value of the hyperlocal model in the discretized modelless current predictive controller and the bandwidth value of the ESO observer are updated. The modelless current predictive controller with updated parameters is used to calculate the (k+1)th current prediction value of the motor system under different switching state control. A second cost function is constructed to evaluate the current prediction error in the dq two-phase rotating coordinate system. Based on the current prediction value at time k+1, the switching state that minimizes the second cost function is selected to control the motor system.

2. The model-free predictive control method for permanent magnet synchronous motors according to claim 1, characterized in that, The method further includes: Calculate the partial derivative of the first cost function with respect to the input gain value of the hyperlocal model to obtain the first descent gradient; and calculate the partial derivative of the first cost function with respect to the bandwidth value of the ESO observer to obtain the second descent gradient; Using the gradient descent method, the first descent gradient and the second descent gradient are weighted and calculated with the learning rate of the gradient descent method to obtain a first optimization value for the input gain of the hyperlocal model and a second optimization value for the bandwidth of the ESO observer. Based on the input gain value of the hyperlocal model and the bandwidth value of the ESO observer at time k, and combined with the first optimization amount of the input gain value of the hyperlocal model and the second optimization amount of the bandwidth value of the ESO observer, the updated input gain value of the hyperlocal model and the bandwidth value of the ESO observer are calculated.

3. The model-free predictive control method for permanent magnet synchronous motors according to claim 2, characterized in that, The method further includes: Step size limiting is applied to the first optimization amount and the second optimization amount, and the step size limiting range of the first optimization amount and the second optimization amount is [- , ].

4. The model-free predictive control method for permanent magnet synchronous motors according to claim 1, characterized in that, The method further includes: The first descent gradient and the second descent gradient are obtained by calculating using the chain rule used in gradient calculation of artificial neural networks.

5. The model-free predictive control method for permanent magnet synchronous motors according to claim 1, characterized in that, The first cost function is expressed by the following formula: ； In the formula, k represents time. This represents the d-axis current prediction error value during the historical control cycle. This represents the q-axis current prediction error value during the historical control cycle. This indicates the number of historical control cycles.

6. The model-free predictive control method for permanent magnet synchronous motors according to claim 1, characterized in that, The second cost function is expressed by the following formula: ； In the formula, This represents the predicted d-axis current value at time k+2. This represents the predicted q-axis current value at time k+2. This indicates the reference value for the q-axis current. This indicates the reference value for the d-axis current.

7. The model-free predictive control method for permanent magnet synchronous motors according to claim 1, characterized in that, The method further includes: Multiple candidate switching states of the motor system are predicted by a finite set. The candidate switching states are then substituted into the model-free current prediction controller with updated parameters to calculate the current prediction value at time k+1 corresponding to each candidate switching state.

8. The model-free predictive control method for permanent magnet synchronous motors according to claim 7, characterized in that, The method further includes: Substitute the predicted current value at time k+1 corresponding to each of the candidate switch states into the discretized model-free current prediction controller, perform delay compensation on the predicted current value at time k+1, and obtain the predicted current value at time k+2. Substitute the predicted current value at time k+2 into the second cost function, and select the switching state that minimizes the second cost function to control the motor system.

9. The model-free predictive control method for permanent magnet synchronous motors according to claim 1, characterized in that, The discretized model-free current predictive controller is represented by the following formula: ； In the formula, This represents the predicted current value at time k. This represents the actual value of the current at time k. This represents the current prediction error value at time k. Indicates the switch status; , This represents the estimated value of the lumped disturbance. Indicates the sampling time of the control cycle; , This represents the input gain value of the hyperlocal model. This represents the input gain value of the hyperlocal model after normalization. , This represents the bandwidth value of the ESO observer. This represents the normalized bandwidth value of the ESO observer.

10. A model-free predictive control device for a permanent magnet synchronous motor, characterized in that, The control device employs the predictive control method according to any one of claims 1 to 9.

Images