Inverter control method, device and system based on neural network predictor

Abstract

The present application discloses an inverter control method, device and system based on a neural network predictor, the method comprising: establishing a super-localized equation according to the three-phase grid-side voltage and the three-phase grid-side current of the inverter, determining the current function according to the super-localized equation; predicting the current function based on the neural network predictor to obtain a reference voltage vector corresponding to the current function; establishing an evaluation function about the three-phase grid-side current, determining the position of the reference voltage vector in the voltage vector partition, traversing a number of candidate voltage vectors in the voltage vector partition where the reference voltage vector is located, obtaining the reference voltage vector when the evaluation function is at the minimum value as a candidate; traversing all control inputs of the inverter, obtaining the control input when the evaluation function is at the minimum value, using it as the optimal choice for controlling the inverter, and adjusting the switch of the inverter. The robustness of the system can be improved while maintaining the excellent dynamic performance of the system.

Description

Technical Field

[0001] The present application relates to the technical field of power electronic converter control, and in particular to an inverter control method, device and system based on a neural network predictor. Background Art

[0002] In recent years, with the rapid development of the national economy and the upgrading of the industrial structure, the imbalance between the power demand of the distribution network and the design of the power grid has become increasingly apparent. The current distribution network faces many problems such as unreasonable structure and limited regulation means, which restricts its operational flexibility and reliability.

[0003] In the prior art, for three-level neutral point clamped (3L-NPC) inverters, the control of the inverter is generally achieved by establishing a predictive model. For example, model predictive control (MPC), as an advanced process control algorithm, is increasingly widely used in industrial processes due to its advantages such as easy modeling, superior control performance, and simple logical structure. With the latest technological advances in powerful modern microprocessors and control fields, the application scope of MPC methods in mechanical transmission and complex power conversion systems has been further broadened. Finite control set model predictive control (FCS-MPC) avoids the modulation process by directly enumerating all candidate options to select the required switching state, while taking into account the discreteness of the power converter. Since FCS-MPC can directly manipulate the converter switch combination, handle multiple constraints, and integrate different control variables, it has become the main solution for multivariable controlled power converter systems.

[0004] However, for the control methods that rely on prediction models, the control performance of the prediction models depends largely on the accuracy of the parameters and the precision of the models. In practical applications, unmodeled dynamic characteristics will inevitably appear, and the parameter values ​​of the physical system may be affected by factors such as the operating point, temperature, and external interference, which will affect the control performance and even cause damage to the terminal equipment. Summary of the invention

[0005] In order to address the deficiencies of the prior art, the purpose of the present application is to provide an inverter control method, device and system based on a neural network predictor, which can improve the robustness of the system while maintaining the excellent dynamic performance of the system.

[0006] To achieve the above objectives, this application adopts the following technical solutions:

[0007] In a first aspect, the present application provides an inverter control method based on a neural network predictor, the method comprising:

[0008] Obtain the three-phase grid-side voltage and three-phase grid-side current of the inverter;

[0009] Clarke transformation is performed on the three-phase grid-side voltage and the three-phase grid-side current respectively to obtain the transformed voltage component and current component, and a hyperlocal model of the inverter is established based on the voltage component and the current component, and a current function characterizing the current component is determined according to the hyperlocal model, wherein the current function changes with the inverter operating condition change;

[0010] Predicting the current function based on a neural network predictor to obtain a reference voltage vector corresponding to the current function;

[0011] An evaluation function for the three-phase grid-side current is established, and a position of a reference voltage vector in a voltage vector partition is determined according to a plurality of pre-established voltage vector partitions, wherein the voltage vector partition includes a plurality of candidate voltage vectors;

[0012] Traverse several candidate voltage vectors in the voltage vector partition where the reference voltage vector is located, obtain the reference voltage vector when the evaluation function is at the minimum value, use it as the optimal choice for controlling the inverter, and adjust the switch of the inverter.

[0013] Furthermore, the current function can be estimated using a neural network, and the current function satisfies the following relationship:

[0014] ; (1)

[0015] In formula (1), represents the pre-set weight, represents a known activation function, Represents the estimation error.

[0016] Furthermore, the current function is predicted based on the neural network predictor, including:

[0017] Establishing an estimated value of a weight matrix between a current component, a voltage component, and an inductance of the inverter, and multiplying the estimated value of the weight matrix by a nonlinear activation function of the current component to obtain first target data;

[0018] Obtaining the sum of the first target data and the voltage component, and dividing the summed data by the inductance of the inverter to obtain the second target data;

[0019] The rate of change of the current component is determined according to a preset threshold value, and the estimated value of the current component is determined according to the rate of change, the estimated value of the current component and the product of the error of the current component and the first fixed coefficient are obtained, and the difference between the second target data and the product is determined to determine the rate of change of the estimated value of the current component, and the rate of change of the estimated value of the current component satisfies the following relationship:

[0020] ; (2)

[0021] In formula (2), represents the rate of change of the estimated value of the current component, represents the estimated value of the current component, represents the transpose of the estimated value of the weight matrix, α is negatively correlated with the inductance of the inverter, Vg represents the voltage component, represents the estimated value of the current component and the error of the current component, and k1 represents the first fixed coefficient.

[0022] Furthermore, predicting the current function based on the neural network predictor also includes:

[0023] Obtain the error between the estimated value of the current component and the current component, and obtain the third target data according to the product of the error and the nonlinear activation function,

[0024] The product between the second fixed coefficient and the estimated value of the weight matrix is ​​obtained, and the change rate of the estimated value of the weight matrix is ​​determined based on the product, the third target data and the third fixed coefficient, and the change rate of the estimated value of the weight matrix satisfies the following relationship:

[0025] ; (3)

[0026] In formula (3), represents the rate of change of the estimated value of the weight matrix, represents the second fixed coefficient, represents the third fixed coefficient, represents the weight matrix.

[0027] Furthermore, based on the estimated value of the current component at the current frame sampling moment and the rate of change of the estimated value of the current component at the current frame sampling moment, and a sampling period, the estimated value of the current component at the next frame sampling moment is calculated.

[0028] Based on the estimated value of the weight matrix at the sampling moment of the current frame and the rate of change of the estimated value of the weight matrix at the sampling moment of the current frame, and a sampling period, calculate the estimated value of the weight matrix at the sampling moment of the next frame;

[0029] Among them, the estimated value of the current component and the estimated value of the weight matrix satisfy the following relationship:

[0030] ; (4)

[0031] In formula (4), k represents the sampling time, represents the estimated value of the current component at time k, represents the current component at time k, represents the estimated value of the weight matrix at time k, represents the weight matrix at time k, It represents the error between the estimated value of the current component at time k and the current component, and Ts is the sampling period.

[0032] Furthermore, the reference voltage vector satisfies the following relationship:

[0033] ; (5)

[0034] In formula (5), Represents the reference voltage vector, and the estimated value of the current component can be obtained by estimating the current component using a neural network.

[0035] In a second aspect, the present application further provides an inverter control device based on a neural network predictor, the control device comprising:

[0036] A first sampling module, used for sampling the three-phase grid-side voltage of the inverter;

[0037] The second sampling module is used to sample the three-phase grid-side current of the inverter;

[0038] A preprocessing module is used to perform Clarke transformation on the three-phase grid-side voltage and the three-phase grid-side current respectively to obtain the transformed voltage component and current component, and establish a hyperlocal model of the inverter based on the voltage component and the current component, and determine a current function characterizing the current component according to the hyperlocal model, wherein the current function changes with the inverter operating condition change;

[0039] A neural network predictor is used to predict the current function to obtain a reference voltage vector corresponding to the current function;

[0040] A control module is used to establish an evaluation function for the three-phase grid-side current, determine the position of the reference voltage vector in the voltage vector partition according to a plurality of pre-established voltage vector partitions, wherein the voltage vector partition includes a plurality of candidate voltage vectors, traverse the plurality of candidate voltage vectors in the voltage vector partition where the reference voltage vector is located, and obtain the reference voltage vector when the evaluation function is at a minimum value as a candidate;

[0041] The control module is also used to traverse all control inputs of the inverter, and solve the evaluation function according to the discrete mathematical model, obtain the control input when the evaluation function is at the minimum value, use it as the optimal choice for controlling the inverter, and adjust the switch of the inverter.

[0042] In a third aspect, the present application also provides an inverter control system based on a neural network predictor, the system comprising an inverter and an inverter control device based on a neural network predictor as in the second aspect.

[0043] In a fourth aspect, the present application also provides an electronic device comprising a memory for storing a computer program; and a processor for implementing the steps of any one of the inverter control methods based on a neural network predictor in the first aspect when executing the computer program.

[0044] In a fifth aspect, the present application also provides a computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the inverter control method based on a neural network predictor as described in any one of the first aspects are implemented.

[0045] The inverter control method based on the neural network predictor proposed in the present invention establishes a hyperlocal model according to the three-phase grid-side voltage and the three-phase grid-side current of the inverter, determines the current function according to the hyperlocal model; predicts the current function based on the neural network predictor to obtain a reference voltage vector corresponding to the current function; traverses several candidate voltage vectors in the voltage vector partition where the reference voltage vector is located, and obtains the reference voltage vector when the evaluation function is at the minimum value as a candidate; traverses all control inputs of the inverter, obtains the control input when the evaluation function is at the minimum value, uses it as the optimal choice for controlling the inverter, and adjusts the switch of the inverter. The robustness of the system can be improved while maintaining the excellent dynamic performance of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a system architecture diagram of an inverter control system based on a neural network predictor in an embodiment of the present application;

[0047] Figure 2 Schematic diagram of the circuit structure of the inverter in the embodiment of the present application;

[0048] Figure 3 This is a flow chart of an inverter control method based on a neural network predictor in an embodiment of the present application;

[0049] Figure 4 A schematic diagram of several voltage vector partitions pre-established in an embodiment of the present application;

[0050] Figure 5 A classification table of several voltage vectors to be selected in the voltage vector partitioning in the embodiment of the present application;

[0051] Figure 6 It is a waveform diagram of the three-phase current on the grid side in the prior art;

[0052] Figure 7 This is a waveform diagram of the grid-side three-phase current after being processed by the inverter control method based on the neural network predictor in the embodiment of the present application;

[0053] Figure 8is a comparison diagram of total harmonic distortion between the prior art and the embodiment of the present application under different inductance states;

[0054] Fig. 9 Schematic diagram of an inverter control device based on a neural network predictor in an embodiment of the present application;

[0055] Fig.10 Schematic diagram of the structure of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the specific implementation manner of the present application will be clearly and completely described below in conjunction with the drawings in the implementation manner of the present application.

[0057] like Figure 1 As shown, a system architecture diagram of an inverter control method based on a neural network predictor is provided in an embodiment of the present application. The system 100 includes an inverter 11 and an inverter control device 12 based on a neural network predictor. The control device 12 samples the working condition of the inverter 11 and adjusts the switch of the inverter 11 in real time according to the working condition of the inverter 11, so that the system 100 is in the optimal working state, that is, the grid-side three-phase current output by the inverter 11 tends to a steady state.

[0058] like Figure 2 As shown, in the embodiment of the present application, the inverter 11 is a three-level neutral point clamped (3L-NPC) inverter, which generates three levels by midpoint clamping and series DC capacitors. In order to obtain three levels, only two adjacent sets of switches (IGBT / Diode) of each phase can be turned on at the same time, and it is necessary to avoid the phenomenon of multiple switches being turned on at the same time, otherwise a short circuit will occur and burn out the components.

[0059] like Figure 3 As shown, the inverter control method based on the neural network predictor provided in the embodiment of the present application includes the following steps:

[0060] Step S100: obtaining the three-phase grid-side voltage and the three-phase grid-side current of the inverter.

[0061] For example, the three-phase grid-side voltage V of the inverter is collected by a voltage sensor and a current sensor respectively. x The three-phase grid-side current i x , is the current of each term of the three-phase current i, It is the voltage of each term of the three-phase voltage V.

[0062] The voltage sensor and the current sensor can be sensors in the control device; the voltage sensor and the current sensor can also be sensors independent of the control device, which receive the three-phase grid-side voltage and the three-phase grid-side current and then send them to the control device. This application does not limit the arrangement of the above sensors.

[0063] Step S200: Clarke transform is performed on the three-phase grid-side voltage and the three-phase grid-side current respectively to obtain the transformed voltage component and current component, and a hyperlocal model of the inverter is established based on the voltage component and the current component, and a current function characterizing the current component is determined according to the hyperlocal model, wherein the current function changes with the change of the inverter operating condition.

[0064] For example, is the value of the three-phase grid-side current ix after Clark transformation, is the value of the three-phase grid-side voltage Vx after Clarke transformation. Among them, the value of the three-phase grid-side current ix after Clarke transformation is the current component of the grid-side three-phase current , the value of the three-phase grid-side voltage Vx after Clark transformation is the voltage component of the three-phase voltage on the grid side .

[0065] The hyperlocal model of the three-phase grid-side voltage and three-phase grid-side current output by the inverter after Clarke change can be established as:

[0066] ;

[0067] In the formula, , , L represents the inductance connected to the inverter.

[0068] In the embodiment of the present application, the current function can be estimated by a neural network, and the current function satisfies the following relationship:

[0069] ; (1)

[0070] In the formula, represents the pre-set weight, represents a known activation function, represents the estimation error. The weight of the ideal neural network The range of is bounded by a known constant, such as ( is to find the norm of the formula), c is a positive real number.

[0071] Step S300: predicting the current function based on the neural network predictor to obtain a reference voltage vector corresponding to the current function.

[0072] Exemplarily, the control device determines the estimated value of the current component based on the current component, establishes an activation function for the current component, and obtains the estimated value of the weight matrix according to the weight of the activation function set in advance. After obtaining the estimated value of the weight matrix, the control device transposes the matrix of the estimated value of the weight matrix, calculates the estimated value of the current component using the estimated value of the transposed weight matrix and the activation function, and establishes a neural network predictor.

[0073] Specifically, predicting the current function based on a neural network predictor includes the following steps:

[0074] Establishing an estimated value of a weight matrix between a current component, a voltage component, and an inductance of the inverter, and multiplying the estimated value of the weight matrix by a nonlinear activation function of the current component to obtain first target data;

[0075] Obtaining the sum of the first target data and the voltage component, and dividing the summed data by the inductance of the inverter to obtain the second target data;

[0076] The rate of change of the current component is determined according to a preset threshold value, and the estimated value of the current component is determined according to the rate of change, the estimated value of the current component and the product of the error of the current component and the first fixed coefficient are obtained, and the difference between the second target data and the product is determined to determine the rate of change of the estimated value of the current component, and the rate of change of the estimated value of the current component satisfies the following relationship:

[0077] ; (2)

[0078] In formula (2), represents the rate of change of the estimated value of the current component, represents the estimated value of the current component, represents the transpose of the estimated value of the weight matrix, α is negatively correlated with the inductance of the inverter, Vg represents the voltage component, represents the estimated value of the current component and the error of the current component, and k1 represents the first fixed coefficient.

[0079] It should be noted that the estimated value of the current component will be updated with different sampling times.

[0080] Furthermore, when predicting the current function based on the neural network predictor, since the estimated value of the weight matrix is ​​updated at different sampling times, in order to determine the estimated value of the updated weight matrix, it is also necessary to determine the rate of change of the estimated value of the weight matrix, which specifically includes the following steps:

[0081] Obtaining an error between an estimated value of the current component and the current component, and obtaining third target data according to a product of the error and a nonlinear activation function;

[0082] The product between the second fixed coefficient and the estimated value of the weight matrix is ​​obtained, and the change rate of the estimated value of the weight matrix is ​​determined based on the product, the third target data and the third fixed coefficient, and the change rate of the estimated value of the weight matrix satisfies the following relationship:

[0083] ; (3)

[0084] In formula (3), represents the rate of change of the estimated value of the weight matrix, represents the second fixed coefficient, represents the third fixed coefficient, represents the weight matrix.

[0085] Exemplarily, the estimated value of the current component at the sampling moment of the current frame is calculated based on the estimated value of the current component at the sampling moment of the current frame and the rate of change of the estimated value of the current component at the sampling moment of the current frame, and a sampling period.

[0086] Based on the estimated value of the weight matrix at the sampling moment of the current frame and the rate of change of the estimated value of the weight matrix at the sampling moment of the current frame, and a sampling period, calculate the estimated value of the weight matrix at the sampling moment of the next frame;

[0087] Among them, the estimated value of the current component and the estimated value of the weight matrix satisfy the following relationship:

[0088] ; (4)

[0089] In formula (4), k represents the sampling time, represents the estimated value of the current component at time k, represents the current component at time k, represents the estimated value of the weight matrix at time k, represents the weight matrix at time k, represents the error between the estimated value of the current component at time k and the current component, and Ts is the sampling period.

[0090] In some embodiments, λ is set to 100, k1 is set to 5000, and k w Set to 0.01.

[0091] According to the above method, based on the discrete mathematical model, the voltage component is first-order discretized to obtain a reference voltage vector, which satisfies the following relationship:

[0092] ; (5)

[0093] In formula (5), Representing the reference voltage vector, the estimated value of the current component can be obtained by estimating the current component using a neural network.

[0094] Step S400: Establish an evaluation function for the three-phase grid-side current, determine the position of the reference voltage vector in the voltage vector partition according to several pre-established voltage vector partitions, wherein the voltage vector partition includes several candidate voltage vectors, traverse the several candidate voltage vectors in the voltage vector partition where the reference voltage vector is located, and obtain the reference voltage vector that makes the evaluation function at the minimum value as a candidate.

[0095] Step S500: traverse all control inputs of the inverter, and solve the evaluation function according to the discrete mathematical model to obtain the control input when the evaluation function is at the minimum value, use it as the optimal choice for controlling the inverter, and adjust the switch of the inverter.

[0096] For example, in order to reduce the computational burden of the processor and improve the operating efficiency, the reference voltage vector After that, the next step is to determine the voltage vector partition. Figure 4 In the embodiment of the present application, the number of voltage vector partitions is 6. Each voltage vector partition includes 10 voltage vectors to be selected (see Figure 5 ).

[0097] In order to find the optimal control input, it is necessary to traverse the 10 candidate voltage vectors in the voltage vector partition where the reference voltage vector is located and select the candidate that minimizes the evaluation function. After the evaluation function is established, all control inputs of the inverter are traversed according to the discrete mathematical model, and the control device calculates the evaluation function value corresponding to each control input. Then, the control input that minimizes the evaluation function is selected as the optimal choice. Next, the optimal switching state is applied to the entire inverter. This method has broad application prospects in dealing with systems with uncertainty and interference. Its main advantage is that it can optimize control performance while ensuring the robustness and stability of the system when there is uncertainty in the model and external environment.

[0098] See also Figures 6 to 8 ,in, Figure 6 This is a diagram showing the effect of optimizing the inverter performance by processing the grid-side three-phase current, grid-side three-phase voltage and inductance of the inverter in the prior art. The three curves represent the grid-side three-phase current output by the inverter. The waveform of each phase current fluctuates greatly and has a high total harmonic distortion (see Figure 8 ). Figure 7 The inverter control method based on the neural network predictor provided in the embodiment of the present application is used to achieve the effect diagram of inverter performance optimization. The fluctuation of the waveform of each phase current is small and the total harmonic distortion is low.

[0099] For example, in a motor system, robust model predictive control (MPC) can ensure that the motor maintains stable output power and achieves precise speed and position control when the load or system parameters change. For common friction changes, external disturbances, and system internal uncertainties in motor systems, robust model predictive control (MPC) can optimize control output, reduce the impact of disturbances on motor operation, and improve the dynamic response of the system.

[0100] like Fig. 9 As shown, the embodiment of the present application further provides an inverter control device 12 based on a neural network predictor, and the control device 12 includes:

[0101] A first sampling module 121, used for sampling the three-phase grid-side voltage of the inverter;

[0102] The second sampling module 122 is used to sample the three-phase grid-side current of the inverter;

[0103] A preprocessing module 123 is used to perform Clarke transformation on the three-phase grid-side voltage and the three-phase grid-side current respectively to obtain the transformed voltage component and current component, and establish a hyperlocal model of the inverter based on the voltage component and the current component, and determine a current function characterizing the current component according to the hyperlocal model, wherein the current function changes with the inverter operating condition change;

[0104] A neural network predictor 124, for predicting the current function to obtain a reference voltage vector corresponding to the current function;

[0105] The control module 125 is used to establish an evaluation function for the three-phase grid-side current, determine the position of the reference voltage vector in the voltage vector partition according to a plurality of pre-established voltage vector partitions, wherein the voltage vector partition includes a plurality of candidate voltage vectors, traverse the plurality of candidate voltage vectors in the voltage vector partition where the reference voltage vector is located, and obtain the reference voltage vector when the evaluation function is at a minimum value as a candidate;

[0106] The control module 125 is also used to traverse all control inputs of the inverter, and solve the evaluation function according to the discrete mathematical model, obtain the control input when the evaluation function is at the minimum value, use it as the optimal choice for controlling the inverter, and adjust the switch of the inverter.

[0107] Please refer to Fig. 9 , Fig. 9 The present invention provides a schematic diagram of the structure of an electronic device 200, the electronic device 200 comprising:

[0108] A memory 21, used for storing computer programs;

[0109] The processor 22 is used to implement the steps of the aforementioned method for voltage equalization of series-connected devices.

[0110] Among them, the processor 22 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 22 can be implemented in at least one hardware form of DSP (Digital Signal Processor), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 22 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a central processing unit; the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 22 may integrate a GPU (graphics processing unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 22 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0111] The memory 21 may include one or more computer-readable storage media, which may be non-transitory. The memory 21 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In this embodiment, the memory 21 is at least used to store the following computer program, wherein, after the computer program is loaded and executed by the processor, it can implement the relevant steps of the inverter control method based on the neural network predictor disclosed in any of the aforementioned embodiments. In addition, the resources stored in the memory 21 may also include an operating system and data, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system may include Windows, Unix, Linux, etc. The data may include, but is not limited to, data of the voltage equalization method of series devices, etc.

[0112] In some embodiments, the electronic device 200 may further include a display screen, an input / output interface, a communication interface, a power supply, and a communication bus.

[0113] It can be understood by those skilled in the art that Fig.10 The structure shown in the figure does not constitute a limitation on the electronic device, and may include more or fewer components than shown in the figure.

[0114] For an introduction to an electronic device provided by the present invention, please refer to the above-mentioned embodiment of the inverter control method based on the neural network predictor, and the present invention will not be repeated here.

[0115] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the inverter control method based on a neural network predictor as described above are implemented.

[0116] It is understandable that if the method in the above embodiment is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application is essentially 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, and the computer software product is stored in a storage medium to execute all or part of the steps of the method described in each embodiment of the present application. Specifically, the computer-readable storage medium may include but is not limited to any type of disk, including a floppy disk, an optical disk, and a mobile hard disk, or any type of medium or device suitable for storing instructions, data, etc., and the present application does not make any special restrictions here.

[0117] For an introduction to a computer-readable storage medium provided by the present invention, please refer to the above-mentioned embodiment of the inverter control method based on the neural network predictor, and the present invention will not be repeated here.

[0118] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0119] It should also be noted that, in this specification, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0120] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0121] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0122] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the claims attached to this application.

Claims

1. An inverter control method based on a neural network predictor, characterized in that: include: Obtain the three-phase grid-side voltage and three-phase grid-side current of the inverter; Clarke transformation is performed on the three-phase grid-side voltage and the three-phase grid-side current respectively to obtain a transformed voltage component and a current component, and a hyperlocal model of the inverter is established based on the voltage component and the current component, and a current function characterizing the current component is determined according to the hyperlocal model, wherein the current function changes with the inverter operating condition change, the current function is obtained by estimating using a neural network, and the current function satisfies the following relationship: （1）； In formula (1), represents the pre-set weight matrix, represents a known activation function, represents the estimation error; The current function is predicted based on the neural network predictor to obtain a reference voltage vector corresponding to the current function, wherein the prediction process includes: Establish an estimated value of a weight matrix between the current component, the voltage component and the inductance of the inverter, and multiply the estimated value of the weight matrix by the nonlinear activation function of the current component to obtain a first target data; obtain the sum of the first target data and the voltage component, and divide the summed data by the inductance of the inverter to obtain a second target data; determine the rate of change of the current component according to a preset threshold, and determine the estimated value of the current component according to the rate of change, obtain the product of the estimated value of the current component and the error of the current component and a first fixed coefficient, determine the difference between the second target data and the product, and determine the rate of change of the estimated value of the current component, and the rate of change of the estimated value of the current component satisfies the following relationship: ；（2） In formula (2), represents the rate of change of the estimated value of the current component, represents the estimated value of the current component, represents the transpose of the estimated value of the weight matrix, α is negatively correlated with the inductance of the inverter, Vg represents the voltage component, represents the error between the estimated value of the current component and the current component, k1 represents the first fixed coefficient; Establishing an evaluation function for the three-phase grid-side current, and determining the position of the reference voltage vector in the voltage vector partition according to a plurality of pre-established voltage vector partitions, wherein the voltage vector partition further includes a plurality of voltage vectors to be selected; The plurality of candidate voltage vectors in the voltage vector partition where the reference voltage vector is located are traversed to obtain the reference voltage vector when the evaluation function is at a minimum value, and the reference voltage vector is used as the optimal choice for controlling the inverter, and the switch of the inverter is adjusted.

2. The inverter control method based on the neural network predictor according to claim 1, characterized in that: The predicting of the current function based on the neural network predictor further includes: Obtaining an error between an estimated value of the current component and the current component, and obtaining third target data according to a product of the error and a nonlinear activation function; The product between the second fixed coefficient and the estimated value of the weight matrix is ​​obtained, and the change rate of the estimated value of the weight matrix is ​​determined based on the product, the third target data and the third fixed coefficient, and the change rate of the estimated value of the weight matrix satisfies the following relationship: ；（3） In formula (3), represents the rate of change of the estimated value of the weight matrix, represents the second fixed coefficient, represents the third fixed coefficient, represents the weight matrix.

3. The inverter control method based on the neural network predictor according to claim 2, characterized in that: Calculate the estimated value of the current component at the next frame sampling moment based on the estimated value of the current component at the current frame sampling moment and the rate of change of the estimated value of the current component at the current frame sampling moment, and a sampling period; Calculate the estimated value of the weight matrix at the next frame sampling time based on the estimated value of the weight matrix at the current frame sampling time and the rate of change of the estimated value of the weight matrix at the current frame sampling time, and the sampling period; The estimated value of the current component and the estimated value of the weight matrix satisfy the following relationship: ；（4） In formula (4), k represents the sampling time, represents the estimated value of the current component at time k, represents the current component at time k, represents the estimated value of the weight matrix at time k, represents the weight matrix at time k, represents the error between the estimated value of the current component at time k and the current component, and Ts is the sampling period.

4. The inverter control method based on a neural network predictor according to claim 1, characterized in that: The reference voltage vector satisfies the following relationship: ；（5） In formula (5), represents the reference voltage vector, and the estimated value of the current component is obtained by estimating the current component using a neural network.

5. An inverter control device based on a neural network predictor, characterized in that: include: A first sampling module, used for sampling the three-phase grid-side voltage of the inverter; A second sampling module, used for sampling the three-phase grid-side current of the inverter; A preprocessing module, used to perform Clarke transformation on the three-phase grid-side voltage and the three-phase grid-side current respectively to obtain a transformed voltage component and a current component, and establish a hyperlocal model of the inverter based on the voltage component and the current component, and determine a current function characterizing the current component according to the hyperlocal model, wherein the current function changes with the inverter operating condition change, the current function is obtained by estimating using a neural network, and the current function satisfies the following relationship: （1）； In formula (1), represents the pre-set weight matrix, represents a known activation function, represents the estimation error; A neural network predictor is used to predict the current function to obtain a reference voltage vector corresponding to the current function, wherein the prediction process includes: establishing an estimated value of a weight matrix between the current component, the voltage component and the inductance of the inverter, and multiplying the estimated value of the weight matrix by the nonlinear activation function of the current component to obtain a first target data; obtaining the sum of the first target data and the voltage component, and dividing the summed data by the inductance of the inverter to obtain a second target data; determining a change rate of the current component according to a preset threshold, and determining an estimated value of the current component according to the change rate, obtaining the product between the estimated value of the current component and the error of the current component and a first fixed coefficient, and determining the difference between the second target data and the product to determine the change rate of the estimated value of the current component, and the change rate of the estimated value of the current component satisfies the following relationship: ；（2） In formula (2), represents the rate of change of the estimated value of the current component, represents the estimated value of the current component, represents the transpose of the estimated value of the weight matrix, α is negatively correlated with the inductance of the inverter, Vg represents the voltage component, represents the error between the estimated value of the current component and the current component, k1 represents the first fixed coefficient; A control module, configured to establish an evaluation function for the three-phase grid-side current, determine the position of the reference voltage vector in the voltage vector partition according to a plurality of pre-established voltage vector partitions, wherein the voltage vector partition further includes a plurality of candidate voltage vectors, traverse the plurality of candidate voltage vectors in the voltage vector partition where the reference voltage vector is located, and obtain the reference voltage vector when the evaluation function is at a minimum value as a candidate; The control module is also used to traverse all control inputs of the inverter, and solve the evaluation function according to the discrete mathematical model, obtain the control input when the evaluation function is at the minimum value, use it as the optimal choice for controlling the inverter, and adjust the switch of the inverter.

6. An inverter control system based on a neural network predictor, characterized in that: It comprises an inverter and an inverter control device based on a neural network predictor as claimed in claim 5.

7. An electronic device, characterized in that: comprising a memory for storing a computer program; A processor, configured to implement the steps of the inverter control method based on a neural network predictor as described in any one of claims 1 to 4 when executing the computer program.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the inverter control method based on a neural network predictor as described in any one of claims 1 to 4 are implemented.

Images