Permanent magnet synchronous motor adaptive predictive control method and system based on neural network

Abstract

The application provides a neural network-based adaptive predictive control method and system for a permanent magnet synchronous motor, relates to the field of power electronics, measures stator voltage and current at a current time, calculates stator current at a time to be predicted, calculates stator voltage corresponding to each switching state, carries out time delay compensation on the stator current at the time to be predicted according to the stator voltage corresponding to each switching state, and selects the optimal switching state corresponding to the minimum cost function value according to the compensated stator current and a preset reference current and applies the optimal switching state to a motor drive controller.The application reduces the parameter dependency of a traditional model predictive controller, does not need to add an additional hardware circuit, has a significant inhibitory effect on the poor system control performance and instability caused by model parameter changes, and greatly improves the robustness and reliability of a power transmission system with the permanent magnet synchronous motor drive as a core equipment.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of power electronics, and particularly relates to a permanent magnet synchronous motor adaptive predictive control method and system based on a neural network. BACKGROUND

[0002] The statements in this section merely provide background information related to the application and do not necessarily constitute prior art.

[0003] Permanent magnet synchronous motor model predictive control (MPC) is widely concerned in the academic field due to its rapid dynamic response, almost no need for adjustment parameters and other advantages. However, similar to other methods that depend on model parameters, permanent magnet synchronous motor model predictive control is very sensitive to changes in circuit model parameters (inductance, resistance, flux linkage, number of pole pairs, etc.), which is often manifested as difficulty in tracking the amplitude or phase of the reference signal, resulting in a decline in system stability and causing significant economic losses and personal safety threats to industrial production and life. Therefore, it is of great scientific and engineering significance to study the parameter desensitization predictive control strategy of the permanent magnet synchronous motor for efficient and reliable operation of energy storage, new energy electric vehicles and other systems.

[0004] Existing improvement methods are generally divided into observer-based methods and artificial intelligence algorithm-based methods. The observer-based method realizes adaptive adjustment of controller parameters to disturbances. However, the use of the observer method increases the number of adjustment parameters of the controller, increases the design difficulty of the system, and has limited adaptive ability to external disturbances. Although the adjustable range of the artificial intelligence method has been increased to a certain extent, there are still problems such as large amount of calculation and complex controller design, resulting in a large application cost.

[0005] Therefore, when the system circuit parameters change, the traditional model predictive control method significantly reduces the prediction performance, which on the one hand significantly reduces the control performance of the controller, making it difficult to accurately track the reference signal and reducing the service life of the motor drive system; on the other hand, it accelerates the aging speed of the capacitor, greatly shortens the service life of the capacitor, and increases the probability of secondary failure; and the existing improvement method has problems such as complex controller design, large amount of calculation, or limited adjustable range, which is difficult to be applied in industry. SUMMARY

[0006] To overcome the shortcomings of the prior art, the application provides a permanent magnet synchronous motor adaptive predictive control method and system based on a neural network, which reduces the dependence of the traditional model predictive controller parameters, does not require additional hardware circuits, and has a significant inhibitory effect on the poor system control performance and instability caused by changes in model parameters, greatly improving the robustness and reliability of the electric drive system with permanent magnet synchronous motor drive as the core equipment.

[0007] To achieve the above object, one or more embodiments of the present application provide the following technical solutions:

[0008] The first aspect of the present application provides a neural network-based adaptive predictive control method for a permanent magnet synchronous motor.

[0009] The neural network-based adaptive predictive control method for a permanent magnet synchronous motor comprises:

[0010] Based on the measured current-time stator voltage and current, the stator current at the predicted time is calculated using the reconstructed stator current prediction equation.

[0011] Traverse a preset number of switching states, and calculate the stator voltage corresponding to each switching state using the working principle of the converter.

[0012] According to the stator voltage corresponding to each switching state, the stator current at the predicted time is compensated for delay.

[0013] According to the compensated stator current and the preset reference current, based on the designed cost function, the optimal switching state corresponding to the minimum cost function value is applied to the motor drive controller.

[0014] The reconstructed stator current prediction equation is obtained by replacing the coefficients related to the circuit parameters in the initial stator current prediction equation with the weights of the neural network according to the PMSM model.

[0015] Further, the PMSM model is constructed based on the stator voltage equation and the stator flux equation, and finally is:

[0016]

[0017] Where, u d , u q are the d-q axis components of the stator voltage; i d , i q are the d-q axis components of the stator current; R s is the stator resistance; ψ d , ψ q are the d-q axis components of the stator flux; ω e is the electrical angular velocity; L d , L q are the d-q axis inductance components; ψ m is the permanent magnet flux.

[0018] Further, the stator voltage equation is:

[0019]

[0020] The stator flux linkage equation is:

[0021]

[0022] wherein u d , u q are d-q axis components of the stator voltage; i d , i q are d-q axis components of the stator current; R s is the stator resistance; ψ d , ψ q are d-q axis components of the stator flux linkage; ω e is the electrical angular velocity; L d , L q are d-q axis inductance components; and ψ m is the permanent magnet flux linkage.

[0023] Further, the reconstructed stator current prediction equation is expressed by the formula:

[0024]

[0025] wherein i is the d-q axis component of the current value at the k+1 moment, i is the d-q axis component of the current value at the k moment, ω x (x=1, 2, …, 7) is the weight actual value.

[0026] Further, the weight of the neural network is obtained by training the neural network, and the training of the neural network is to correct the weight actual value of the linear adaptive neural network by using the error between the actual output vector and the expected output vector, specifically:

[0027]

[0028]

[0029]

[0030] wherein F is a quadratic form error function, used to calculate the error between the actual output vector and the expected output vector; is the unit value of the weight correction value at the k moment; is the unit value of the weight correction value at the k-1 moment; is the unit value of the weight at the k+1 moment; and β is a first momentum term coefficient.

[0031] Further, the delay compensation is to introduce one beat of delay compensation on the premise of the stator current to be predicted, specifically:

[0032] The stator voltage corresponding to each switching state and the stator current at the to-be-predicted moment are input into the reconstructed stator current prediction equation to calculate and return the stator current at the next moment of the to-be-predicted moment.

[0033] Further, the cost function, in particular:

[0034]

[0035] Wherein, λ lim is a weight coefficient, i lim is a maximum current limit, and is a reference current, is the current value dq-axis component at the k+2 moment, i.e., the compensated stator current.

[0036] The second aspect of the application provides a neural network-based adaptive predictive control system of a permanent magnet synchronous motor.

[0037] The neural network-based adaptive predictive control system of the permanent magnet synchronous motor comprises a current prediction module, a voltage calculation module, a delay compensation module and an optimal selection module.

[0038] The current prediction module is configured to calculate the stator current at the to-be-predicted moment based on the measured stator voltage and current at the current moment by using the reconstructed stator current prediction equation.

[0039] The voltage calculation module is configured to calculate the stator voltage corresponding to each switching state by using the working principle of the converter by traversing a preset number of switching states.

[0040] The delay compensation module is configured to perform delay compensation on the stator current at the to-be-predicted moment according to the stator voltage corresponding to each switching state.

[0041] The optimal selection module is configured to select the optimal switching state corresponding to the minimum cost function value based on the designed cost function according to the compensated stator current and the preset reference current, and apply the optimal switching state to the motor drive controller.

[0042] The reconstructed stator current prediction equation is obtained by using the weight of the neural network to replace the coefficients related to the circuit parameters in the initial stator current prediction equation according to the PMSM model.

[0043] The third aspect of the application provides a computer readable storage medium having a program stored thereon, the program being executed by a processor to implement the steps in the neural network-based adaptive predictive control method of the permanent magnet synchronous motor according to the first aspect of the application.

[0044] The fourth aspect of the present application provides an electronic device, comprising a memory, a processor and a program stored in the memory and executable on the processor, wherein the processor implements the steps in the neural network-based adaptive predictive control method for permanent magnet synchronous motor according to the first aspect of the present application when executing the program.

[0045] The one or more technical solutions above have the following beneficial effects:

[0046] The present application proposes a neural network-based adaptive predictive control method for permanent magnet synchronous motor in the systems of micro-grid, energy storage, new energy electric vehicle, etc. The method mainly improves the parameter sensitivity of the classical model predictive control, replaces the coefficients related to the circuit parameters in the model predictive control with the weight coefficients in the neural network. Meanwhile, in order to solve the problem of multiple learning rates, the weight coefficients are normalized, i.e. only one learning rate parameter needs to be adjusted. The adaptive predictive control of the permanent magnet synchronous motor drive under the conditions of parameter mismatch and external disturbance is realized.

[0047] The present application reduces the parameter dependence of the traditional model predictive controller, and does not need to add extra hardware circuit. The present application has a significant inhibitory effect on the poor system control performance and instability caused by the change of model parameters, and greatly improves the robustness and reliability of the electric power transmission system with permanent magnet synchronous motor drive as the core equipment.

[0048] The advantages of the additional aspects of the present application will be partially given in the following description, partially become obvious from the following description, or be known by the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0049] The accompanying drawings, which form a part of the present application, are used to provide further understanding of the present application, and the illustrative embodiments of the present application and their description are used to explain the present application, and do not constitute improper limitations on the present application.

[0050] Figure 1 The method flowchart of the first embodiment.

[0051] Figure 2 The PMSM classical model predictive control flowchart of the first embodiment.

[0052] Figure 3 The nonlinear mathematical model diagram of the neuron of the first embodiment.

[0053] Figure 4 The linear adaptive neural training flowchart of the first embodiment.

[0054] Figure 5 The PMSM predictive control flowchart of the first embodiment based on the normalized linear adaptive neural network. Detailed Implementation

[0055] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0056] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0057] This invention is based on predictive control theory. It replaces all coefficients related to system circuit parameters in traditional model predictive control with neural network weight coefficients. By measuring the three-phase AC voltage and current values ​​and the DC bus voltage, a linear adaptive neural network is trained online. The trained weight coefficients predict the system's state over a future period. An appropriate cost function is designed, and the optimal switching state corresponding to the minimum cost function value is selected and applied to the motor drive controller. Simultaneously, to reduce the number of controller parameters that need adjustment, the weight coefficients in the linear adaptive neural network are normalized by per-unit scaling.

[0058] This invention effectively overcomes the problems of difficulty in tracking reference signals caused by fluctuations in system circuit parameters in traditional solutions. Without adding extra hardware, it establishes a new predictive control method, improves the reliability of motor drive systems in complex environments, and avoids economic losses and safety accidents caused by significant degradation in motor drive control performance in energy storage and new energy electric vehicles.

[0059] Example 1

[0060] One embodiment of this disclosure provides an adaptive predictive control method for permanent magnet synchronous motors based on neural networks, such as... Figure 1 As shown, it includes the following steps:

[0061] Step S1: Based on the measured stator voltage and current at the current moment, calculate the stator current at the moment to be predicted using the reconstructed stator current prediction equation.

[0062] Step S2: Iterate through a preset number of switching states and calculate the stator voltage corresponding to each switching state using the working principle of the converter.

[0063] Step S3: according to the stator voltage corresponding to each switch state, the stator current at the to-be-predicted moment is time-delay compensated.

[0064] Step S4: according to the compensated stator current and a preset reference current, an optimal switch state corresponding to a minimum cost function value is selected based on a designed cost function and applied to the motor drive controller.

[0065] The reconstructed stator current prediction equation is obtained by using the weight of the neural network to replace the coefficients related to the circuit parameters in the initial stator current prediction equation constructed according to the PMSM model.

[0066] Further, the PMSM model is constructed based on a stator voltage equation and a stator flux linkage equation, and finally is:

[0067]

[0068] wherein u d , u q are d-q axis components of the stator voltage; i d , i q are dq axis components of the stator current; R s is a stator resistance; ψ d , ψ q are d-q axis components of the stator flux linkage; ω e is an electrical angular velocity; L d , L q are d-q axis inductance components; and ψ m is a permanent magnet flux linkage.

[0069] Further, the stator voltage equation is:

[0070]

[0071] The stator flux linkage equation is:

[0072]

[0073] wherein u d , u q are d-q axis components of the stator voltage; i d , i q are dq axis components of the stator current; R s is a stator resistance; ψ d , ψ q are d-q axis components of the stator flux linkage; ω e is an electrical angular velocity; L d , L q are d-q axis inductance components; and ψ mis a permanent magnet flux linkage.

[0074] Further, the reconstructed stator current prediction equation is expressed by a formula as follows:

[0075]

[0076] wherein, is a current value dq-axis component at k+1 moment, is a current value dq-axis component at k moment, ω x (x=1, 2, …, 7) is a weight actual value.

[0077] Further, the weight of the neural network is obtained by training the neural network, and the training of the neural network is to correct the weight actual value of the linear adaptive neural network by using the error between the actual output vector and the expected output vector, specifically as follows:

[0078]

[0079]

[0080]

[0081] wherein, F is a quadratic error function, used for calculating the error between the actual output vector and the expected output vector; is a unit value of weight correction value at k moment; is a unit value of weight correction value at k-1 moment; is a unit value of weight at k+1 moment; β is a first momentum term coefficient.

[0082] Further, the delay compensation is to introduce one beat of delay compensation on the premise of the stator current at the to-be-predicted moment, specifically as follows:

[0083] The stator voltage corresponding to each switching state and the stator current at the to-be-predicted moment are input into the reconstructed stator current prediction equation, and the stator current at the next moment at the to-be-predicted moment is calculated and returned.

[0084] Further, the cost function is specifically as follows:

[0085]

[0086] wherein, λ lim is a weight coefficient, i lim is a current limit maximum value, and is a reference current, is a current value dq-axis component at k+2 moment, i.e. the compensated stator current.

[0087] The implementation process of the adaptive predictive control method of the permanent magnet synchronous motor based on the neural network of the embodiment will be described in detail below.

[0088] For a permanent magnet synchronous motor (PMSM) drive system, the embodiment proposes a predictive control method based on a normalized linear adaptive neural network, which ensures that the PMSM maintains high-performance operation under complex working conditions. In the following, taking an interior PMSM driven by a two-level inverter as an example, the PMSM model construction, the classical predictive current control algorithm, the basic principles of linear adaptive neural network, and the specific implementation of the adaptive predictive control method of the permanent magnet synchronous motor based on the neural network are introduced respectively.

[0089] 1. PMSM model construction

[0090] The PMSM model is used to calculate the stator voltage under different switching states. To simplify the analysis, first make idealized assumptions for a three-phase PMSM:

[0091] (1) Ignore the saturation of the motor core;

[0092] (2) Do not consider the eddy current and hysteresis loss in the motor;

[0093] (3) The current in the motor is a symmetrical three-phase sinusoidal current.

[0094] For ease of controller design and application, the mathematical model in the synchronous rotating coordinate system d-q is selected, and the stator voltage equation is expressed as:

[0095]

[0096] The stator flux equation is:

[0097]

[0098] Then the final PMSM model is:

[0099]

[0100] Where, u d , u q are the d-q axis components of the stator voltage; i d , i q are the dq axis components of the stator current; R s is the stator resistance; ψ d , ψ q are the d-q axis components of the stator flux; ω e is the electrical angular velocity; L d , Lq These are the dq-axis inductance components; ψ m It is a permanent magnet flux linkage.

[0101] 2. PMSM Classic Predictive Current Control Algorithm

[0102] This is an existing control algorithm whose advantage lies in its ability to predict the system's state changes (i.e., current changes) over the next n time steps based on measurement and estimation data combined with a PMSM model. It then performs rolling optimization according to the principle of minimizing the cost function to determine the optimal operation at that moment, i.e., the switching state. This method, which considers future states, has many advantages over traditional methods, such as rapid dynamic response, strong online optimization capability, simple structure, and ease of adding constraints. Figure 2 As shown, where ω * The target rotational speed is given by ω, and the reference current is obtained through the speed control loop PI controller. and J is the cost function value, consistent with the cost mentioned below; ω e θ is the electric rotational speed, and θ is the rotor angle. e denoted as the rotor electrical angle, and n as the number of pole pairs of the motor.

[0103] Obtain the stator current i at time k from motor M. abc (k), rotational speed ω and rotor angle θ, stator current i abc (k) After the Park transformation, the dq-axis components i of the stator current are obtained. d i q Based on the rotational speed ω and the rotor angle θ, the electric rotational speed ω is calculated. e and rotor electrical angle θ e .

[0104] Based on the parameter values ​​obtained above, the current value at the time to be predicted (i.e., time k+1) is calculated using the prediction equation. The prediction equations are constructed using existing system circuit parameters, specifically:

[0105]

[0106] Where T is the control period, R s It is the stator resistance, L d L q These are the dq axis inductance components, respectively.

[0107] In practical engineering, predictive control inherently involves a larger computational load compared to traditional methods, and there are delays generated during sensor and signal transmission. Therefore, delay compensation is required, i.e., a one-beat delay compensation is introduced. At this point, the new future current prediction value... The predicted value at k+2 is used instead of the predicted value at k+1.

[0108] The stator voltage corresponding to each switching state is calculated by using the working principle of the converter, i.e. the stator voltage at the next sampling time corresponding to each switching state:

[0109] u d (k+1)=V dc *cosθ e *v α (n)+V dc *sinθ e *v β (n)

[0110] u q (k+1)=-V dc *sinθ e *v α (n)+V dc *cosθ e *v β (n)

[0111] wherein V dc is the voltage of the motor driving DC power supply, θ e is the rotor electric angle, v α [n]={0,0.6667,0.3333,-0.6667,-0.3333,0.3333,0},v β [n]={0,0,0.5774,0.5774,0,-0.5774,-0.5774,0} are the voltage vector coefficients corresponding to the 8 switching states, n=1,2…,8

[0112] The stator voltage corresponding to each switching state is calculated by using the working principle of the converter, i.e. the stator voltage at the next sampling time corresponding to each switching state:

[0113]

[0114] Finally, the predicted current value and the reference current and are used to calculate the cost of each switching state according to the following cost function, and the switching state with the minimum cost is used as the switching state at the next time:

[0115]

[0116] wherein λ lim is the weight coefficient, which is generally set by experience to optimize the control effect, and in this case, to suppress the current overrun, i lim is the maximum current limit.

[0117] 3. Linear adaptive neural network basic principle

[0118] The above-mentioned PMSM classical predictive current control algorithm predicts the future current value through the existing system circuit parameters, and is very sensitive to the changes of circuit model parameters (inductance, resistance, flux linkage, number of pole pairs, etc.), so the prediction accuracy is not high. Therefore, a linear adaptive neural network is introduced to intelligently predict the current value.

[0119] The basic principle of linear adaptive neural network will be introduced below. Linear adaptive neural network belongs to artificial neural network technology. Artificial neural network technology abstracts and simulates the human brain neural network, extracts features from historical data, learns and trains, and completes specific tasks such as parameter estimation, pattern recognition, fault diagnosis, and predictive control. Artificial neural network includes linear neural network, radial basis neural network, competitive learning neural network, back propagation neural network, etc. Among them, M-P neuron is the most widely used basic unit in linear neural network, as shown in Figure 3 , mainly composed of input signal x1…x n , connection weight w1…w n , summation unit ∑, threshold θ, activation function and output signal y.

[0120] In the field of power electronics and power transmission, due to the demand for rapid real-time response of the control system, and the limited computing power of the current controller, etc., a series of intelligent algorithms represented by neural network are difficult to realize online training and learning on the controller. Compared with other types of neural networks, linear adaptive neural network has the advantages of simple structure, small calculation burden, and easy parameter design, and is relatively easy to design for the control system of a single machine system. Therefore, under the existing hardware performance conditions, it has certain practical significance and application prospect.

[0121] The learning process of linear adaptive neural network is a typical teacher learning method, which adopts Widrow-Hoff learning rule (LMS learning rule). In the training process, the network is constantly trained with the pattern pairs in the training set. When the neural network training pattern is given, the output unit will produce an actual output vector, and the error between the actual output vector and the expected output vector is used to correct the weight of the neural network. The specific training process is shown in Figure 4 , during online learning, the data set sampled at the last time is used as the training set, assuming that the sample is p, which has input / output pattern pairs {x p}, {t p}, the input of the jth neuron of the output layer under the action of sample p is represented as:

[0122]

[0123] where θ j is the threshold of the output layer neuron j, and N is the input number of the jth neuron.

[0124] The output of the jth neuron of the output layer is:

[0125]

[0126] where M represents the number of neurons, and f is a linear activation function, and when the neuron is activated, the output is equal to the input, that is:

[0127]

[0128] The quadratic error function is constructed for the current sample:

[0129]

[0130] where e j represents the error of the jth neuron.

[0131] The weight coefficient correction formula of any neuron j of the output layer is:

[0132]

[0133] The above process is essentially to use the gradient steepest descent method, and the weight is changed along the negative gradient direction of the error function to obtain the final weight of the neural network.

[0134] 4. Adaptive predictive control algorithm based on normalized linear neural network

[0135] Based on the above linear adaptive neural network basic principle, the embodiment proposes a permanent magnet synchronous motor predictive control scheme based on normalized linear adaptive neural network, and the advantage is that it can mine information from measured data and estimated data, and train the neural network online, and the permanent magnet synchronous motor drive system can adaptively track the changes of system circuit parameters, such as Figure 5 As shown in the figure, specifically:

[0136] According to the PMSM model, the initial stator current prediction equation, that is, formula (4), is constructed, the weight of the neural network is used to replace the coefficients related to the circuit parameters in the prediction equation, and the stator current prediction equation is reconstructed.

[0137] Combined with the stator voltage and current measured by the sensor at the current moment, the stator current at the next sampling moment is predicted;

[0138] Eight switching states are traversed, and according to the working principle of the converter, the stator voltage corresponding to each switching state at the next sampling moment is calculated and predicted;

[0139] Based on the predicted stator voltage corresponding to various switch states, the predicted stator current at the next moment is delayed to obtain the predicted stator current value at the next sampling moment;

[0140] According to the compensated stator current value and the preset reference current, a cost function is constructed, and the optimal switch corresponding to the minimum cost function is selected for application in motor driving;

[0141] Among them, the reconstruction of the predicted stator current equation is to replace the 7 coefficients related to the circuit parameters in the formula 5 predicted equation in the classical predictive current control algorithm of PMSM with 7 weight initial values (i.e. weight reference value) in the linear adaptive neural network, specifically:

[0142]

[0143] 0 is taken as the threshold value of the linear adaptive neural network, and is taken as the activation function, the data collected by the sensor during operation is taken as the input data set (training set) of the neural network, the predicted current value is taken as the output, the neural network is trained, and based on the setting of the learning rate, the 7 weights in the neural network are continuously optimized.

[0144] In the process of setting the learning rate, due to the different dimensions of each weight, the design of the learning rate is relatively complex, therefore, the embodiment adopts the normalization concept, normalizes the seven weights, and then changes the original seven learning rates to be set into only one learning rate:

[0145]

[0146] Among them, ω Nx (x=1, 2, …, 7) is the weight reference value, ω x (x=1, 2, …, 7) is the weight actual value, is the weight unit value after normalization.

[0147] At the same time, in order to speed up the training speed of the neural network, a momentum term is introduced, therefore, the weight update process, i.e. the linear adaptive neural network, is represented by the formula:

[0148]

[0149] Among them, F is a quadratic error function; is the unit value of the weight correction value at k moment; is the unit value of the weight correction value at k-1 moment; is the unit value of the weight at k+1 moment; β is the coefficient of the first momentum term, and due to the normalization processing, the same coefficient value is used for the seven weights.

[0150] Embodiment two

[0151] In an embodiment of the present disclosure, a neural network-based adaptive predictive control system for permanent magnet synchronous motor is provided, comprising a current prediction module, a voltage calculation module, a delay compensation module and an optimal selection module:

[0152] The current prediction module is configured to calculate the stator current at the predicted time based on the measured stator voltage and current at the current time, and using the reconstructed stator current prediction equation.

[0153] The voltage calculation module is configured to calculate the stator voltage corresponding to each switching state by traversing a preset number of switching states and using the working principle of the inverter.

[0154] The delay compensation module is configured to compensate the stator current at the predicted time according to the stator voltage corresponding to each switching state.

[0155] The optimal selection module is configured to select the optimal switching state corresponding to the minimum cost function value based on the designed cost function according to the compensated stator current and the preset reference current, and apply the optimal switching state to the motor drive controller.

[0156] The reconstructed stator current prediction equation is obtained by using the weight of the neural network to replace the coefficients related to the circuit parameters in the initial stator current prediction equation constructed according to the PMSM model.

[0157] Embodiment three

[0158] The purpose of the present embodiment is to provide a computer-readable storage medium.

[0159] The computer-readable storage medium stores a computer program, which is executed by a processor to implement the steps of the neural network-based adaptive predictive control method for permanent magnet synchronous motor according to Embodiment One of the present disclosure.

[0160] Embodiment four

[0161] The purpose of the present embodiment is to provide an electronic device.

[0162] The electronic device comprises a memory, a processor and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the neural network-based adaptive predictive control method for permanent magnet synchronous motor according to Embodiment One of the present disclosure.

[0163] The above merely provides the preferred embodiments of the present application, and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the principles and technical scope of the present application shall fall into the scope of the present application.

Claims

1. A neural network-based adaptive predictive control method for permanent magnet synchronous motor, characterized in that, The method comprises the following steps: Based on the measured current moment stator voltage and current, the reconstructed stator current prediction equation is used to calculate the stator current at the predicted moment; Traverse a preset number of switching states, and calculate the stator voltage corresponding to each switching state by using the working principle of the converter; According to the stator voltage corresponding to each switching state, the stator current at the predicted moment is compensated for delay; According to the compensated stator current and the preset reference current, the optimal switching state corresponding to the minimum cost function value is selected based on the designed cost function and applied to the motor drive controller; The cost function is specifically: wherein, is a weight coefficient, is a current limit maximum value, and is a reference current, , is the dq-axis component of the current value at the instant, i.e. the compensated stator current; Wherein, the reconstructed stator current prediction equation is obtained by replacing the coefficients related to the circuit parameters in the initial stator current prediction equation with the weights of the neural network according to the PMSM model; the weights of the neural network are obtained by training the neural network, and the training of the neural network is to correct the actual value of the weights of the linear adaptive neural network by using the error between the actual output vector and the expected output vector, specifically: wherein, is a quadratic error function for calculating the error between the actual output vector and the desired output vector; is the unit value of the weight correction value at time k; is the unit value of the weight correction value at time k-1; is the unit value of the weight at time k+1; is a coefficient of the first momentum term.

2. The neural network-based adaptive predictive control method for permanent magnet synchronous motor according to claim 1, wherein, The PMSM model is constructed based on the stator voltage equation and the stator flux equation, and finally: in, , These are the dq-axis components of the stator voltage, respectively. , These are the dq-axis components of the stator current, respectively. It is the stator resistance; , For the dq-axis components of the stator flux linkage; It is electric angular velocity; , These are the dq axis inductance components; It is a permanent magnet flux linkage.

3. The neural network-based adaptive predictive control method for permanent magnet synchronous motor according to claim 2, wherein, The stator voltage equation is: The stator flux equation is: in, , These are the dq-axis components of the stator voltage, respectively. , These are the dq-axis components of the stator current, respectively. It is the stator resistance; , For the dq-axis components of the stator flux linkage; It is electric angular velocity; , These are the dq axis inductance components; It is a permanent magnet flux linkage.

4. The neural network-based adaptive predictive control method for permanent magnet synchronous motor according to claim 1, wherein, The reconstructed stator current prediction equation is expressed by formula as: wherein, , is the current value dq-axis component at the time instant, is the current value dq-axis component at the time instant, , is the current value dq-axis component at the time instant, is the current value dq-axis component at the time instant, is the weight value actual value.

5. The neural network-based adaptive predictive control method for permanent magnet synchronous motor according to claim 1, wherein, The delay compensation is a one-beat delay compensation under the premise of the stator current at the predicted moment, specifically: The stator voltage corresponding to each switching state and the stator current at the predicted moment are input into the reconstructed stator current prediction equation, and the stator current at the next moment at the predicted moment is calculated and returned.

6. A neural network-based adaptive predictive control system for permanent magnet synchronous motor, characterized in that, It comprises a current prediction module, a voltage calculation module, a delay compensation module and an optimal selection module: The current prediction module is configured to calculate the stator current at the predicted moment based on the measured current moment stator voltage and current by using the reconstructed stator current prediction equation; The voltage calculation module is configured to traverse a preset number of switching states, and calculate the stator voltage corresponding to each switching state by using the working principle of the converter; The delay compensation module is configured to compensate for the delay of the stator current at the predicted moment according to the stator voltage corresponding to each switching state; The optimal selection module is configured to select the optimal switching state corresponding to the minimum cost function value based on the designed cost function according to the compensated stator current and the preset reference current, and apply it to the motor drive controller; the cost function is specifically: wherein is a weight coefficient, is a current limit maximum value, and is a reference current, , is the dq-axis component of the current value at the instant, i.e. the compensated stator current; Wherein, the reconstructed stator current prediction equation is obtained by replacing the coefficients related to the circuit parameters in the initial stator current prediction equation with the weights of the neural network according to the PMSM model; the weights of the neural network are obtained by training the neural network, and the training of the neural network is to correct the actual value of the weights of the linear adaptive neural network by using the error between the actual output vector and the expected output vector, specifically: wherein, is a quadratic error function for calculating the error between the actual output vector and the desired output vector; is the unit value of the weight correction value at time k; is the unit value of the weight correction value at time k-1; is the unit value of the weight at time k+1; is a coefficient of the first momentum term.

7. An electronic device, characterized in that it comprises: a memory for non-transitory storage of computer readable instructions; and a processor for running the computer readable instructions, The computer readable instructions, when executed by the processor, perform the method of any one of claims 1-5.

8. A storage medium characterized by, Non-transitory computer readable instructions are stored, wherein the non-transitory computer readable instructions, when executed by a computer, perform the instructions of the method of any one of claims 1-5.

Images