Gray box impedance identification method and device for grid-connected inverters adapting to multiple operating points

Abstract

The present application provides a method and device for identifying the gray box impedance of a grid-connected inverter that is adaptable to multiple operating points, including: determining an analytical expression of the sequence impedance of the grid-connected inverter considering the frequency coupling characteristics based on a control scheme including dual-loop control; and extracting a first nonlinear functional relationship and a second nonlinear functional relationship; using a feedforward neural network to replace the first nonlinear functional relationship and the second nonlinear functional relationship, respectively, to construct a first neural network and a second neural network; injecting a voltage disturbance into the grid-connected point to obtain a first data set and a second data set; training the first neural network and the second neural network based on the first data set and the second data set, respectively, to obtain a gray box impedance model of the grid-connected inverter that is adaptable to multiple operating points, and calculating the positive sequence impedance amplitude and phase angle of the grid-connected inverter considering the frequency coupling characteristics. The method can improve the accuracy of neural network model training under a limited amount of data, and realize accurate identification of the impedance of the grid-connected inverter under multiple operating points.

Description

Technical Field

[0001] The present application belongs to the technical field of new energy grid-connected systems, and in particular, relates to a method and device for identifying gray box impedance of a grid-connected inverter that is adaptable to multiple working points. Background Art

[0002] The proportion of new energy represented by wind turbines and photovoltaics in the power grid is increasing year by year. Although the large-scale access of distributed new energy has achieved the goals of energy conservation, emission reduction and green environmental protection, it has also made the dynamic characteristics of the distribution network more complicated. As the main interface device connecting distributed new energy power generation and the public power grid, the grid-connected inverter not only transmits the electric energy generated by renewable energy to the power grid, but also undertakes the important mission of stable operation of the distributed power generation system. In different power generation systems, each grid-connected inverter is mostly designed independently, and its parameters such as capacity and equivalent output impedance are different. The interactive coupling between the inverters of new energy power generation equipment and between the inverters and the power grid can easily cause a series of broadband oscillation problems, which seriously affects the reliable power supply of new energy and the safe and stable operation of the power system. Impedance analysis is the main means to study the stability of the interactive system between the grid-connected inverter of new energy and the power grid. In order to further analyze the frequency domain stability of the grid-connected system of new energy and improve the power quality of the power system containing new energy, it is very important to obtain accurate and reliable grid-connected inverter impedance.

[0003] In actual industrial applications, the working conditions of grid-connected inverters may vary over a wide range. The "white box" analytical model of the grid-connected inverter ignores the dead zone effect and non-ideal characteristics, so the model accuracy is difficult to meet the requirements. In addition, due to technical confidentiality, the "white box" impedance modeling method is no longer applicable because the parameters are not completely known. The impedance value of the grid-connected inverter system can be obtained by the impedance measurement method, but since the operating point of the grid-connected inverter often changes, it is difficult to perform online real-time impedance measurement of the grid-connected inverter across multiple operating points, making it difficult to obtain the complete electrical dynamic characteristics of the grid-connected inverter. "Black box" methods such as neural networks can obtain the broadband impedance characteristics of the grid-connected inverter based on the port current and voltage parameters, and have advantages in multi-operating impedance identification, but the training of the neural network model relies on a large amount of data, and the generated model is difficult to intuitively explain its physical meaning.

[0004] Through investigation, it was found that new energy grid-connected inverters basically adopt a dual-loop control structure. In this context, the present invention combines the control mechanism of the grid-connected inverter with measurable data, and proposes a gray box impedance modeling method that uses the typical control structure of the grid-connected inverter to guide the construction of the neural network model. This method improves the accuracy of neural network model training under limited data volume and realizes accurate identification of the impedance of the grid-connected inverter under multiple operating points. Summary of the invention

[0005] The main purpose of the embodiments of the present invention is to provide a gray box impedance identification method and device for a grid-connected inverter that is adaptable to multiple operating points, which can improve the accuracy of neural network model training with a limited amount of data and realize accurate identification of the impedance of the grid-connected inverter under multiple operating points.

[0006] In a first aspect, a method for identifying gray-box impedance of a grid-connected inverter that is adaptable to multiple operating points is provided, the method comprising: determining an analytical expression of sequence impedance of the grid-connected inverter that takes frequency coupling characteristics into consideration based on a control scheme including dual-loop control; extracting a first nonlinear functional relationship in the frequency domain expression of the output port voltage of any phase of the grid-connected inverter and a second nonlinear functional relationship that takes frequency coupling characteristics into consideration in the impedance analytical model of the grid-connected inverter based on the analytical expression of sequence impedance; mining and replacing the first nonlinear functional relationship using a feedforward neural network to construct a first neural network and a corresponding loss function; mining and replacing the second nonlinear functional relationship using a feedforward neural network to construct a second neural network and a corresponding loss function; injecting voltage disturbances into the grid-connected point of a gray-box grid-connected inverter simulation model that adopts a dual-loop control structure but has unknown control parameters and performing data acquisition to obtain a first data set and a second data set that are required for training the first neural network and the second neural network respectively, having N groups of data; training the first neural network and the second neural network respectively based on the first data set and the second data set to obtain a gray-box impedance model of the grid-connected inverter that is adaptable to multiple operating points, and calculating the positive-sequence impedance amplitude and phase angle of the grid-connected inverter that takes frequency coupling characteristics into consideration.

[0007] In a possible implementation, extracting a first nonlinear functional relationship of a frequency domain expression for characterizing the voltage of any phase output port of the grid-connected inverter and a second nonlinear functional relationship for characterizing the admittance matrix of the grid-connected inverter considering the frequency coupling characteristics in the impedance analytical model of the grid-connected inverter based on the sequence impedance analytical expression includes: extracting a frequency domain equation for the voltage of any phase output port of the grid-connected inverter based on the sequence impedance analytical expression:

[0008]

[0009] To apply positive sequence disturbance frequency Rear grid-connected inverter port The frequency components of the phase voltage, To apply positive sequence disturbance frequency Rear grid-connected inverter port The frequency component of the coupled component of the phase voltage, , , represents the imaginary unit, is the fundamental frequency, is the coupling frequency component, is the PWM modulation coefficient, , The sampling delay is simulated. is the sampling period, , They are respectively the grid-connected inverter in the stationary coordinate system Positive sequence disturbance frequency of phase modulated signal Negative sequence coupling frequency The frequency component at the output port of any phase of the grid-connected inverter is simplified to obtain the first nonlinear functional relationship; the main circuit equation and the admittance matrix model of any phase in the frequency domain are extracted based on the sequence impedance analytical expression; the main circuit equation, the first nonlinear functional relationship and the admittance matrix model are combined to perform matrix calculation to obtain the admittance matrix representation of the grid-connected inverter considering the frequency coupling characteristics; the admittance matrix representation is simplified to obtain the second nonlinear functional relationship.

[0010] In another possible implementation, a feedforward neural network is used to mine and replace the first nonlinear functional relationship, and a first neural network and a corresponding loss function are constructed, including: the first nonlinear functional relationship is

[0011]

[0012] In the formula To apply positive sequence disturbance frequency Rear grid-connected inverter port Phase voltage at positive sequence disturbance frequency The frequency component at To apply positive sequence disturbance frequency Rear grid-connected inverter port The coupling component of the phase voltage is at the negative sequence coupling frequency The frequency component at ~ is with , , , , The related nonlinear function, ~ is with , , , , The related nonlinear function, , and are the fundamental frequency, positive sequence disturbance frequency and negative sequence coupling frequency respectively, , , , , fundamental frequency current amplitude , fundamental frequency voltage amplitude , Fundamental current initial phase angle represents the working point, and They are the positive sequence frequency disturbance voltage vector and the negative sequence coupling response voltage vector respectively; and are respectively the positive sequence frequency disturbance current vector and the negative sequence coupling response current vector; a feedforward neural network is used to replace the first nonlinear function relationship, and a hidden layer is applied to capture the mapping characteristics between frequency, frequency domain offset, working point and nonlinear function expression to construct a first neural network, which includes a hidden layer, an input layer and an output layer; the transfer function of the output voltage of any phase of the grid-connected inverter port is expressed as the form of division of two polynomials, and the number of neurons in the hidden layer is determined according to the highest power of the denominator polynomial and the numerator polynomial; the output port voltage of any phase of the grid-connected inverter is related to , , , , , The voltage of any phase output port of the grid-connected inverter in the expected frequency band is calculated according to the first nonlinear function relationship, and the voltage of any phase output port of the grid-connected inverter in the expected frequency band is selected according to the mean square error between the calculated value and the measured value. and The average value of the mean square error in is used as the loss function for the first neural network training.

[0013] In another possible implementation, a feedforward neural network is used to mine and replace the second nonlinear functional relationship, and a second neural network and a corresponding loss function are constructed, including: the second nonlinear functional relationship is

[0014]

[0015] In the formula ~ is the first nonlinear function ~ , ~ and A nonlinear function composed of: using a feedforward neural network to replace the second nonlinear functional relationship, applying a hidden layer to capture the mapping relationship characteristics of the second nonlinear functional relationship, and constructing a second neural network, wherein the output of the second neural network is the base frequency current amplitude , fundamental frequency voltage amplitude The elements of the broadband admittance matrix of the grid-connected inverter at the working point are obtained by simple programming based on the analytical expression of the sequence impedance. According to the mean square error between the calculated value and the measured value, the average value of the mean square error in the amplitude and phase angle of the positive sequence impedance is selected as the loss function during the training of the second neural network.

[0016] In another possible implementation, a voltage disturbance is injected into the grid connection point of a gray box grid-connected inverter simulation model that adopts a dual-loop control structure but has unknown control parameters, and data is collected to obtain a voltage disturbance with unknown control parameters. N The first data set and the second data set required for training the first neural network and the second neural network respectively include: determining the selection range of the working point of the grid-connected system of the grid-connected inverter according to the requirements of avoiding overload of the grid-connected inverter and satisfying the stable operation of the power system, the working point includes the fundamental frequency voltage amplitude, fundamental frequency current and its initial phase angle at the PCC point; selecting within the selection range N working points; a positive-sequence disturbance voltage with a frequency equal to the positive-sequence disturbance frequency is injected into the PCC point of the grid-connected inverter, the positive-sequence current responded by the output port is measured, and the ratio of the positive-sequence disturbance voltage to the positive-sequence current is calculated to obtain the impedance of the grid-connected inverter at the positive-sequence disturbance frequency; a voltage disturbance including multiple frequencies is injected into the PCC point of a gray-box grid-connected inverter simulation model that adopts a dual-loop control structure but has unknown control parameters, a frequency scan is performed within the expected frequency band, and a first data set and a second data set required for training the first neural network and the second neural network are collected.

[0017] In another possible implementation, at each working point, the first data set includes: input data of the first neural network: m disturbance frequencies , m disturbance frequency offsets and , positive sequence frequency disturbance current vector at m disturbance frequencies , voltage vector and negative sequence coupling response current vector , response voltage vector , including the fundamental frequency current amplitude , fundamental frequency current phase angle and fundamental frequency voltage amplitude The three PCC point measurements and the label data of the first neural network: the m frequency components of any phase voltage of the grid-connected inverter port after filtering and m frequency components of the coupled component ; and write the three PCC point measurement values ​​as In the form of; at each working point, the second data set includes: the input data of the second neural network: m disturbance frequencies , the grid side impedance calculated at m disturbance frequencies, the m groups of coefficients of the first nonlinear relationship output by the second neural network and , and the label data of the second neural network: m positive sequence frequency disturbance currents measured by the PCC point ,Voltage Calculated positive sequence impedance amplitude of grid-connected inverter considering frequency coupling characteristics and phase angle .

[0018] In another possible implementation, the first neural network and the second neural network are trained based on the first data set and the second data set respectively to obtain a gray box impedance model of a grid-connected inverter that adapts to multiple working points, including: inputting the input data in the first data set into the first neural network for model training to obtain predicted label data output by the first neural network; calculating a first loss function based on the predicted label data and the label data in the first data set, and judging whether a first training end condition is met; if the first training end condition is not met, returning to the step of inputting the input data in the first data set into the first neural network for model training, otherwise outputting the coefficient of the first nonlinear relationship output by the first neural network; inputting the input data in the second data set into the second neural network for model training to obtain predicted label data output by the second neural network; calculating a second loss function based on the predicted label data and the label data in the second data set, and judging whether a second training end condition is met; if the second training end condition is not met, returning to the step of inputting the input data in the second data set into the second neural network for model training, otherwise obtaining a gray box impedance model of a grid-connected inverter that adapts to multiple working points.

[0019] In a second aspect, a gray box impedance identification device for a grid-connected inverter adapting to multiple working points is provided, comprising: a sequence impedance analysis unit, for determining a sequence impedance analytical expression of a grid-connected inverter considering frequency coupling characteristics based on a control scheme including dual-loop control; a nonlinear relationship determination unit, for extracting a first nonlinear functional relationship for characterizing a frequency domain expression of an output port voltage of any phase of the grid-connected inverter and a second nonlinear functional relationship for characterizing an admittance matrix considering frequency coupling characteristics of the grid-connected inverter in an impedance analytical model of the grid-connected inverter based on the sequence impedance analytical expression; a network construction unit, for mining and replacing the first nonlinear functional relationship using a feedforward neural network to construct a first neural network and a corresponding loss function; mining and replacing the second nonlinear functional relationship using a feedforward neural network to construct a second neural network and a corresponding loss function; a data collection unit, for injecting voltage disturbances into a grid-connected point of a gray box grid-connected inverter simulation model that adopts a dual-loop control structure but has unknown control parameters and performing data collection to obtain a voltage disturbance having a frequency domain expression of an output port voltage of any phase of the grid-connected inverter and a second nonlinear functional relationship to characterize an admittance matrix considering frequency coupling characteristics of the grid-connected inverter. NThe first and second data sets are respectively used for training the first neural network and the second neural network; the gray box impedance identification unit is used to train the first neural network and the second neural network based on the first and second data sets, respectively, to obtain a gray box impedance model of the grid-connected inverter that adapts to multiple working points, and to calculate the positive sequence impedance amplitude and phase angle of the grid-connected inverter considering the frequency coupling characteristics.

[0020] In a third aspect, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a gray box impedance identification method for a grid-connected inverter adaptable to multiple operating points as provided in the first aspect is implemented.

[0021] In a fourth aspect, a non-transitory computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a gray box impedance identification method for a grid-connected inverter that is adaptable to multiple operating points as provided in the first aspect is implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in describing the embodiments of the present application are briefly introduced below.

[0023] Figure 1 A schematic flow chart of a method for identifying gray box impedance of a grid-connected inverter adapted to multiple operating points provided by an embodiment of the present invention;

[0024] Figure 2 Schematic diagram of a grid-connected system structure and a corresponding grid-connected inverter dual-loop control structure in an embodiment of the present invention, wherein (a) is the grid-connected system structure, and (b) is the grid-connected inverter dual-loop control structure;

[0025] Figure 3 Schematic diagram of the dual-loop control process of the grid-connected inverter in an embodiment of the present invention;

[0026] Figure 4 A schematic diagram of data collection and use of a first data set and a second data set in an embodiment of the present invention;

[0027] Figure 5 Schematic diagram of the training process of the first neural network in an embodiment of the present invention;

[0028] Figure 6 is a schematic diagram of the training process of the second neural network in an embodiment of the present invention;

[0029] Figure 7 A schematic diagram of the process from constructing a gray box impedance model of a grid-connected inverter to performing impedance identification of multiple working points of the grid-connected inverter in an embodiment of the present invention;

[0030] Figure 8 A schematic structural diagram of a gray box impedance identification device for a grid-connected inverter adapted to multiple operating points provided by another embodiment of the present invention;

[0031] Fig. 9 A schematic diagram of the physical structure of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0032] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar modules or modules with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present invention.

[0033] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, modules and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, modules, components and / or groups thereof. It should be understood that when we refer to a module as being "connected" or "coupled" to another module, it may be directly connected or coupled to the other modules, or there may be intermediate modules. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any modules and all combinations of one or more associated listed items.

[0034] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation method of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0035] The technical solution of the present application and how to solve the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0036] like Figure 1 FIG. 1 is a flowchart of a gray box impedance identification method for a grid-connected inverter adapted to multiple operating points provided by an embodiment of the present invention. The method is applicable to a current-type grid-connected inverter adopting a dual-loop control structure. The grid-connected system structure and the corresponding grid-connected inverter dual-loop control structure are as follows: Figure 2As shown, (a) is the grid-connected system structure, and (b) is the grid-connected inverter dual-loop control structure. The grid-connected system structure includes a new energy power generation unit and a grid impedance. The new energy power generation unit includes a grid-connected inverter and a filter unit. The new energy power generation unit is connected to the grid impedance through a point of common coupling (PCC) and is connected to the large power grid through the grid impedance. The gray box impedance identification method of the grid-connected inverter that adapts to multiple working points of the embodiment of the present invention is applied to the grid-connected inverter dual-loop control structure. Figure 1 As shown, the gray box impedance identification method of a grid-connected inverter adapting to multiple working points according to an embodiment of the present invention includes:

[0037] Step S11, based on a control scheme including dual-loop control, determining an analytical expression of the sequence impedance of the grid-connected inverter taking into account the frequency coupling characteristics.

[0038] Specifically, the machine-side converter is ignored and based on the general control scheme including dual-loop control, an analytical expression for the sequence impedance of the grid-connected inverter considering the frequency coupling characteristics is derived.

[0039] Step S12, extracting a first nonlinear functional relationship in the frequency domain expression of the output port voltage of any phase of the grid-connected inverter and a second nonlinear functional relationship in the admittance matrix of the grid-connected inverter considering the frequency coupling characteristics in the impedance analytical model of the grid-connected inverter based on the sequence impedance analytical expression.

[0040] Step S13, using a feedforward neural network to mine and replace the first nonlinear function relationship, constructing a first neural network and a corresponding loss function; using a feedforward neural network to mine and replace the second nonlinear function relationship, constructing a second neural network and a corresponding loss function.

[0041] Step S14, injecting voltage disturbance into the grid connection point of the gray box grid-connected inverter simulation model which adopts the dual-loop control structure but the control parameters are unknown and performing data collection to obtain N The sets of data are used for training the first data set and the second data set required for the first neural network and the second neural network respectively.

[0042] Step S15, training the first neural network and the second neural network based on the first data set and the second data set respectively, obtaining a gray box impedance model of the grid-connected inverter that adapts to multiple working points, and calculating the positive sequence impedance amplitude and phase angle of the grid-connected inverter considering the frequency coupling characteristics.

[0043] The existing gray box impedance model of the grid-connected inverter only considers the wide-band impedance model under fixed working conditions. The gray box impedance identification method of the grid-connected inverter that adapts to multiple working points in the embodiment of the present invention combines a neural network with a strong nonlinear relationship fitting ability with the knowledge of the grid-connected inverter structure, and can obtain a sequence impedance model that adapts to multiple working points. It can improve the accuracy of neural network model training under a limited amount of data, and realize the accurate identification of the impedance of the grid-connected inverter under multiple working points. By relying on the advantages of the neural network in processing multi-dimensional data and mining data features, the constructed model can characterize the impedance transfer function of the grid-connected inverter within a wide working range under limited data conditions. It has strong interpretability and engineering use value, and provides an accurate model basis for the study of broadband oscillation problems of new energy grid connection.

[0044] In the embodiment of the present invention, the commonly used dual-loop control scheme is the grid voltage oriented vector control, and its specific control process is as follows: Figure 3 As shown, the grid voltage phase angle is followed by the phase-locked control link, the grid voltage phase angle obtained by phase locking is used as the reference phase angle of coordinate transformation, and the decoupling control of active power and reactive power is realized by controlling the dq axis current in the rotating coordinate system. Figure 3 middle is the reference rotation angle output by the grid-connected inverter phase-locked loop, and They are the grid-connected inverter grid-connected point current The d-axis and q-axis components in the synchronously rotating coordinate system, and They are the grid-connected inverter grid-connected point voltage The d-axis and q-axis components in the synchronously rotating coordinate system, , and is the modulation signal in the stationary coordinate system, , and is the output port voltage of the grid-connected inverter, , and They are the filter inductor, filter capacitor and damping resistor of the grid-connected inverter respectively.

[0045] In step S11, firstly, the frequency coupling characteristics of the grid-connected inverter of the new energy power generation device are considered, and a positive voltage small disturbance frequency is applied at the grid connection point. In addition to the positive sequence current response with the same frequency, the AC port of the grid-connected inverter also generates a frequency of The negative sequence current response. Phase as an example, at this time the voltage of the PCC point of the grid-connected inverter , Current The time domain expression is as follows:

[0046] (1)

[0047] In the formula , , and are the amplitudes of fundamental frequency voltage, positive sequence frequency disturbance voltage and negative sequence coupling voltage respectively; , and are the amplitudes of fundamental frequency current, positive sequence frequency disturbance current and negative sequence coupling current respectively; , and They are the fundamental frequency, positive sequence disturbance frequency and negative sequence coupling frequency respectively; and are the initial phase angles of the positive-sequence disturbance voltage and the negative-sequence disturbance voltage respectively; , and They are the initial phase angles of the fundamental current, positive-sequence disturbance current and negative-sequence disturbance current respectively.

[0048] Corresponding grid-connected inverter PCC point Phase voltage , Current The frequency domain expression is as follows:

[0049] (2)

[0050] The frequency components are in bold, and , , , , , .

[0051] Then, based on the control scheme including dual-loop control, the admittance matrix represented by the frequency components of voltage and current is obtained considering the frequency coupling characteristics. Under small disturbances, let , the admittance matrix of the grid-connected inverter considering the frequency coupling characteristics The frequency components of voltage and current can be expressed as follows:

[0052] (3)

[0053] Among them, the admittance matrix The diagonal elements of and It reflects the influence of voltage disturbance on the current components with the same frequency and phase sequence. The off-diagonal elements and It reflects the influence of voltage disturbance on the coupled current component.

[0054] based on Figure 2 The relationship between the output port voltage of the grid-connected inverter and the voltage and current at the PCC point is as follows:

[0055] (4)

[0056] , , They are Mutually, Mutually, Phase voltage of the phase.

[0057] Next, we model the control link:

[0058] The d-axis current reference value is regarded as a constant, and the frequency domain equation of the output port voltage of any phase of the grid-connected inverter and the main circuit equation in the frequency domain of any phase are obtained.

[0059] The output of the voltage outer loop will be used as the d-axis current reference value. Therefore, under the influence of the DC voltage disturbance component, the d-axis current reference value will also have a disturbance component. The expression of the d-axis current reference value in the frequency domain is as follows:

[0060] (5)

[0061] In the formula, is the d-axis current reference value of the current inner loop, is the input voltage, is the transfer function of the voltage outer loop PI regulator, is the DC voltage reference value of the voltage outer loop.

[0062] In stable operation, the voltage fluctuation on the DC side is very small, so when deriving the impedance analysis model, d The shaft current reference value is regarded as a constant. The relationship between the grid-connected inverter modulation signal and the PCC point voltage and current can be expressed as ：

[0063] (6)

[0064] in, is the transfer function of the current inner loop PI regulator, is the current inner loop proportional coefficient, is the voltage outer loop proportional coefficient. , and , The expression in the frequency domain is as follows:

[0065] (7)

[0066] (8)

[0067] (9)

[0068] (10)

[0069] In the formula is the initial phase angle of fundamental current, , , , is the phase-locked loop proportionality factor, is the phase-locked loop integral coefficient.

[0070] (11)

[0071] (12)

[0072] Where:

[0073]

[0074]

[0075] make , , After the inverse Park transformation, the grid-connected inverter is obtained in the stationary coordinate system. Frequency domain expression of phase modulated signal:

[0076] (13)

[0077] (14)

[0078] Combining equations (6) to (14) we can get the grid-connected inverter The frequency domain equation of the phase output port voltage can be expressed as follows:

[0079] (15)

[0080] In the formula To apply positive sequence disturbance frequency Rear grid-connected inverter port Positive sequence disturbance frequency of phase voltage The frequency component at To apply positive sequence disturbance frequency Rear grid-connected inverter port Negative sequence coupling frequency of the coupling component of the phase voltage The frequency component at , , represents the imaginary unit, is the fundamental frequency, is the coupling frequency component, is the PWM modulation coefficient, , The sampling delay is simulated. is the sampling period, , They are respectively the grid-connected inverter in the stationary coordinate system Positive sequence disturbance frequency of phase modulated signal Negative sequence coupling frequency The frequency component at .

[0081] According to the relationship between the output port voltage of the grid-connected inverter and the voltage and current of the PCC point and the transfer function of the output voltage of any phase of the grid-connected inverter port, the main circuit equation in the frequency domain of any phase is determined. Specifically, according to equation (4) and equation (23), it can be obtained The main circuit equation in the phase-frequency domain is:

[0082] (16)

[0083] In the formula .

[0084] According to the admittance matrix and the main circuit equation, the admittance matrix model of the grid-connected inverter considering the frequency coupling characteristics is obtained by matrix calculation. Specifically, according to the matrix calculation of equations (3) and (16), the admittance matrix model of the grid-connected inverter considering the frequency coupling characteristics can be obtained:

[0085] (17)

[0086] Specifically, in order to facilitate impedance measurement and verification, the grid impedance can be used Eliminate the frequency coupling component and convert the coupling term in the admittance matrix model into the non-coupling term, so that the single-input multi-output system is simplified to a single-input single-output system. The conversion process can be expressed by the following formula:

[0087] (18)

[0088] Among them, the grid impedance Formula (18) can be regarded as an analytical expression of the positive-sequence impedance model of the grid-connected inverter after equivalent conversion considering the frequency coupling characteristics.

[0089] Based on the conversion relationship between positive-sequence impedance and negative-sequence impedance, the negative-sequence impedance model of the grid-connected inverter considering the frequency coupling characteristics can be obtained:

[0090] (19)

[0091] In step S12, the frequency domain equation of the output port voltage of any phase of the grid-connected inverter is extracted based on the sequence impedance analytical expression, such as formula (15); the frequency domain equation of the output port voltage of any phase of the grid-connected inverter is simplified to obtain the first nonlinear functional relationship. Specifically, by observing formula (15), the grid-connected inverter The frequency domain expression of the phase output port voltage is simplified to the following formula:

[0092] (20)

[0093] In the formula To apply positive sequence disturbance frequency Rear grid-connected inverter port Phase voltage at positive sequence disturbance frequency The frequency component at For grid-connected inverter port The coupling component of the phase voltage is at the negative sequence coupling frequency The frequency component at ~ is with , , , , The related nonlinear function, ~ is with , , , , The related nonlinear function, , and are the fundamental frequency, positive sequence disturbance frequency and negative sequence coupling frequency respectively, , , , , fundamental frequency current amplitude , fundamental frequency voltage amplitude , Fundamental current initial phase angle represents the working point, and They are the positive sequence frequency disturbance voltage vector and the negative sequence coupling response voltage vector respectively; and are respectively the positive sequence frequency disturbance current vector and the negative sequence coupling response current vector. Thus, the first nonlinear function relationship is obtained.

[0094] The frequency domain equation of the output port voltage of any phase of the grid-connected inverter is simplified to obtain the first nonlinear functional relationship; the main circuit equation and the admittance matrix model of any phase in the frequency domain are extracted based on the sequence impedance analytical expression; the main circuit equation, the first nonlinear functional relationship and the admittance matrix model are combined for matrix calculation to obtain the admittance matrix representation of the grid-connected inverter considering the frequency coupling characteristics. That is, according to equations (16), (17) and (20), the admittance matrix of the grid-connected inverter considering the frequency coupling characteristics can be expressed as follows after matrix calculation:

[0095] (twenty one)

[0096] The elements of the admittance matrix can be expressed as follows:

[0097] (twenty two)

[0098] In the formula ~ is the first nonlinear function ~ , ~ and The nonlinear function composed of . Simplifying the representation of the admittance matrix, we get the second nonlinear function relationship.

[0099] It can be seen from formula (21) that if the filtering parameters are known, the admittance matrix considering the frequency coupling characteristics can be directly calculated by formula (21).

[0100] In the embodiment of the present invention, the neural network is good at extracting features from raw data, exploring nonlinear relationships between data, and adaptively adjusting network weights through data features, and can approximate and fit any complex nonlinear function. According to equations (20) and (22), the impedance model of the grid-connected inverter using dual-loop control is essentially a multivariable nonlinear model. Considering the powerful nonlinear fitting ability, flexibility and feature learning ability of the neural network, the neural network can be used to fit and approximate some nonlinear relationships in the impedance model. The setting of the neural network structure is guided by the grid-connected inverter control structure, which can effectively reduce the dependence of the neural network on the amount of data.

[0101] For the training of the neural network, the first step is to select the structure of the neural network. Impedance identification is to establish the mapping relationship between the frequency and working point of the grid-connected inverter and the corresponding impedance. Considering that the input-output relationship of the function in the two nonlinear relationships is progressive, the present invention chooses to use a feedforward neural network (FNN) instead of solving the nonlinear expression.

[0102] The next step is to design the structure of the feedforward neural network. According to equations (18), (20) and (22), the gray box impedance model of the grid-connected inverter contains two nonlinear steps and one conversion. The impedance conversion can be calculated using an exact equation, and the nonlinear function can be represented by a neural network.

[0103] In step S13, the first nonlinear functional relationship is equation (20). The nonlinear process of equation (20) is defined as nonlinear step 1, and the feedforward neural network FNN is used to replace the first nonlinear functional relationship. The output of FNN is and Since the mapping between frequency, frequency domain offset, working point and nonlinear function expression is continuous and surjective, a small number of hidden layers are applied to capture the characteristics of the mapping between frequency, frequency domain offset, working point and nonlinear function expression to construct the first neural network. The first neural network includes a hidden layer, an input layer and an output layer.

[0104] The transfer function of the output voltage of any phase of the grid-connected inverter port is expressed as the division of two polynomials, and the number of neurons in the hidden layer is determined according to the highest power of the denominator polynomial and the numerator polynomial. The transfer function of the phase output voltage is expressed in the general form of the division of two polynomials:

[0105] (twenty three)

[0106] In the formula and are the coefficients of the denominator and numerator polynomials, respectively. and are the highest powers of the denominator and numerator polynomials respectively. In order to enable the neural network to fully represent the nonlinear relationship between each term in the numerator and denominator polynomials, the number of neurons Should meet To improve the generality of the neural network model, the number of neurons in the hidden layer Set as .

[0107] The voltage at any phase output port of the grid-connected inverter is , , , , , The voltage of any phase output port of the grid-connected inverter in the expected frequency band is calculated according to the first nonlinear function relationship, and the voltage of any phase output port of the grid-connected inverter in the expected frequency band is selected according to the mean square error between the calculated value and the measured value. and The average value of the mean square error in is used as the loss function for the first neural network training. For example,

[0108] (twenty four)

[0109] (25)

[0110] is the loss function during the training of the first neural network.

[0111] In the embodiment of the present invention, a feedforward neural network is used to replace the second nonlinear functional relationship, a small number of hidden layers are applied to capture the mapping relationship characteristics of the second nonlinear functional relationship, and a second neural network is constructed. The output of the second neural network is the base frequency current amplitude. , fundamental frequency voltage amplitude The second nonlinear function relationship is formula (22), the nonlinear calculation process of each element of the matrix of formula (22) is defined as nonlinear step 2, and the nonlinear process is replaced by FNN, and the output of FNN is the positive sequence impedance at multiple frequency points.

[0112] According to formula (21), it can be seen that the mapping between the input and output of the nonlinear calculation process replaced by FNN is strongly correlated and surjective, so a small number of hidden layers is sufficient to capture the characteristics of this mapping relationship. The output of FNN is the current amplitude at the base frequency , fundamental frequency voltage amplitude The elements of the broadband admittance matrix of the grid-connected inverter at the working point are obtained. In order to simplify the model verification and data collection process, the grid impedance is used to eliminate the frequency coupling component. The calculation process is shown in formula (18) to obtain the positive sequence impedance of the grid-connected inverter considering the frequency coupling characteristics. Next, the amplitude and phase angle of the positive sequence impedance are obtained through simple programming of formula (26). The loss function of training FNN is shown in formula (27), and the loss function of the second neural network is obtained. :

[0113] (26)

[0114] (27)

[0115] In step S14, firstly, according to the requirements of avoiding overload of the grid-connected inverter and satisfying the stable operation of the power system, the selection range of the working point of the grid-connected system of the grid-connected inverter is determined, and the working point includes the fundamental frequency voltage amplitude, fundamental frequency current and its initial phase angle at the PCC point; Noperating point. For the grid-connected system of the grid-connected inverter, the operating point (OP) involves the fundamental frequency voltage amplitude, fundamental frequency current and its initial phase angle at the PCC point. It can be seen from the impedance analytical equations (11) and (12) of the grid-connected inverter that when the operating conditions of the grid-connected inverter change, the operating point will shift and the impedance value will also change. In order to avoid overloading of the grid-connected inverter and meet the requirements of stable operation of the power system, the operating point can be set within the following range:

[0116] (28)

[0117] In the formula, P is the input power, It refers to the ratio of the actual value of the voltage in the power system to its reference value, also known as the per unit value. To train and verify the model, select N working points, among which N1 working points are used for model training, N2 operating points for model validation ( ).

[0118] Then, a positive-sequence disturbance voltage with a frequency equal to the positive-sequence disturbance frequency is injected at the PCC point of the grid-connected inverter, the positive-sequence current of the port response is measured, and the ratio of the positive-sequence disturbance voltage to the positive-sequence current is calculated to obtain the impedance of the grid-connected inverter at the positive-sequence disturbance frequency. The embodiment of the present invention adopts a sweep frequency method to obtain the measured impedance of the grid-connected inverter: the frequency of the positive-sequence disturbance voltage injected at the grid-connected point port of the grid-connected inverter is The three-phase positive sequence (negative sequence) frequency disturbance voltage vector , measure the positive sequence (negative sequence) disturbance current vector of its port , then the grid-connected inverter The impedance can be given by and In order to improve the efficiency of frequency sweep measurement, the sinusoidal voltage disturbance input signal used in the present invention contains multiple frequencies. Through Fourier decomposition (FFT) processing, the impedance at multiple frequencies can be obtained, that is:

[0119] (29)

[0120] The nonlinear processes replaced by the first neural network and the second neural network in the embodiment of the present invention are different from the physical meanings represented, and the data sets used for training are also different. Optionally, a voltage disturbance including multiple frequencies is injected into the PCC point of a gray box grid-connected inverter simulation model that adopts a dual-loop control structure but has unknown control parameters, and a frequency scan is performed within the expected frequency band to collect the first data set and the second data set required for training the first neural network and the second neural network. Specifically, at a preset operating point n within the selected working range, the expected frequency band is 1~10000HZ, and a disturbance voltage vector is injected into the PCC point of a gray box grid-connected inverter simulation model that adopts a dual-loop control structure but has unknown control parameters. , perform frequency scanning, the frequency points scanned are , representing m different perturbation frequencies, and collecting the data set required to train the neural network.

[0121] At each operating point, the first data set includes: Input data of the first neural network: m perturbation frequencies , m disturbance frequency offsets and , positive sequence frequency disturbance current vector at m disturbance frequencies , disturbance voltage vector and negative sequence coupling response current vector , response voltage vector , including the fundamental frequency current amplitude , fundamental frequency current phase angle and fundamental frequency voltage amplitude The three PCC point measurements and the label data of the first neural network: the m frequency components of any phase voltage of the grid-connected inverter port after filtering and m frequency components of the coupled component ; and write the three PCC point measurement values ​​as Therefore, the first data set at each working point contains a dimension of data.

[0122] At each operating point, the second data set includes: Input data of the second neural network: m perturbation frequencies , the grid side impedance calculated at m disturbance frequencies , the m groups of coefficients output by the second neural network represent the first nonlinear relationship and , and the label data of the second neural network: m positive sequence frequency disturbance current vectors measured by the PCC point , disturbance voltage vector Calculated positive sequence impedance amplitude of grid-connected inverter considering frequency coupling characteristics and phase angle Therefore, the second data set at each working point contains a dimension of The data collection and use process of the first and second data sets are as follows: Figure 4 shown.

[0123] In step S15, the first neural network is first trained based on the first data set. N1 The group data is applied to the first neural network to obtain the coefficients representing the nonlinear relationship at multiple working points. and The specific training process is as follows: Figure 5 As shown, the input data in the first data set , , , , , , , , , Input the first neural network for model training and get the output of the first neural network and , substitute into formula (20) to obtain the predicted label data and . Calculate the first loss function based on the predicted label data and the label data in the first data set, and determine whether the first training end condition is met. , , , Substitute into formula (25) to calculate the first loss function, and determine whether the error is small enough to meet the requirements, that is, whether the first training end condition is met. If the first training end condition is not met, return to the step of inputting the input data in the first data set into the first neural network for model training, otherwise output the coefficient of the first nonlinear relationship output by the first neural network and .

[0124] Then, the second neural network model is trained based on the second data set. The coefficients representing the nonlinear relationship under multiple working points are , And in Dataset 2 N1 The group data is applied to the neural network 2 constructed in step S3 to obtain a gray box sequence impedance model of a grid-connected inverter that is adaptable to multiple working points and adopts a dual-loop control structure. Figure 6 As shown, the input data in the second data set , , and Input the second neural network for model training and get the output of the second neural network , Substitute into equation (18) and equation (26) to get the predicted label data and ; Calculate the second loss function based on the predicted label data and the label data in the second data set, and determine whether the second training end condition is met. and , and Substitute into formula (27) to calculate the second loss function and determine whether the error meets the requirement, that is, the second training end condition. If the second training end condition is not met, return to the step of inputting the input data in the second data set into the second neural network for model training. Otherwise, obtain the gray box impedance model of the grid-connected inverter that adapts to multiple working points, and calculate the positive sequence impedance amplitude of the grid-connected inverter considering the frequency coupling characteristics. and phase angle .

[0125] Finally, the first and second data sets N2 The group data is applied to the trained model to verify the model. In the present invention, the process from the construction of the grid-connected inverter gray box impedance model to the multi-operating point impedance identification of the grid-connected inverter is as follows: Figure 7 As shown, including:

[0126] Step 100: Start.

[0127] Step 101: Based on a general control scheme including dual-loop control, an analytical expression of the sequence impedance of the grid-connected inverter considering the frequency coupling characteristics is derived.

[0128] For the specific derivation process, please refer to the specific content of the previous step S11. The final analytical expression of the sequence impedance of the grid-connected inverter considering the frequency coupling characteristics is as shown in formula (18).

[0129] Step 102: extract key nonlinear functional relationships in the impedance analysis model, and use a neural network to mine and replace the nonlinear functional relationships.

[0130] The key nonlinear functional relationships in the impedance analysis model are extracted, such as equations (20) and (22). The first neural network is used to mine and replace equation (20), and the second neural network is used to mine and replace equation (22).

[0131] Step 103: Design a neural network structure based on the impedance analysis model characteristics and construct a loss function 1 ( ) and loss function 2 ( ).

[0132] The loss function 1 of the first neural network is obtained by referring to the previous formula (25), and the loss function 2 of the second neural network is obtained by referring to the previous formula (27).

[0133] Step 104: Set up according to reasonable power system operation requirements N working points, among which N1 working points as training set, N2 working points as the test set ( N1+N2=N ).

[0134] According to reasonable power system operation requirements, set N working points to facilitate subsequent data collection.

[0135] Step 105: For a gray-box grid-connected inverter at working point n that adopts a dual-loop control structure but has unknown control parameters, a voltage disturbance is injected into the grid-connected point.

[0136] At any operating point n, a voltage disturbance containing multiple frequencies is injected into the grid-connected point.

[0137] Step 106: Disturbance frequency s, frequency offset ( ), obtain the filtered response voltage of the grid-connected inverter port ( , ), grid connection point voltage ( V 1 , V P , V P2 ) and current ( , , , ) constitute the first data set.

[0138] Data for training the first neural network at working point n is collected and added to the first data set.

[0139] Step 107: Disturb frequency s, and perturb voltage vector by frequency V P , and the disturbance current vector I p The calculated sequence impedance amplitude and phase angle of the grid-connected inverter and the grid-side impedance A second data set is formed.

[0140] Data for training the second neural network at the working point n is collected and added to the second data set.

[0141] Step 108: n+1.

[0142] Set the next operating point n+1.

[0143] Step 109: Determine whether n> N If yes, then execute step 110 , otherwise return to step 105 .

[0144] If n> N , indicating that it has been completed N The data of the working points are collected, and the model training can be carried out later. N , indicating that it is not yet completed N The data collection of each working point needs to be continued, so the process returns to step 105 to collect the data of the next working point.

[0145] Step 110: Substitute the first data set into N1 The first neural network is trained by using a set of data to identify the nonlinear functional relationship between the port voltage of the grid-connected inverter and the frequency, voltage, and current variables. A i and B i (i=1,2,3,4) .

[0146] Apply the first dataset N1 The first neural network is trained with the data set. The specific training process is shown in Figure 5 , I will not go into details about this.

[0147] Step 111: Substitute the second data set into N1 Group data, A i and B i (i=1,2,3,4) The second neural network is trained to identify the sequence impedance amplitude and phase angle of the grid-connected inverter considering the frequency coupling characteristics.

[0148] Apply the second dataset N1 The data set and the output of the first neural network A i and B i (i=1,2,3,4) Train the second neural network. See Figure 6 , I will not go into details about this.

[0149] Step 112: Put the N2 The group data is substituted into the trained model for verification.

[0150] According to the first data set and the second data set N2 The trained first neural network and the trained second neural network are verified by using a set of data.

[0151] Step 113: A grid-connected inverter impedance gray box impedance model adapted to multiple operating points.

[0152] After completing the training and verification of the first neural network and the second neural network, the impedance gray box impedance model of the grid-connected inverter that adapts to multiple working points is obtained, that is, the elements of the admittance matrix can be obtained. Further, according to equations (18) and (26), the positive sequence impedance amplitude and phase angle of the grid-connected inverter considering the frequency coupling characteristics can be calculated.

[0153] Step 114: End.

[0154] At this point, the gray box impedance identification of the grid-connected inverter that can adapt to multiple operating points is completed.

[0155] The above is only a preferred embodiment of the present invention. The selected FNN can be replaced by other neural networks with better effects, and is not intended to limit the protection scope of the present invention.

[0156] The gray box impedance identification method for grid-connected inverters adapting to multiple operating points in an embodiment of the present invention combines a neural network with powerful nonlinear relationship fitting capabilities with the structural knowledge of the grid-connected inverter, and can obtain a sequence impedance model adapting to multiple operating points. Relying on the advantages of neural networks in processing multidimensional data and mining data features, the constructed model can characterize the impedance transfer function of the grid-connected inverter within a wide working range under limited data conditions. It has strong interpretability and engineering use value, and provides an accurate model basis for the study of broadband oscillation problems of new energy grid connection.

[0157] The embodiment of the present invention determines the sequence impedance analytical expression of the grid-connected inverter considering the frequency coupling characteristics based on a control scheme including dual-loop control; extracts the first nonlinear functional relationship of the frequency domain expression of the output port voltage of any phase of the grid-connected inverter and the second nonlinear functional relationship of the admittance matrix considering the frequency coupling characteristics of the grid-connected inverter in the impedance analytical model of the grid-connected inverter based on the sequence impedance analytical expression; utilizes a feedforward neural network to mine and replace the first nonlinear functional relationship, and constructs a first neural network and a corresponding loss function; utilizes a feedforward neural network to mine and replace the second nonlinear functional relationship, and constructs a second neural network and a corresponding loss function; injects voltage disturbance into the grid connection point of the gray box grid-connected inverter simulation model that adopts a dual-loop control structure but has unknown control parameters, and performs data acquisition to obtain a voltage disturbance having a frequency domain expression of the output port voltage of any phase of the grid-connected inverter and a second nonlinear functional relationship. NThe two sets of data are used to train the first data set and the second data set required by the first neural network and the second neural network respectively; the first neural network and the second neural network are trained based on the first data set and the second data set respectively to obtain a gray box impedance model of a grid-connected inverter that adapts to multiple working points, and calculate the positive sequence impedance amplitude and phase angle of the grid-connected inverter considering the frequency coupling characteristics, which can improve the accuracy of neural network model training with a limited amount of data and realize accurate identification of the impedance of the grid-connected inverter under multiple working points.

[0158] like Figure 8 FIG. 1 is a structural diagram of a gray box impedance identification device for a grid-connected inverter adapted to multiple operating points provided by an embodiment of the present invention. The gray box impedance identification device for a grid-connected inverter adapted to multiple operating points includes:

[0159] A sequence impedance analysis unit 801 is used to determine an analytical expression of the sequence impedance of the grid-connected inverter considering the frequency coupling characteristics based on a control scheme including dual-loop control;

[0160] A nonlinear relationship determination unit 802 is used to extract a first nonlinear functional relationship in the frequency domain expression for characterizing the voltage of any phase output port of the grid-connected inverter and a second nonlinear functional relationship for characterizing the admittance matrix of the grid-connected inverter considering the frequency coupling characteristics in the impedance analytical model of the grid-connected inverter based on the sequence impedance analytical expression;

[0161] The network construction unit 803 is used to use a feedforward neural network to mine and replace the first nonlinear function relationship, to construct a first neural network and a corresponding loss function; and use a feedforward neural network to mine and replace the second nonlinear function relationship, to construct a second neural network and a corresponding loss function;

[0162] The data collection unit 804 is used to inject voltage disturbance into the grid connection point of the gray box grid-connected inverter simulation model that adopts a dual-loop control structure but has unknown control parameters and collect data to obtain N The first data set and the second data set of the group data are respectively used for training the first neural network and the second neural network;

[0163] The gray box impedance identification unit 805 is used to train the first neural network and the second neural network based on the first data set and the second data set respectively, obtain a gray box impedance model of the grid-connected inverter that adapts to multiple working points, and calculate the positive sequence impedance amplitude and phase angle of the grid-connected inverter considering the frequency coupling characteristics.

[0164] The device of the above embodiment is applied to the corresponding method in the above embodiment, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0165] Fig. 9 An example of a physical structure diagram of an electronic device is shown in FIG. Fig. 9 As shown, the electronic device may include: a processor (processor) 901, a communication interface (Communications Interface) 902, a memory (memory) 903 and a communication bus 904, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus. The processor may call the logic instructions in the memory to execute a gray box impedance identification method for a grid-connected inverter that is adapted to multiple operating points. The method comprises: determining a sequence impedance analytical expression of a grid-connected inverter considering frequency coupling characteristics based on a control scheme including dual-loop control; extracting a first nonlinear functional relationship in a frequency domain expression for characterizing a voltage of an output port of any phase of the grid-connected inverter and a second nonlinear functional relationship for characterizing an admittance matrix of the grid-connected inverter considering frequency coupling characteristics in an impedance analytical model of the grid-connected inverter based on the sequence impedance analytical expression; mining and replacing the first nonlinear functional relationship using a feedforward neural network to construct a first neural network and a corresponding loss function; mining and replacing the second nonlinear functional relationship using a feedforward neural network to construct a second neural network and a corresponding loss function; injecting voltage disturbance into a grid-connected point of a gray box grid-connected inverter simulation model that adopts a dual-loop control structure but has unknown control parameters and performing data acquisition to obtain a voltage disturbance having a frequency domain expression of a voltage at any phase of the grid-connected inverter and a second nonlinear functional relationship using a feedforward neural network to construct a second neural network and a corresponding loss function; and obtaining a voltage disturbance having a frequency domain expression of a voltage at any phase of the grid-connected inverter and a second nonlinear functional relationship using a feedforward neural network to construct a second neural network and a corresponding loss function. N The first and second sets of data are used to train the first and second neural networks respectively; the first and second neural networks are trained based on the first and second sets of data respectively to obtain a gray box impedance model of a grid-connected inverter that adapts to multiple working points, and the positive sequence impedance amplitude and phase angle of the grid-connected inverter considering the frequency coupling characteristics are calculated.

[0166] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0167] On the other hand, an embodiment of the present invention further provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, and when the program instructions are executed by a computer, the computer can execute a gray box impedance identification method for a grid-connected inverter that is adapted to multiple operating points provided by the above-mentioned method embodiments. The method comprises: determining a sequence impedance analytical expression of a grid-connected inverter that takes into account frequency coupling characteristics based on a control scheme including dual-loop control; extracting a first nonlinear functional relationship in the frequency domain expression of the output port voltage of any phase of the grid-connected inverter and a second nonlinear functional relationship for characterizing the admittance matrix of the grid-connected inverter that takes into account frequency coupling characteristics in the impedance analytical model of the grid-connected inverter based on the sequence impedance analytical expression; using a feedforward neural network to mine and replace the first nonlinear functional relationship, constructing a first neural network and a corresponding loss function; using a feedforward neural network to mine and replace the second nonlinear functional relationship, constructing a second neural network and a corresponding loss function; injecting voltage disturbances into the grid-connected point of a gray box grid-connected inverter simulation model that adopts a dual-loop control structure but has unknown control parameters and performing data acquisition to obtain a voltage disturbance having a frequency domain expression of the output port voltage of any phase of the grid-connected inverter and a second nonlinear functional relationship to characterize the admittance matrix of the grid-connected inverter that takes into account frequency coupling characteristics. N The first and second sets of data are used to train the first and second neural networks respectively; the first and second neural networks are trained based on the first and second sets of data respectively to obtain a gray box impedance model of a grid-connected inverter that adapts to multiple working points, and the positive sequence impedance amplitude and phase angle of the grid-connected inverter considering the frequency coupling characteristics are calculated.

[0168] On the other hand, an embodiment of the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute a gray box impedance identification method for a grid-connected inverter that is adaptable to multiple operating points provided in the above embodiments. The method comprises: determining a sequence impedance analytical expression of a grid-connected inverter considering frequency coupling characteristics based on a control scheme including dual-loop control; extracting a first nonlinear functional relationship of a frequency domain expression of a voltage of an output port of any phase of the grid-connected inverter and a second nonlinear functional relationship of an admittance matrix considering frequency coupling characteristics of the grid-connected inverter in an impedance analytical model of the grid-connected inverter based on the sequence impedance analytical expression; mining and replacing the first nonlinear functional relationship by a feedforward neural network to construct a first neural network and a corresponding loss function; mining and replacing the second nonlinear functional relationship by a feedforward neural network to construct a second neural network and a corresponding loss function; injecting voltage disturbance into a grid-connected point of a gray box grid-connected inverter simulation model adopting a dual-loop control structure but with unknown control parameters and performing data acquisition to obtain a first data set and a second data set with N groups of data required for training the first neural network and the second neural network respectively; training the first neural network and the second neural network respectively based on the first data set and the second data set to obtain a gray box impedance model of a grid-connected inverter adapting to multiple working points, and calculating the positive sequence impedance amplitude and phase angle of the grid-connected inverter considering frequency coupling characteristics.

[0169] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0170] The above is only a partial implementation of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A gray box impedance identification method for grid-connected inverters adapting to multiple operating points, characterized in that: Applied to a client, the method comprises: Based on the control scheme including dual-loop control, the analytical expression of the sequence impedance of the grid-connected inverter considering the frequency coupling characteristics is determined; Extracting a first nonlinear functional relationship in the frequency domain expression for characterizing the voltage of any phase output port of the grid-connected inverter and a second nonlinear functional relationship for characterizing the admittance matrix of the grid-connected inverter considering the frequency coupling characteristics in the impedance analytical model of the grid-connected inverter based on the sequence impedance analytical expression; A feedforward neural network is used to mine and replace the first nonlinear function relationship, and a first neural network and a corresponding loss function are constructed; a feedforward neural network is used to mine and replace the second nonlinear function relationship, and a second neural network and a corresponding loss function are constructed; The first neural network is constructed by using a feedforward neural network to replace the first nonlinear function relationship, applying a hidden layer to capture the mapping characteristics between frequency, frequency domain offset, working point and nonlinear function expression; the second nonlinear function relationship is replaced by a feedforward neural network, and applying a hidden layer to capture the mapping relationship characteristics between frequency and the second nonlinear function; The voltage disturbance is injected into the grid connection point of the gray box grid-connected inverter simulation model with a dual-loop control structure but unknown control parameters, and data is collected to obtain the N A first data set and a second data set required for training the first neural network and the second neural network respectively; The first neural network and the second neural network are trained based on the first data set and the second data set respectively to obtain a gray box impedance model of a grid-connected inverter that adapts to multiple operating points, and to calculate the positive sequence impedance amplitude and phase angle of the grid-connected inverter considering frequency coupling characteristics.

2. The method according to claim 1, characterized in that The method of extracting a first nonlinear functional relationship in the frequency domain expression for characterizing the voltage of any phase output port of the grid-connected inverter and a second nonlinear functional relationship for characterizing the admittance matrix of the grid-connected inverter considering the frequency coupling characteristics in the impedance analytical model of the grid-connected inverter based on the sequence impedance analytical expression comprises: Based on the analytical expression of sequence impedance, the frequency domain equation of the output port voltage of any phase of the grid-connected inverter is extracted: In the formula To apply positive sequence disturbance frequency Rear grid-connected inverter port The frequency components of the phase voltage, To apply positive sequence disturbance frequency Rear grid-connected inverter port The frequency component of the coupled component of the phase voltage, , , represents the imaginary unit, is the fundamental frequency, is the coupling frequency component, is the PWM modulation coefficient, , The sampling delay is simulated. is the sampling period, , They are respectively the grid-connected inverter in the stationary coordinate system Positive sequence disturbance frequency of phase modulated signal Negative sequence coupling frequency The frequency component at , s is the disturbance frequency, is the disturbance frequency offset; The frequency domain equation of the output port voltage of any phase of the grid-connected inverter is simplified to obtain a first nonlinear function relationship; Extracting the main circuit equation and admittance matrix model in any phase frequency domain based on the sequence impedance analytical expression; Performing matrix calculation in combination with the main circuit equation, the first nonlinear functional relationship and the admittance matrix model to obtain an admittance matrix representation of the grid-connected inverter taking into account frequency coupling characteristics; The admittance matrix representation is simplified to obtain a second nonlinear functional relationship.

3. The method according to claim 1, characterized in that The method of using a feedforward neural network to mine and replace the first nonlinear functional relationship and constructing a first neural network and a corresponding loss function includes: The first nonlinear functional relationship is: In the formula To apply positive sequence disturbance frequency Rear grid-connected inverter port Phase voltage at positive sequence disturbance frequency The frequency component at To apply positive sequence disturbance frequency Rear grid-connected inverter port The coupling component of the phase voltage is at the negative sequence coupling frequency The frequency component at ~ is with , , , , The related nonlinear function, ~ is with , , , , The related nonlinear function, , and are the fundamental frequency, positive sequence disturbance frequency and negative sequence coupling frequency respectively, ; , , , fundamental frequency current amplitude , fundamental frequency voltage amplitude , the initial phase angle of the fundamental frequency current represents the working point, and They are the positive sequence frequency disturbance voltage vector and the negative sequence coupling response voltage vector respectively; and are respectively the positive sequence frequency disturbance current vector and the negative sequence coupling response current vector, s is the disturbance frequency, is the disturbance frequency offset; Using a feedforward neural network to replace the first nonlinear function relationship, applying a hidden layer to capture the characteristics of the mapping between the frequency, the frequency domain offset, the working point and the nonlinear function expression, and constructing a first neural network, wherein the first neural network includes a hidden layer, an input layer and an output layer; The transfer function of the output voltage of any phase of the grid-connected inverter port is expressed as a form of division of two polynomials, and the number of neurons in the hidden layer is determined according to the highest power of the denominator polynomial and the numerator polynomial; The voltage at any phase output port of the grid-connected inverter is , , , , , , , , , , , The voltage of any phase output port of the grid-connected inverter in the expected frequency band is calculated according to the first nonlinear function relationship, and the voltage of any phase output port of the grid-connected inverter in the expected frequency band is selected according to the mean square error between the calculated value and the measured value. and The average value of the mean square error in is used as the loss function during the training of the first neural network.

4. The method according to claim 3, characterized in that The method of using a feedforward neural network to mine and replace the second nonlinear functional relationship and constructing a second neural network and a corresponding loss function includes: The second nonlinear functional relationship is: In the formula ~ is the first nonlinear function ~ , ~ and The nonlinear function of the composition; A feedforward neural network is used to replace the second nonlinear functional relationship, and a hidden layer is applied to capture the mapping relationship characteristics of the second nonlinear functional relationship to construct a second neural network. The output of the second neural network is the base frequency current amplitude. , fundamental frequency voltage amplitude The elements of the broadband admittance matrix of the grid-connected inverter at the working point; The amplitude and phase angle of the positive sequence impedance are obtained by simple programming based on the analytical expression of the sequence impedance. According to the mean square error between the calculated value and the measured value, the average value of the mean square error in the amplitude and phase angle of the positive sequence impedance is selected as the loss function during the training of the second neural network.

5. The method according to claim 1, characterized in that The method injects voltage disturbance into the grid connection point of the gray box grid-connected inverter simulation model which adopts a dual-loop control structure but has unknown control parameters and performs data acquisition to obtain N The first data set and the second data set required for training the first neural network and the second neural network, respectively, include: According to the requirements of avoiding overload of the grid-connected inverter and satisfying the stable operation of the power system, the selection range of the working point of the grid-connected system of the grid-connected inverter is determined, and the working point includes the fundamental frequency voltage amplitude, fundamental frequency current amplitude and its initial phase angle at the PCC point; Select from the selection range N Working points; A positive-sequence disturbance voltage having a frequency equal to the positive-sequence disturbance frequency is injected at the PCC point of the grid-connected inverter, the positive-sequence current responded by the output port is measured, and the ratio of the positive-sequence disturbance voltage to the positive-sequence current is calculated to obtain the impedance of the grid-connected inverter at the positive-sequence disturbance frequency; Voltage disturbances including multiple frequencies are injected into the PCC point of a gray-box grid-connected inverter simulation model that adopts a dual-loop control structure but has unknown control parameters, and a frequency scan is performed within the expected frequency band to collect the first data set and the second data set required for training the first neural network and the second neural network.

6. The method according to claim 1, characterized in that At each working point, the first data set includes: the input data of the first neural network: m , m and , positive sequence frequency disturbance current vector at m disturbance frequencies , positive sequence frequency disturbance voltage vector and negative sequence coupling response current vector , negative sequence coupling response voltage vector , including the fundamental frequency current amplitude , the initial phase angle of the fundamental frequency current and fundamental frequency voltage amplitude The three PCC point measurement values ​​and the label data of the first neural network: the m frequency components of any phase voltage of the grid-connected inverter port after filtering and m frequency components of the coupled component ; and write the three PCC point measurement values ​​as form; At each working point, the second data set includes: the input data of the second neural network: m , the grid side impedance calculated at m disturbance frequencies, the m groups of coefficients of the first nonlinear relationship output by the first neural network , , , and , , , , and the label data of the second neural network: m positive sequence frequency disturbance current vectors measured by the PCC point , positive sequence frequency disturbance voltage vector Calculated positive sequence impedance amplitude of grid-connected inverter considering frequency coupling characteristics and phase angle , where s is the disturbance frequency, is the disturbance frequency offset.

7. The method according to claim 6, characterized in that The first neural network and the second neural network are trained respectively based on the first data set and the second data set to obtain a gray box impedance model of a grid-connected inverter adaptable to multiple working points, including: Inputting the input data in the first data set into the first neural network for model training to obtain predicted label data output by the first neural network; Calculating a first loss function based on the predicted label data and the label data in the first data set, and determining whether a first training end condition is met; If the first training end condition is not met, returning to the step of inputting the input data in the first data set into the first neural network for model training, otherwise outputting the coefficient of the first nonlinear relationship output by the first neural network; Inputting the input data in the second data set into the second neural network for model training to obtain predicted label data output by the second neural network; Calculating a second loss function according to the predicted label data and the label data in the second data set, and determining whether a second training end condition is met; If the second training end condition is not met, return to the step of inputting the input data in the second data set into the second neural network for model training; otherwise, obtain a gray box impedance model of the grid-connected inverter that adapts to multiple operating points.

8. A gray box impedance identification device for grid-connected inverters adapting to multiple working points, characterized in that: Executing the method according to claim 1, the device comprises: A sequence impedance analysis unit, used to determine an analytical expression of the sequence impedance of the grid-connected inverter taking into account the frequency coupling characteristics based on a control scheme including dual-loop control; A nonlinear relationship determination unit, used for extracting a first nonlinear functional relationship of a frequency domain expression for characterizing the voltage of any phase output port of the grid-connected inverter and a second nonlinear functional relationship for characterizing the admittance matrix of the grid-connected inverter considering frequency coupling characteristics in the impedance analytical model of the grid-connected inverter based on the sequence impedance analytical expression; A network construction unit is used to mine and replace the first nonlinear function relationship using a feedforward neural network to construct a first neural network and a corresponding loss function; and to mine and replace the second nonlinear function relationship using a feedforward neural network to construct a second neural network and a corresponding loss function; The data collection unit is used to inject voltage disturbance into the grid connection point of the gray box grid-connected inverter simulation model with a dual-loop control structure but unknown control parameters and collect data to obtain N A first data set and a second data set required for training the first neural network and the second neural network respectively; A gray box impedance identification unit is used to train the first neural network and the second neural network based on the first data set and the second data set respectively, to obtain a gray box impedance model of a grid-connected inverter that adapts to multiple operating points, and to calculate the positive sequence impedance amplitude and phase angle of the grid-connected inverter considering the frequency coupling characteristics.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, a gray box impedance identification method for a grid-connected inverter adaptable to multiple operating points is implemented as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, a gray box impedance identification method for a grid-connected inverter adaptable to multiple operating points is implemented as described in any one of claims 1 to 7.

Images