A model-free predictive control method and system for permanent magnet synchronous motor

Abstract

This invention provides a model-free predictive control method and system for permanent magnet synchronous motors (PMSMs). The method includes: constructing a discretized current prediction equation for the PMSM in a dq rotating coordinate system; determining an ideal voltage input containing unknown motor parameters based on the discretized current prediction equation and the state tracking target; constructing a neural network approximator to approximate the ideal voltage input online; constructing a weight matrix update rule for the neural network approximator, wherein the weight matrix update rule uses the state tracking error as a learning signal and does not depend on unknown motor parameters, but only on measurable data; updating the weight matrix of the neural network approximator based on the current measurable data in each control cycle, and determining the output voltage of the neural network approximator based on the weight matrix; and driving the PMSM to run based on the output voltage. This invention achieves control of a PMSM without requiring a precise motor model or offline training.

Description

Technical Field

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

[0002] Permanent magnet synchronous motors (PMSMs) possess advantages such as high efficiency, high power density, and high reliability, and are widely used in industrial manufacturing, new energy electric vehicles, and other fields. Due to their faster dynamic response and better steady-state performance, model predictive control (MPC) has become one of the most widely used methods in PMSM control. However, MPC methods are highly dependent on model parameters; when the controller parameters are inconsistent with the actual motor system parameters, control performance deteriorates. To address the parameter mismatch problem, researchers have proposed several methods. The first type of method is based on parameter identification, which updates the controller structure by identifying motor parameters in real time, thereby enhancing the system's parameter robustness. However, this type of method has high algorithm complexity and computational burden. The second type of method is based on disturbance observers. This method treats the adverse effects of parameter mismatch as disturbances and constructs observers (such as sliding mode observers and extended state observers) to observe disturbances and compensate them into the controller, thereby enhancing the system's parameter robustness. While this type of method reduces the amount of computation, it still uses the mathematical model of the motor, just like parameter identification.

[0003] In recent years, model-free predictive control (MRC) methods have received increasing attention due to their lack of reliance on system models, offering greater versatility and flexibility compared to other methods. Common model-free control algorithms include hyperlocal models, dynamically linearized models, and lookup table models. These methods drive controller updates through sampled data to control the motor, but they also have limitations. For example, lookup table models are highly dependent on the accuracy of motor sampling and suffer from stagnation effects; hyperlocal models still rely on inductor parameters for hyperlocal gain design; and dynamically linearized models have complex initial parameter tuning issues. With the development of artificial intelligence and digital signal processors, model-free MRC methods based on reinforcement learning, imitation learning, and deep learning have been proposed. While these learning-based methods eliminate the use of model parameters to improve the robustness of controller parameters, all three require large amounts of data for offline training, which significantly increases the difficulty of implementation and hinders deployment.

[0004] Therefore, there is an urgent need to provide a model-free predictive control method and system for permanent magnet synchronous motors, which can solve the technical problems encountered by current learning-based methods, such as the need for offline training with large amounts of data and the inability to achieve real-time control. Summary of the Invention

[0005] In view of this, it is necessary to provide a model-free predictive control method and system for permanent magnet synchronous motors to solve the technical problem that existing model-free predictive control methods require a large amount of data for offline training and cannot be implemented in real time.

[0006] To address the aforementioned technical problems, in a first aspect, the present invention provides a model-free predictive control method for a permanent magnet synchronous motor, comprising: Discretized current prediction equations for permanent magnet synchronous motors in the dq rotating coordinate system are constructed. Based on the discretized current prediction equation and the state tracking target, the ideal voltage input containing unknown motor parameters is determined. A neural network approximator is constructed for online approximation of the ideal voltage input. The input of the neural network approximator is a measurable motor state quantity and state tracking error, and the output is an actual voltage command that approximates the ideal voltage input. A weight matrix update rule for the neural network approximator is constructed. The weight matrix update rule uses the state tracking error as the learning signal and does not depend on the unknown motor parameters, but only on the measurable data. In each control cycle, the weight matrix of the neural network approximator is updated based on the currently measurable data, and the output voltage of the neural network approximator is determined based on the weight matrix; The permanent magnet synchronous motor is driven to operate based on the output voltage.

[0007] In one possible implementation, the weight matrix update rule is as follows:

[0008] In the formula, This is the weight matrix at time k+1; Let be the weight matrix at time k; For design parameters; This refers to the state tracking error; For the neuron function of the neural network approximator; This is the input to the neural network approximator.

[0009] In one possible implementation, the construction of the discretized current prediction equation for the permanent magnet synchronous motor in the dq rotating coordinate system includes: Real-time acquisition of three-phase stator current and rotor electric angular velocity of permanent magnet synchronous motor; Based on the three-phase stator current and the rotor electric angular velocity, the continuous voltage equation of the permanent magnet synchronous motor in the dq rotating coordinate system is constructed. The continuous voltage equation is discretized using the forward Euler method to obtain the discretized current prediction equation.

[0010] In one possible implementation, the state tracking error is the difference between the reference current and the actual current; then, determining the ideal voltage input containing unknown motor parameters based on the discretized current prediction equation and the state tracking target includes: A Lyapunov function is constructed based on the difference between the reference current and the actual current. Based on the difference of the Lyapunov function being negative definite or semi-negative definite, the decay law of the state tracking error is generated; Substituting the attenuation law into the discretized current prediction equation, the ideal voltage input containing unknown motor parameters is solved in reverse.

[0011] In one possible implementation, the network architecture of the neural network approximator is a radial basis function neural network.

[0012] In one possible implementation, the method further includes: The measurable data and the output voltage are input into a pre-built extended state observer to obtain the estimated state, estimated disturbance, and disturbance rate of the permanent magnet synchronous motor. Based on the estimated state, estimated perturbation, and perturbation change rate, a robust compensation voltage is generated to suppress the inherent approximation error of the neural network approximator and external perturbations. The sum of the output voltage and the robust compensation voltage is used as the target control voltage; The permanent magnet synchronous motor is driven to operate based on the target control voltage.

[0013] In one possible implementation, generating a robust compensation voltage based on the estimated state, estimated perturbation, and perturbation rate of change to suppress the inherent approximation error of the neural network approximator and external perturbations includes: The current tracking error and the rate of change of the current tracking error are determined based on the estimated state, and the filtering error is determined based on the current tracking error and the rate of change of the current tracking error. The compensation gain is determined based on the estimated disturbance and the rate of change of the disturbance, and the robust compensation voltage is determined based on the compensation gain and the filtering error.

[0014] In one possible implementation, the robust compensation voltage is:

[0015]

[0016]

[0017] In the formula, For robust compensation voltage; This is the filtering error; To compensate for the gain; The rate of change of current tracking error; For current tracking error; These are the filter coefficients; It is a symbolic function; This is a design constant.

[0018] In one possible implementation, driving the permanent magnet synchronous motor based on the output voltage includes: The output voltage is spatially vector modulated to obtain a pulse width modulation signal, and the pulse width modulation signal is inverted to obtain a current signal that drives the permanent magnet synchronous motor.

[0019] Secondly, the present invention also provides a model-free predictive control system for a permanent magnet synchronous motor, comprising: Discretized current prediction equation construction unit, used to construct the discretized current prediction equation of permanent magnet synchronous motor in dq rotating coordinate system; An ideal voltage input determination unit is used to determine the ideal voltage input containing unknown motor parameters based on the discretized current prediction equation and the state tracking target. A neural network approximator construction unit is used to construct a neural network approximator for online approximation of the ideal voltage input. The input of the neural network approximator is a measurable motor state quantity and state tracking error, and the output is an actual voltage command that approximates the ideal voltage input. The weight matrix update rule construction unit is used to construct the weight matrix update rule of the neural network approximator. The weight matrix update rule uses the state tracking error as the learning signal and does not depend on the unknown motor parameters, but only on the measurable data. An output voltage determination unit is used to update the weight matrix of the neural network approximator based on the currently measurable data in each control cycle, and to determine the output voltage of the neural network approximator based on the weight matrix. A motor control unit is used to drive the permanent magnet synchronous motor to operate based on the output voltage.

[0020] The beneficial effects of this invention are as follows: The model-free predictive control method for permanent magnet synchronous motors provided by this invention constructs a neural network approximator for online approximation of the ideal voltage input, thus eliminating the influence of changes or mismatches in motor parameters on the control of the permanent magnet synchronous motor. Compared with the problems of traditional model predictive control being highly sensitive to parameters and experiencing performance degradation when parameters are mismatched, this invention fundamentally improves robustness and stability under parameter uncertainty, ensuring control accuracy even when motor parameters are time-varying or initially inaccurate.

[0021] Meanwhile, the weight matrix update rule of the neural network approximator in this invention is set to use the state tracking error as the learning signal, relying solely on measurable data and not on unknown motor parameters. It requires no offline training phase, overcoming the drawback of existing advanced control methods based on deep learning and reinforcement learning that require massive amounts of data for long-term offline pre-training, significantly reducing the complexity and cost of implementation. Furthermore, the neural network approximator can self-adjust and learn in real time during motor operation, possessing online adaptive capabilities, making it more suitable for practical engineering applications.

[0022] In summary, this invention combines unsupervised learning mechanisms with predictive control frameworks, successfully achieving high-performance and robust control of permanent magnet synchronous motors without the need for precise motor models and offline training. This has significant theoretical value and broad engineering application prospects. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A schematic flowchart of an embodiment of the model-free predictive control method for permanent magnet synchronous motors provided by the present invention; Figure 2 For the present invention Figure 1 A schematic diagram of an embodiment of S101; Figure 3 For the present invention Figure 1 A schematic diagram of an embodiment of S102; Figure 4 A schematic flowchart of an embodiment of the present invention for determining robust interference terms; Figure 5 A schematic diagram of a control flow for the model-free predictive control method for permanent magnet synchronous motors provided by the present invention; Figure 6This is a schematic diagram of an embodiment of the model-free predictive control system for permanent magnet synchronous motors provided by the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0026] It should be understood that the illustrative drawings are not drawn to scale. The flowcharts used in this invention illustrate operations implemented according to some embodiments of the invention. It should be understood that the operations in the flowcharts may be implemented out of order, and steps without logical contextual relationships may be reversed or performed simultaneously. Furthermore, those skilled in the art, guided by the content of this invention, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor systems and / or microcontroller systems.

[0027] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0028] This invention provides a model-free predictive control method and system for permanent magnet synchronous motors, which will be described below.

[0029] Figure 1 This is a schematic flowchart of an embodiment of the model-free predictive control method for permanent magnet synchronous motors provided by the present invention, as shown below. Figure 1 As shown, the model-free predictive control method for permanent magnet synchronous motors includes: S101. Construct the discretized current prediction equation for the permanent magnet synchronous motor in the dq rotating coordinate system; S102. Based on the discretized current prediction equation and the state tracking target, determine the ideal voltage input containing unknown motor parameters.

[0030] Specifically, the expression for the ideal voltage input is:

[0031] G ( k ) = T s / L d In the formula, The ideal voltage input is at time k+1; For time k, there are unknowns related to the permanent magnet synchronous motor; Let k be the ideal state value of the permanent magnet synchronous motor at time k; L d Stator resistance; T s To control the cycle.

[0032] S103. Construct a neural network approximator for online approximation of the ideal voltage input. The input of the neural network approximator is the measurable motor state quantity and state tracking error, and the output is the actual voltage command that approximates the ideal voltage input. S104. Construct the weight matrix update rule for the neural network approximator. The weight matrix update rule uses the state tracking error as the learning signal and does not depend on unknown motor parameters, but only on measurable data.

[0033] Specifically, the process of constructing the weight matrix update rule is as follows: First, the neural network approximator satisfies:

[0034] In the formula, The ideal output of the neural network approximator; This is the ideal weight matrix for the neural network approximator; For neuron functions; This is the input to the neural network approximator at time k; Let be the inherent approximation error of the neural network approximator at time k.

[0035] The goal is to make the actual output of the neural network approximator With rational output Approximately, therefore It can be represented as:

[0036] In the formula, This is the actual weight matrix of the neural network approximator.

[0037] It can be seen that only by obtaining The update rule can then yield an approximate output, and the state tracking error is defined as:

[0038] In the formula, Let represent the tracking error of the weight matrix. To obtain the update rule for the weight matrix and minimize the error, a cost function is defined with respect to the output tracking error. The update rule for the weight matrix is ​​obtained using gradient descent:

[0039] As can be seen from the above expression, this update rule still depends on the ideal input and cannot achieve online updating of the weight matrix. Therefore, it is assumed that there exists Make:

[0040] The weight matrix update rule can then be rewritten as:

[0041] In the formula, This is the weight matrix at time k+1; Let be the weight matrix at time k; For design parameters; This refers to the state tracking error; For the neuron function of the neural network approximator; This is the input to the neural network approximator.

[0042] It should be noted that the tracking state error is the difference between the reference state and the real-time state acquired by the sensor. The reference state refers to the available parameters. Therefore, the weight matrix update rule is only related to the state parameters, which are all measurable data, with no unknown parameters. Thus, online updates of the weight matrix can be achieved without training.

[0043] It should be noted that: to prove the validity of the approximation hypothesis, the cost function can be... Find the derivative and determine if it is less than or equal to zero. If it is less than or equal to zero, it means that the convergence has occurred, that is, the weight matrix update rule obtained by using the gradient descent method is correct and reasonable.

[0044] S105. In each control cycle, update the weight matrix of the neural network approximator based on the current measurable data, and determine the output voltage of the neural network approximator based on the weight matrix. S106, Drive the permanent magnet synchronous motor based on the output voltage.

[0045] It should be understood that the model-free predictive control method for permanent magnet synchronous motors in this embodiment of the invention can be implemented in any device based on the model-free predictive control method for permanent magnet synchronous motors, such as a permanent magnet synchronous motor control device. Specifically, the model-free predictive control method for permanent magnet synchronous motors is stored in the aforementioned device as a pre-programmed program. When the device is started, the program is invoked, and the model-free predictive control method for permanent magnet synchronous motors is implemented.

[0046] Compared with existing technologies, the model-free predictive control method for permanent magnet synchronous motors provided in this invention constructs a neural network approximator for online approximation of the ideal voltage input, thus eliminating the influence of motor parameter changes or mismatches on the control of the permanent magnet synchronous motor. Compared to the problems of traditional model predictive control being highly sensitive to parameters and experiencing performance degradation due to parameter mismatches, this invention fundamentally improves robustness and stability under parameter uncertainty, ensuring control accuracy even when motor parameters are time-varying or initially inaccurate.

[0047] Meanwhile, in this embodiment of the invention, the weight matrix update rule of the neural network approximator is set to use the state tracking error as the learning signal, relying solely on measurable data and not on unknown motor parameters. It requires no offline training phase, overcoming the drawback of existing advanced control methods based on deep learning and reinforcement learning that require massive amounts of data for long-term offline pre-training, significantly reducing the complexity and cost of implementation. Furthermore, the neural network approximator can self-adjust and learn in real time during motor operation, possessing online adaptive capabilities, making it more suitable for practical engineering applications.

[0048] In summary, the embodiments of the present invention combine unsupervised learning mechanisms with predictive control frameworks, and successfully achieve high-performance and robust control of permanent magnet synchronous motors without the need for accurate motor models and offline training. This has significant theoretical value and broad engineering application prospects.

[0049] In some embodiments of the present invention, such as Figure 2 As shown, step S101 includes: S201: Real-time acquisition of three-phase stator current and rotor electric angular velocity of permanent magnet synchronous motor.

[0050] Specifically, the three-phase stator current and rotor electric angular velocity during the operation of the permanent magnet synchronous motor are collected based on the current sensor and electric angular velocity sensor installed on the permanent magnet synchronous motor.

[0051] It should be understood that the sampling period can be set according to actual needs. In a specific embodiment of the present invention, the sampling period can be set to 100 microseconds.

[0052] S202. Based on the three-phase stator current and rotor electric angular velocity, construct the continuous voltage equation of the permanent magnet synchronous motor in the dq rotating coordinate system.

[0053] Specifically, firstly, the three-phase stator current is converted into d-axis and q-axis currents in the dq rotating coordinate system based on the Clarke-Park transformation.

[0054] Then, using coordinate transformation theory and constant amplitude as constraints, the continuous voltage equation of the permanent magnet synchronous motor in the dq coordinate system is obtained:

[0055] In the formula, u d The voltage along the d-axis; u q This is the q-axis voltage; i d This refers to the d-axis current. i q This is the q-axis current; R s , L d , L q , ψ f These are the stator resistance, stator d-axis inductance, stator q-axis inductance, and rotor permanent magnet flux linkage, respectively. w e ω is the rotor's electric angular velocity.

[0056] S203. Discretize the continuous voltage equation based on the forward Euler method to obtain the discretized current prediction equation.

[0057] Among them, the discretized current prediction equation can deduce the current change trend at the next sampling time based on the current value, voltage command and motor speed change at the previous sampling time, and serve as the basis for subsequent control.

[0058] Specifically, the discretized current prediction equation is as follows:

[0059] In the formula, k and k+1 are time k and time k+1, respectively; T s To control the cycle.

[0060] It should be noted that the ideal voltage input includes the d-axis ideal voltage input and the q-axis ideal voltage input, meaning that the d-axis and q-axis are controlled separately.

[0061] In some embodiments of the present invention, the state tracking error is the difference between the reference current and the actual current; then, as... Figure 3As shown, step S102 includes: S301. Construct the Lyapunov function based on the difference between the reference current and the actual current.

[0062] Specifically, the Lyapunov function is:

[0063] In the formula, This is the difference between the reference current and the actual current.

[0064] S302. Based on the difference of the Lyapunov function being negative definite or semi-negative definite, generate the decay law of the state tracking error.

[0065] S303. Substitute the attenuation law into the discretized current prediction equation to solve for the ideal voltage input containing unknown motor parameters.

[0066] The embodiments of the present invention derive an ideal voltage input based on Lyapunov stability theory, providing a clear and stable learning target for the neural network approximator, and theoretically guaranteeing the stability of the closed-loop system and the exponential convergence characteristics of the error.

[0067] Since the goal of this invention is to achieve unsupervised, real-time tracking, in a preferred embodiment of this invention, the network architecture of the neural network approximator is a radial basis function neural network.

[0068] Radial basis function neural networks possess local response characteristics and a linear output structure. Their weight update rules are extremely simple, enabling forward computation and online learning to be completed within a very short control cycle. This choice significantly reduces computational complexity while ensuring powerful nonlinear approximation capabilities, thus achieving real-time online control of permanent magnet synchronous motors.

[0069] In practical engineering, the control performance of permanent magnet synchronous motors will also decrease when they are disturbed. At present, most of the research focuses on improving the robustness of parameters, but does not consider how to improve the control performance when the sampled data is disturbed.

[0070] To improve control robustness under disturbance, in some embodiments of the present invention, such as Figure 4 As shown, the model-free predictive control method for permanent magnet synchronous motors also includes: S401. Input the measurable data and output voltage into the pre-built extended state observer to obtain the estimated state, estimated disturbance, and disturbance rate of the permanent magnet synchronous motor.

[0071] Specifically, the extended state observer is:

[0072] In the formula,z 1 represents the estimated state; z 2 is for estimating disturbances; z 3 represents the rate of change of the disturbance; β 1. β 2. β 3 are all coefficients of the extended observer; x Measurable data; u This refers to the output voltage. and b For configurable values, specifically, It is 100. b It is 200.

[0073] By discretizing the above equations using the Euler method, we can obtain:

[0074] By properly setting the parameters of the extended observer, the estimated state, estimated disturbance, and rate of change of disturbance at any given time can be obtained.

[0075] S402. Generate a robust compensation voltage for suppressing the inherent approximation error of the neural network approximator and external disturbances based on the estimated state, estimated disturbance and disturbance change rate. S403, use the sum of the output voltage and the robust compensation voltage as the target control voltage.

[0076] That is, the target control voltage for:

[0077] In the formula, This refers to the output voltage. For robust compensation voltage.

[0078] S404, based on the target control voltage, drives the permanent magnet synchronous motor to operate.

[0079] By introducing a robust compensation term, this invention can effectively observe and compensate for disturbances such as the inherent approximation error of the neural network, unmodeled dynamics, and sudden changes in external load in real time. This significantly enhances the anti-interference capability and steady-state accuracy of the control process while ensuring a fast dynamic response, thus improving control robustness.

[0080] In a specific embodiment of the present invention, step S402 includes: The current tracking error and the rate of change of the current tracking error are determined based on the estimated state, and the filtering error is determined based on the current tracking error and the rate of change of the current tracking error. The compensation gain is determined based on the estimated disturbance and the rate of change of the disturbance, and the robust compensation voltage is determined based on the compensation gain and the filtering error.

[0081] Specifically, the robust compensation voltage is:

[0082]

[0083]

[0084] In the formula, For robust compensation voltage; This is the filtering error; To compensate for the gain; The rate of change of current tracking error; For current tracking error; These are the filter coefficients; It is a symbolic function; This is a design constant.

[0085] In a specific embodiment of the present invention, step S106 specifically includes: The output voltage is spatially vector modulated to obtain a pulse width modulation signal, and the pulse width modulation signal is inverted to obtain the current signal that drives the permanent magnet synchronous motor.

[0086] To illustrate the overall flow of the model-free predictive control method for permanent magnet synchronous motors proposed in this embodiment of the invention, the following description is provided: Figure 5 As shown, the current tracking error is determined based on the reference current output by the speed loop and the actual current collected by the sensor. Then, it is input into the radial basis neural network approximator to obtain the output voltage. The output voltage and the current tracking error are input into the extended state observer to obtain the robust compensation voltage. The sum of the output voltage and the robust compensation voltage is used as the target control voltage. The output voltage is subjected to space vector modulation to obtain the pulse width modulation signal. The pulse width modulation signal is then inverted to obtain the three-phase current signal driving the permanent magnet synchronous motor.

[0087] In summary, the model-free predictive control method for permanent magnet synchronous motors proposed in this invention solves for the ideal voltage input, establishes an ideal neural network to equivalently substitute for the ideal voltage input, and designs an actual neural network to approximate the ideal voltage input. A tracking error function is established between the two, and a gradient descent method is designed to obtain the weight matrix update rule. By designing the relationship between state error and tracking error, the weight matrix update rule is made only relevant to the sampled data, thus achieving online network updates. Simultaneously, to compensate for the inherent approximation error neglected in the above derivation process and to compensate for external disturbances, a robust compensation term based on an extended state observer is designed. These settings significantly improve the parameter robustness and anti-interference capability of the predictive control of permanent magnet synchronous motors, eliminating the drawbacks of traditional learning-based methods that are complex to design and cannot be implemented online. This provides a reference for promoting the real-time application of artificial intelligence technology in industry and has strong engineering significance.

[0088] On the other hand, embodiments of the present invention also provide a model-free predictive control system for permanent magnet synchronous motors, such as... Figure 6 As shown, the model-free predictive control system 600 for permanent magnet synchronous motors includes: Discretized current prediction equation building unit 601 is used to build the discretized current prediction equation of permanent magnet synchronous motor in dq rotating coordinate system; The ideal voltage input determination unit 602 is used to determine the ideal voltage input containing unknown motor parameters based on the discretized current prediction equation and the state tracking target. The neural network approximator building unit 603 is used to build a neural network approximator for online approximation of an ideal voltage input. The input of the neural network approximator is a measurable motor state quantity and state tracking error, and the output is the actual voltage command that approximates the ideal voltage input. The weight matrix update rule construction unit 604 is used to construct the weight matrix update rule of the neural network approximator. The weight matrix update rule uses the state tracking error as the learning signal and does not depend on unknown motor parameters, but only on measurable data. The output voltage determination unit 605 is used to update the weight matrix of the neural network approximator based on the currently measurable data in each control cycle, and determine the output voltage of the neural network approximator based on the weight matrix. The motor control unit 606 is used to drive the permanent magnet synchronous motor based on the output voltage.

[0089] The permanent magnet synchronous motor model-free predictive control system 600 provided in the above embodiments can realize the technical solutions described in the above embodiments of the permanent magnet synchronous motor model-free predictive control method. The specific implementation principles of each module or unit can be found in the corresponding content in the above embodiments of the permanent magnet synchronous motor model-free predictive control method, which will not be repeated here.

[0090] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0091] The above provides a detailed description of the model-free predictive control method and system for permanent magnet synchronous motors provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A model-free predictive control method for a permanent magnet synchronous motor, characterized in that, include: Discretized current prediction equations for permanent magnet synchronous motors in the dq rotating coordinate system are constructed. Based on the discretized current prediction equation and the state tracking target, the ideal voltage input containing unknown motor parameters is determined. A neural network approximator is constructed for online approximation of the ideal voltage input. The input of the neural network approximator is a measurable motor state quantity and state tracking error, and the output is an actual voltage command that approximates the ideal voltage input. A weight matrix update rule for the neural network approximator is constructed. The weight matrix update rule uses the state tracking error as the learning signal and does not depend on the unknown motor parameters, but only on the measurable data. In each control cycle, the weight matrix of the neural network approximator is updated based on the currently measurable data, and the output voltage of the neural network approximator is determined based on the weight matrix; The permanent magnet synchronous motor is driven to operate based on the output voltage.

2. The model-free predictive control method for permanent magnet synchronous motors according to claim 1, characterized in that, The weight matrix update rule is as follows: In the formula, This is the weight matrix at time k+1; Let be the weight matrix at time k; For design parameters; This refers to the state tracking error; For the neuron function of the neural network approximator; This is the input to the neural network approximator.

3. The model-free predictive control method for permanent magnet synchronous motors according to claim 1, characterized in that, The construction of the discretized current prediction equation for the permanent magnet synchronous motor in the dq rotating coordinate system includes: Real-time acquisition of three-phase stator current and rotor electric angular velocity of permanent magnet synchronous motor; Based on the three-phase stator current and the rotor electric angular velocity, the continuous voltage equation of the permanent magnet synchronous motor in the dq rotating coordinate system is constructed. The continuous voltage equation is discretized using the forward Euler method to obtain the discretized current prediction equation.

4. The model-free predictive control method for permanent magnet synchronous motors according to claim 1, characterized in that, The state tracking error is the difference between the reference current and the actual current; therefore, determining the ideal voltage input containing unknown motor parameters based on the discretized current prediction equation and the state tracking target includes: A Lyapunov function is constructed based on the difference between the reference current and the actual current. Based on the difference of the Lyapunov function being negative definite or semi-negative definite, the decay law of the state tracking error is generated; Substituting the attenuation law into the discretized current prediction equation, the ideal voltage input containing unknown motor parameters is solved in reverse.

5. The model-free predictive control method for permanent magnet synchronous motors according to claim 1, characterized in that, The neural network approximator uses a radial basis function neural network architecture.

6. The model-free predictive control method for permanent magnet synchronous motors according to claim 1, characterized in that, The method further includes: The measurable data and the output voltage are input into a pre-built extended state observer to obtain the estimated state, estimated disturbance, and disturbance rate of the permanent magnet synchronous motor. Based on the estimated state, estimated perturbation, and perturbation change rate, a robust compensation voltage is generated to suppress the inherent approximation error of the neural network approximator and external perturbations. The sum of the output voltage and the robust compensation voltage is used as the target control voltage; The permanent magnet synchronous motor is driven to operate based on the target control voltage.

7. The model-free predictive control method for permanent magnet synchronous motors according to claim 6, characterized in that, The generation of a robust compensation voltage for suppressing the inherent approximation error of the neural network approximator and external disturbances based on the estimated state, estimated disturbance, and rate of change of the disturbance includes: The current tracking error and the rate of change of the current tracking error are determined based on the estimated state, and the filtering error is determined based on the current tracking error and the rate of change of the current tracking error. The compensation gain is determined based on the estimated disturbance and the rate of change of the disturbance, and the robust compensation voltage is determined based on the compensation gain and the filtering error.

8. The model-free predictive control method for permanent magnet synchronous motors according to claim 7, characterized in that, The robust compensation voltage is: In the formula, For robust compensation voltage; This is the filtering error; To compensate for the gain; The rate of change of current tracking error; For current tracking error; These are the filter coefficients; It is a symbolic function; This is a design constant.

9. The model-free predictive control method for permanent magnet synchronous motors according to claim 1, characterized in that, The method of driving the permanent magnet synchronous motor based on the output voltage includes: The output voltage is spatially vector modulated to obtain a pulse width modulation signal, and the pulse width modulation signal is inverted to obtain a current signal that drives the permanent magnet synchronous motor.

10. A model-free predictive control system for a permanent magnet synchronous motor, characterized in that, include: Discretized current prediction equation construction unit, used to construct the discretized current prediction equation of permanent magnet synchronous motor in dq rotating coordinate system; An ideal voltage input determination unit is used to determine the ideal voltage input containing unknown motor parameters based on the discretized current prediction equation and the state tracking target. A neural network approximator construction unit is used to construct a neural network approximator for online approximation of the ideal voltage input. The input of the neural network approximator is a measurable motor state quantity and state tracking error, and the output is an actual voltage command that approximates the ideal voltage input. The weight matrix update rule construction unit is used to construct the weight matrix update rule of the neural network approximator. The weight matrix update rule uses the state tracking error as the learning signal and does not depend on the unknown motor parameters, but only on the measurable data. An output voltage determination unit is used to update the weight matrix of the neural network approximator based on the currently measurable data in each control cycle, and to determine the output voltage of the neural network approximator based on the weight matrix. A motor control unit is used to drive the permanent magnet synchronous motor to operate based on the output voltage.

Images