Double-fed motor double-vector model predictive control method and system based on neural network

Abstract

This invention provides a neural network-based two-vector model predictive control method and system for a doubly-fed induction generator (DFIG), relating to the field of DFIG control technology. The method includes: acquiring the stator current, rotor current, slip angular velocity, and rotor voltage of the DFIG's two-vector model in a two-phase rotating coordinate system at the current moment, inputting these values ​​into a constructed rotor current prediction model to predict the rotor current at the next moment; selecting the optimal vector based on the predicted rotor current, substituting the optimal vector and the zero vector into a value function to obtain the optimal duty cycle; and using the optimal vector and optimal duty cycle to control the DFIG's two-vector model. This invention effectively solves the problems of high modeling complexity and poor parameter matching accuracy in current DFIG models.

Description

Technical Field

[0001] This invention relates to the field of doubly-fed motor control technology, and in particular to a doubly-fed motor dual-vector model predictive control method and system based on neural networks. Background Technology

[0002] A doubly-fed induction generator (DFIG) is a wound-rotor asynchronous motor whose stator windings are directly connected to the power grid, while the rotor windings are connected to the grid via a back-to-back converter. The DFIG dual-vector model, by introducing a dual-vector control strategy, enables decoupling and coordinated control of physical quantities such as voltage, current, and flux linkage of the stator and rotor in the dq rotating coordinate system, thereby achieving effective motor control.

[0003] In the field of doubly-fed induction generator (DFIG) control, traditional control methods typically rely on highly accurate motor models. However, these motor models often exhibit highly nonlinear and complex dynamic characteristics, making modeling extremely difficult. Furthermore, in actual operation, complex motor model systems can suffer from parameter mismatches, such as motor parameters deviating from design values ​​due to temperature and load variations, leading to severely weakened control performance. This manifests not only as significant current ripple, affecting motor operating efficiency and stability, but in more severe cases, it can even cause the entire system to become completely uncontrollable, posing significant risks and economic losses to industrial production and equipment operation.

[0004] While existing technologies have attempted to alleviate the aforementioned problems to some extent by optimizing control algorithm parameters and employing more accurate model identification methods, these methods have not fundamentally solved the constraints on motor model control performance caused by high modeling complexity and poor parameter matching accuracy. Furthermore, most existing technologies utilize single-vector control, which cannot flexibly adjust the amplitude and direction of the voltage vector, resulting in poor dynamic response and control accuracy. Under non-ideal conditions, it is difficult to ensure stable and efficient system operation. Therefore, a new control method is urgently needed to overcome the bottlenecks of existing technologies and open up new paths for improving the performance of doubly-fed motor models, in order to meet the stringent robustness requirements of practical industrial applications. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a neural network-based dual-vector model predictive control method and system for doubly-fed motors, which solves the problems of high modeling complexity and poor parameter matching accuracy in current doubly-fed motor models, avoids motor runaway, and improves the parameter robustness of motor model predictive control.

[0006] This invention provides a two-vector model predictive control method for a doubly-fed induction generator based on a neural network, the method comprising: The stator current, rotor current, slip angular velocity, and rotor voltage of the doubly fed motor dual-vector model in the two-phase rotating coordinate system at the current moment are collected and input into the constructed rotor current prediction model to predict the rotor current at the next moment. The optimal vector is selected based on the predicted rotor current. The optimal vector and the zero vector are substituted into the value function to obtain the optimal duty cycle. The optimal vector and the optimal duty cycle are used to control the doubly fed motor dual-vector model.

[0007] Furthermore, the method for constructing the rotor current prediction model includes: Record the data during the operation of the doubly-fed induction generator (DFIG) dual-vector model to generate a sample training set; wherein, the sample training set includes multiple sets of input data and output data; the input data and the output data correspond one-to-one, the input data are the stator current, rotor current, slip angular velocity and rotor voltage collected at time k during motor operation; the output data is the rotor current collected at time k+1 during motor operation; The generated sample training set is input into the neural network model for training, and the rotor current prediction model is obtained after convergence.

[0008] Furthermore, a dual vector is generated based on the switching state of the inverter bridge arm in the dual-vector model of the doubly fed motor; wherein the dual vector includes an effective vector and a zero vector; and the optimal vector is one of the vectors selected from the effective vectors.

[0009] Furthermore, the value function for: ; in, , These represent the rotor current command values ​​in the two-phase rotating coordinate systems, respectively. , Let represent the rotor current at time k; , These represent the change in rotor current during one sampling period at time k.

[0010] Furthermore, the change in rotor current over one sampling period is: ; in, Indicates the duty cycle of the effective vector; , This represents the change in rotor current within one sampling period of the effective vector; , This represents the change in rotor current within one sampling period of the zero vector.

[0011] Furthermore, the optimal duty cycle is obtained in the following ways: By taking the derivative of the value function, the minimum value of the value function is obtained when the derivative is zero, and thus the corresponding duty cycle is obtained. If duty cycle If so, the duty cycle will be output directly; If duty cycle Then let , exist The minimum value in the interval is 1; If duty cycle Then let , exist The minimum value in the interval is 0.

[0012] Furthermore, the correspondence between the switching state and the dual vector is as follows: ; in, , , , , , It is an effective vector, a non-zero vector, and plays an actual control role in the doubly-fed motor dual-vector model; , It is a zero vector; , , This indicates the switching status of the three pairs of bridge arms of a three-phase two-level inverter; , , Representing voltage vectors respectively The three bits of the binary code.

[0013] Furthermore, the loss function of the rotor current prediction model for: ; in, This indicates the number of data sets in the training set. Indicates the first The first class of data One data point; Indicates the first The first class of data One data point; , All are positive integers; 1 / 2 is a constant; This represents the rotor current prediction model. Indicates the weighting coefficient. This represents the bias coefficient.

[0014] Furthermore, the gradient descent method is used to train the weight coefficients and bias coefficients of the neural network; The weighting coefficient and bias coefficient The update formula is: ; ; in, Indicates the learning rate; This indicates calculating the gradient; Represents the loss function; Indicates to Find the partial derivative.

[0015] On the other hand, the present invention proposes a doubly fed motor dual-vector model predictive control system based on neural networks, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of any of the above-described methods.

[0016] In summary, this invention provides a method and system for two-vector model predictive control of a doubly-fed motor based on neural networks. Compared with existing technologies, the technical solution conceived in this invention can achieve the following beneficial effects: (1) This invention introduces a neural network model into the model predictive control of a doubly-fed induction generator (DFIG), proposing a model-free predictive control method. By collecting historical data of motor operation, the stator current, rotor current, slip angular velocity, and rotor voltage collected at time k during motor operation are used as input data for the neural network model, and the rotor current collected at time k+1 during motor operation is used as output data for the neural network model. In this way, a rotor current prediction model is trained to directly predict the rotor current at the next time step. This avoids the deterioration of control performance caused by excessive system modeling complexity or parameter mismatch, and greatly optimizes the dynamic response stability of the motor under non-ideal conditions, thereby improving the robustness of motor control.

[0017] (2) This invention introduces effective vector and zero vector, and adopts a dual vector control method combining effective vector and zero vector to control the motor. The amplitude and direction of the voltage vector can be flexibly adjusted to improve its dynamic response stability and control accuracy. In addition, the introduction of zero vector can not only greatly reduce the exponential calculation pressure and control pulsation, and reduce the computing power requirements of the controller, but also avoid the steady-state effect caused by the long vector action time, thus meeting the stringent requirements for motor control in actual industrial applications. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the method steps of a doubly-fed motor dual-vector model predictive control method and system based on neural networks provided by the present invention; Figure 2 This is a schematic diagram illustrating the principle of doubly-fed motor dual-vector model predictive control, a method and system for doubly-fed motors based on neural networks, provided by this invention. Figure 3 It is a predictive control method for a single-vector model of a doubly-fed motor in the existing technology; Figure 4 This is a schematic diagram of current ripples in a single-vector model predictive control under ideal conditions; Figure 5 This is a schematic diagram of a three-phase two-level inverter structure of a doubly-fed motor dual-vector model predictive control method and system based on neural networks provided by the present invention. Figure 6 This invention provides a three-phase two-level inverter voltage vector diagram of a doubly-fed motor dual-vector model predictive control method and system based on neural networks; Figure 7 This is a schematic diagram of the neuron structure of a doubly fed motor dual-vector model predictive control method and system based on neural networks provided by the present invention; Figure 8 This is a schematic diagram of the neural network model structure of a doubly fed motor dual-vector model predictive control method and system based on neural networks provided by the present invention. Figure 9 This is a schematic diagram of rotor current prediction model training for a doubly fed motor dual-vector model predictive control method and system based on neural networks provided by the present invention. Figure 10 This is a schematic diagram of the output flow of the optimal vector and optimal duty cycle of a doubly fed motor dual-vector model predictive control method and system based on neural networks provided by the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0021] It should be noted that, in the description of the embodiments of the present invention, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a method, step, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to the method, step, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the method, step, or apparatus that includes that element.

[0022] To address the problems of high modeling complexity and poor parameter matching accuracy in current doubly-fed motor models, this invention proposes a neural network-based dual-vector model predictive control method and system for doubly-fed motors to improve the parameter robustness of motor model predictive control.

[0023] Specifically, such as Figure 1 and Figure 2 As shown, the method includes: S101: Collect the stator current, rotor current, slip angular velocity, and rotor voltage of the doubly fed motor dual-vector model in the two-phase rotating coordinate system at the current moment, input them into the constructed rotor current prediction model, and predict the rotor current at the next moment.

[0024] As an example, the establishment of the doubly-fed motor dual-vector model includes: obtaining the mathematical model of the doubly-fed motor in a two-phase rotating coordinate system, obtaining the state equation with rotor current as the state variable; and using the state equation to establish the doubly-fed motor dual-vector model.

[0025] It should be noted that the mathematical model includes voltage equations and flux linkage equations.

[0026] The voltage equation can be: ; in, , These represent the stator voltages, specifically the stator dq-axis voltages; , These represent the rotor voltages, specifically the dq-axis voltages of the rotor. , These represent the stator currents, specifically the stator dq-axis currents; , These represent the rotor currents, specifically the dq-axis currents of the rotor. , These represent the stator flux linkage, specifically the stator dq axis flux linkage; , These represent rotor flux linkages, specifically the flux linkages along the rotor's d and q axes. Indicates the stator winding resistance; Indicates the rotor winding resistance; Indicates the angular velocity of the stator current. This indicates the rotor current angular velocity.

[0027] The flux linkage equation can be: ; in, Indicates the self-inductance of the stator winding; Indicates the self-inductance of the rotor winding; This indicates the mutual inductance between the stator and rotor windings.

[0028] Based on the voltage equation and the flux linkage equation, the state equation with rotor current as the state variable is obtained. The state equation can be: ; in, ; Indicates slip angular velocity, , , These represent the angles between the stator / rotor phase A and the direct axis of the reference coordinate system, respectively.

[0029] like Figure 3 As shown, this is a predictive control method for a single-vector model of a doubly-fed motor in the prior art. It mainly sorts the value functions corresponding to all possible switching combinations by exhaustive search, and then selects the voltage vector that minimizes the cost function to obtain the corresponding switching combination, which is used as the switching state of the inverter at the next moment.

[0030] However, the motor in one sampling period Within, the selected voltage vector will cause the current to change. When the actual current exceeds the reference value, but the sampling period has not yet ended, the current will continue to increase until the next sampling period when the voltage vector changes. This causes significant pulsation in the control, placing high demands on the controller's computing power and affecting motor stability due to the excessively long vector action time. Figure 4 The figure shows a schematic diagram of current ripple in a single-vector model predictive control under ideal conditions. However, in actual control, due to the diversification of control objectives and the existence of parameter mismatch, the error is more significant. Therefore, reducing the error is extremely important for improving the control effect.

[0031] Without changing the controller cycle, multi-vector control can also be used within a single sampling period. However, the typical logic of multi-vector control involves dividing a sampling period into N equal segments, selecting one vector for each segment, exhaustively enumerating 7N voltage vectors to obtain 7N corresponding switch combinations, calculating the value function for each switch combination, and then selecting the voltage vector corresponding to the switch combination with the smallest value function. While this multi-vector approach reduces ripple, the exponentially increased computational load places stricter demands on the controller's computing power.

[0032] Therefore, in order to reduce control pulsation without increasing computational burden, this invention utilizes a dual-vector control method that combines effective vector and zero vector.

[0033] As an example, a dual vector is generated based on the switching state of the inverter arm in the dual-vector model of the doubly fed motor; wherein the dual vector includes an effective vector and a zero vector; and the optimal vector is one of the vectors selected from the effective vectors.

[0034] like Figure 5 The diagram shown is a schematic of a three-phase two-level inverter. a, b, and c represent three pairs of bridge arms. , , This indicates the switching status of the three pairs of bridge arms; These represent the three bits of the binary code for the voltage vector; in each pair of bridge arms, the upper bridge arm being on and the lower bridge arm being off indicates a switch state of 1, while the upper bridge arm being off and the lower bridge arm being on indicates a switch state of 0.

[0035] As an example, the correspondence between the switch state and the two vectors is as follows: ; in, , , , , , It is an effective vector, a non-zero vector, and plays an actual control role in the two-vector model of the doubly-fed motor; , It is a zero vector; , , This indicates the switching status of the three pairs of bridge arms of a three-phase two-level inverter; , , Representing voltage vectors respectively The three bits of the binary code.

[0036] For example, based on the switching state of the bridge arm, it is possible to generate, such as Figure 6 The voltage vector diagram of the three-phase two-level inverter shown includes 6 active vectors and 2 zero vectors. The 6 active vectors are as follows: , , , , , It plays an actual control role for the motor; the two zero vectors are: , It does not output control voltage.

[0037] When the motor is running, the stator current, rotor current, slip angular velocity, and rotor voltage of the doubly fed motor dual-vector model in the two-phase rotating coordinate system at the current moment are collected and input into the constructed rotor current prediction model to predict the rotor current at the next moment.

[0038] It should be noted that the trained neural network model constructs a network structure consisting of multiple interconnected processing units (neurons). These neurons, through weighted connections and nonlinear activation functions, are able to perform multi-level abstraction and feature extraction of input data, thereby achieving efficient learning and modeling of complex data.

[0039] like Figure 7 The diagram shows a neuron structure that receives signals from other neurons. It can multiply the input signals (input 1, input 2, input 3) with the corresponding weights (weight 1, weight 2, weight 3), sum the results, and then generate the output through a non-linear activation function.

[0040] like Figure 8 The diagram shows a complete neural network model structure. The input layer receives external input data. Hidden layers extract and transform features from the input data; more layers result in better fitting, but also increase computational cost. The output layer produces the final output of the neural network, and the number of neurons in the output layer depends on the task type.

[0041] As an example, the method for constructing a rotor current prediction model includes: Record the data during the operation of the doubly fed motor dual-vector model to generate a sample training set.

[0042] The sample training set includes multiple sets of input and output data; the input data and output data correspond one-to-one. The input data are the stator current, rotor current, slip angular velocity and rotor voltage collected at time k when the motor is running; the output data is the rotor current collected at time k+1 when the motor is running.

[0043] The generated sample training set is input into the neural network model for training, and the rotor current prediction model is obtained after convergence.

[0044] It should be noted that this invention uses the mean square error as the loss function. As an example, the loss function of the rotor current prediction model... for: ; in, This indicates the number of data sets in the training set. Indicates the first The first class of data One data point; Indicates the first The first class of data One data point; , All are positive integers; 1 / 2 is a constant used to cancel out the 2 multiplied when calculating the gradient, which facilitates subsequent calculations and does not affect the final result; This represents the rotor current prediction model. Indicates the weighting coefficient. This represents the bias coefficient.

[0045] Furthermore, the gradient descent method is used to train the weight coefficients and bias coefficients of the neural network; Weighting coefficient and bias coefficient The update formula is: ; ; in, Indicates the learning rate; This indicates calculating the gradient; Represents the loss function; Indicates to Find the partial derivative.

[0046] Data from the operation of the doubly-fed induction generator (DFIG) dual-vector model is recorded to generate a sample training set. Based on the update rate, the model is trained until W and B converge, thus obtaining the rotor current prediction model. For example... Figure 9 As shown, the input values ​​for this model are: , , , , , , The corresponding output value is: , .

[0047] S102: Select the optimal vector based on the predicted rotor current, substitute the optimal vector and the zero vector into the value function to obtain the optimal duty cycle; use the optimal vector and the optimal duty cycle to realize the control of the doubly fed motor dual-vector model.

[0048] It should be noted that the optimal vector is one of the vectors selected from the effective vectors.

[0049] The optimal vector is selected as follows: obtain the effective vector and the zero vector, obtain the switch combination corresponding to the voltage vector, calculate the value function of each switch combination, and then select the voltage vector corresponding to the switch combination with the smallest value function. This voltage vector is the optimal vector.

[0050] As an example, a quadratic function is selected as the value function, and the rotor dq-axis current is used as the control target. No value function weights are set between the dq-axis current control targets. Then the value function... for: ; in, , These represent the rotor current command values ​​in the two-phase rotating coordinate systems, respectively. , Let represent the rotor current at time k; , These represent the change in rotor current during one sampling period at time k.

[0051] Since the current change within a sampling period is composed of the effects of the selected effective vector and the zero vector, if the selected effective vector is D, and the remaining time is affected by the zero vector, then the rotor current change within a sampling period is further: ; in, Indicates the duty cycle of the effective vector; , This represents the change in rotor current within one sampling period of the effective vector; , This represents the change in rotor current within one sampling period of the zero vector.

[0052] Furthermore, when the effective vector is applied, the change in rotor current within one sampling period is: ; When the zero vector is applied, the voltage will naturally drop, affecting the steady state; this factor also needs to be considered. Therefore, the change in rotor current during one sampling period due to the zero vector is: ; Rotor current at time k+1 and This was obtained by discretizing the current-state equations using the forward Euler method. Specifically: ; The forward Euler method is a commonly used method for discretization: ;in, Indicates the sampling period.

[0053] It should be noted that after selecting an effective vector, the effect time of the voltage vector and the deviation of the current have a simple linear relationship. Therefore, the duty cycle of the selected voltage vector can be calculated to avoid poor steady-state performance due to excessively long vector effect time.

[0054] As an example, the optimal duty cycle can be obtained in the following ways: Since the value function is a quadratic function of the duty cycle, taking the derivative of the value function and finding its minimum when the derivative is zero yields the corresponding optimal duty cycle. Furthermore, as... Figure 10 As shown, if the duty cycle Select the optimal vector corresponding to the minimum value of the value function, and calculate the rotor current at time k+1 under the action of the optimal vector. Then by calculate This yields the optimal duty cycle.

[0055] If duty cycle If the duty cycle is 0, then the duty cycle is output directly; if the duty cycle is 10, then the duty cycle is output Then let , exist The minimum value in the interval is 1; If duty cycle Then let , exist The minimum value in the interval is 0.

[0056] Finally, the optimal vector and the corresponding optimal duty cycle are applied to the doubly fed motor two-vector model.

[0057] On the other hand, the present invention also proposes a doubly fed motor dual-vector model predictive control system based on neural networks, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of any of the above-described methods.

[0058] In summary, this invention introduces a neural network model into the model predictive control of a doubly-fed induction generator (DFIG), proposing a model-free predictive control method. By collecting historical data from motor operation, the stator current, rotor current, slip angular velocity, and rotor voltage collected at time k are used as input data to the neural network model, while the rotor current collected at time k+1 is used as the output data. This allows for the training of a rotor current prediction model, which can directly predict the rotor current at the next time step. This avoids the deterioration of control performance caused by excessive system modeling complexity or parameter mismatch, significantly optimizing the dynamic response stability of the motor under non-ideal conditions and improving the robustness of motor control.

[0059] It should be noted that, for the sake of simplicity, the foregoing embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0060] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0061] In the several embodiments provided in this application, it should be understood that the disclosed methods or systems can be implemented in other ways. For example, the embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0062] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0063] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0064] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0065] Those skilled in the art will understand that all or part of the circuits in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0066] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of embodiments of this disclosure upon considering the specification and practicing the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

[0067] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0068] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A two-vector model predictive control method for a doubly-fed induction generator based on a neural network, characterized in that, The method includes: The stator current, rotor current, slip angular velocity, and rotor voltage of the doubly fed motor dual-vector model in the two-phase rotating coordinate system at the current moment are collected and input into the constructed rotor current prediction model to predict the rotor current at the next moment. The optimal vector is selected based on the predicted rotor current. The optimal vector and the zero vector are substituted into the value function to obtain the optimal duty cycle. The optimal vector and the optimal duty cycle are used to control the doubly fed motor dual-vector model.

2. The method for two-vector model predictive control of a doubly-fed induction generator based on a neural network according to claim 1, characterized in that, The method for constructing the rotor current prediction model includes: Record the data during the operation of the doubly-fed induction generator (DFIG) dual-vector model to generate a sample training set; wherein, the sample training set includes multiple sets of input data and output data; the input data and the output data correspond one-to-one, the input data are the stator current, rotor current, slip angular velocity and rotor voltage collected at time k during motor operation; the output data is the rotor current collected at time k+1 during motor operation; The generated sample training set is input into the neural network model for training, and the rotor current prediction model is obtained after convergence.

3. The method for two-vector model predictive control of a doubly-fed induction generator based on a neural network according to claim 1, characterized in that, The dual vectors are generated based on the switching states of the inverter bridge arms in the dual-vector model of the doubly fed motor; wherein the dual vectors include an effective vector and a zero vector; and the optimal vector is one of the vectors selected from the effective vectors.

4. The method for two-vector model predictive control of a doubly-fed induction generator based on a neural network according to claim 1, characterized in that, The value function for: ； in, , These represent the rotor current command values ​​in the two-phase rotating coordinate systems, respectively. , Let represent the rotor current at time k; , These represent the change in rotor current during one sampling period at time k.

5. The method for two-vector model predictive control of a doubly-fed induction generator based on a neural network according to claim 4, characterized in that, The change in rotor current over one sampling period is: ； in, Indicates the duty cycle of the effective vector; , This represents the change in rotor current within one sampling period of the effective vector; , This represents the change in rotor current within one sampling period of the zero vector.

6. The method for two-vector model predictive control of a doubly-fed induction generator based on a neural network according to claim 4, characterized in that, The optimal duty cycle is obtained in the following ways: By taking the derivative of the value function, the minimum value of the value function is obtained when the derivative is zero, and thus the corresponding duty cycle is obtained. If duty cycle If so, the duty cycle will be output directly; If duty cycle Then let , exist The minimum value in the interval is 1; If duty cycle Then let , exist The minimum value in the interval is 0.

7. The method for two-vector model predictive control of a doubly-fed induction generator based on a neural network according to claim 3, characterized in that, The correspondence between the switching states and the dual vectors is as follows: ； in, , , , , , It is an effective vector, a non-zero vector, and plays an actual control role in the doubly-fed motor dual-vector model; , It is a zero vector; , , This indicates the switching status of the three pairs of bridge arms of a three-phase two-level inverter; , , Representing voltage vectors respectively The three bits of the binary code.

8. A method for two-vector model predictive control of a doubly-fed induction generator based on a neural network according to claim 1 or 2, characterized in that, The loss function of the rotor current prediction model for: ； in, This indicates the number of data sets in the training set. Indicates the first The first class of data One data point; Indicates the first The first class of data One data point; , All are positive integers; 1 / 2 is a constant; This represents the rotor current prediction model. Indicates the weighting coefficient. This represents the bias coefficient.

9. The method for two-vector model predictive control of a doubly-fed induction generator based on a neural network according to claim 8, characterized in that, The weights and biases of the neural network are trained using gradient descent. The weighting coefficient and bias coefficient The update formula is: ； ； in, Indicates the learning rate; This indicates calculating the gradient; Represents the loss function; Indicates to Find the partial derivative.

10. A doubly-fed motor dual-vector model predictive control system based on neural networks, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 9.

Images