A wideband impedance identification model, method and system for a flexible low frequency power transmission system

Abstract

The present application relates to the technical field of flexible power transmission, and in particular to a wideband impedance identification model, method and system of a flexible low-frequency power transmission system. The present application proposes a training method of a wideband impedance identification model of a flexible low-frequency power transmission system, introduces a power conservation loss in the model training process, establishes a wideband impedance identification model of a flexible low-frequency power transmission system considering the M3C low-frequency side coupling effect, and can accurately analyze the wideband oscillation generation mechanism of the flexible low-frequency power transmission system considering the M3C low-frequency side coupling effect. The present application solves the problem that due to the complex M3C topology structure, the existence of the low-frequency side coupling effect, and the limitation of commercial technology secrecy, the related parameters cannot be accurately obtained, and it is difficult to establish the wideband impedance identification model through theoretical derivation.

Description

Technical Field

[0001] This invention relates to the field of flexible power transmission technology, and in particular to a broadband impedance identification model, method and system for flexible low-frequency power transmission systems. Background Technology

[0002] Flexible low-frequency transmission systems (FLPS) are a new type of transmission system with a high proportion of power electronics. They contain numerous power electronic devices, such as AC converters, and nonlinear control components. Compared to DC transmission systems, they exhibit inter-frequency interaction characteristics, making their dynamic characteristics more complex. Dynamic interactions occur between these power electronic devices and between them and the AC network, leading to the risk of broadband oscillations, which can cause power quality deterioration, equipment damage, and even system shutdowns. Impedance analysis, due to its clear physical meaning and simple application, is currently the primary analytical method for studying broadband oscillations in power electronic power systems. Impedance modeling is a fundamental prerequisite for using impedance analysis to study broadband oscillations in flexible FLPS systems.

[0003] Currently, for the broadband oscillation stability problem of M3C-based flexible low-frequency transmission systems, impedance modeling often references DC impedance models and AC / DC impedance models, considering only the influence of the M3C (Modular Multilevel Matrix Converter) low-frequency or power frequency side on the AC network. It neglects the power frequency coupling relationship of the M3C itself and the influence of inter-frequency energy exchange between M3C arms, establishing only a single-sided impedance model. This leads to low accuracy in impedance modeling, thus affecting the accuracy of research on broadband oscillation problems in flexible low-frequency transmission systems. Furthermore, if general neural networks (such as BP neural networks) are used, they are purely data-driven methods, relying solely on training data samples to obtain certain distribution patterns. This requires a large number of training samples, and because the M3C converter involves coupling on the power and low frequency sides, existing training methods lack interpretability, making it difficult to obtain an accurate broadband impedance identification model. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, such as low accuracy and poor interpretability of wideband impedance identification of M3C converters using existing machine learning methods, this invention proposes a training method for a wideband impedance identification model of flexible low-frequency transmission systems. This method can establish an accurate and efficient wideband impedance identification model of flexible low-frequency transmission systems that considers the mutual coupling between the M3C and low-frequency sides.

[0005] This invention proposes a training method for a broadband impedance identification model of a flexible low-frequency transmission system. The broadband impedance identification model identifies the broadband impedance of a modular multilevel matrix converter (M3C) based on its operating current, voltage, and frequency f.

[0006] The loss function during the training process of the broadband impedance identification model is:

[0007] Loss = Loss a +Loss phy +Loss bc_ld +Loss bc_lq +Loss bc_sd +Loss bc_sq

[0008] Among them, Loss a Loss is the model's prediction loss. phy Loss is due to power conservation. bc_ld Loss is the d-axis loss on the low-frequency side. bc_lq For low-frequency q-axis loss; Loss bc_sd Loss is the d-axis loss on the power frequency side. bc_sq This refers to the q-axis loss on the power frequency side;

[0009] Power conservation loss, Loss phy In the M3C submodule capacitor voltage U dc Above calculation, U dc The power conservation constraint condition is satisfied:

[0010]

[0011] Among them, C dc Indicates the equivalent capacitance value of the M3C submodule; U sd and I sd U represents the voltage and current on the d-axis of the M3C's power frequency side. sq and I sq U represents the voltage and current on the q-axis of the M3C's power frequency side. ld and I ld U represents the voltage and current on the d-axis of the low-frequency side of the M3C. lq and I lq This represents the voltage and current on the d-axis of the low-frequency side of the M3C.

[0012] Preferred, Loss bc_ld and Loss bc_lq The d-axis common-mode voltage U on the low-frequency side of M3C m3cd and q-axis common-mode voltage U m3cq Above calculation; U m3cd and U m3cq Satisfy the low-frequency side constraint conditions:

[0013]

[0014] Where L represents the equivalent inductance of the bridge arm of the flexible low-frequency transmission system, w l Indicates the low-frequency side angle frequency of the M3C; U ld and I ldU represents the voltage and current on the d-axis of the low-frequency side of the M3C. lq and I lq This represents the voltage and current on the d-axis of the low-frequency side of the M3C.

[0015] Preferred, Loss bc_sd and Loss bc_sq The common-mode voltage U on the d-axis of the M3C power frequency side is respectively comd and q-axis common-mode voltage U comq Above calculation, U comd and U comq

[0016] The following power frequency side constraints must be met:

[0017]

[0018] Among them, w s U represents the power frequency side angular frequency of the M3C. sd and I sd U represents the voltage and current on the d-axis of the M3C's power frequency side. sq and I sq This indicates the voltage and current on the q-axis of the M3C's power frequency side.

[0019] Preferred, Loss bc_ld Using the M3C low-frequency side d-axis common-mode voltage U m3cd The low-frequency side d-axis common-mode voltage U represented by the hidden layer output of the model. m3cd Cross-entropy loss of process quantities; Loss bc_lq Using the low-frequency side q-axis common-mode voltage U of M3C m3cq The low-frequency side q-axis common-mode voltage U represented by the hidden layer output of the model. m3cq Cross-entropy loss of process quantities; Loss bc_sd Using M3C power frequency side d-axis common mode voltage U comd The hidden layer output of the model represents the d-axis common-mode voltage U on the power frequency side. comd Cross-entropy loss of process quantities; Loss bc_sq Using M3C power frequency side q-axis common mode voltage U comq The hidden layer output of the model represents the q-axis common-mode voltage U on the power frequency side. comq The cross-entropy loss of the process quantity.

[0020] Preferred, Loss phy The submodule capacitor voltage U of M3C dc The capacitance voltage U of the representation submodule output by the hidden layer of the model dc The cross-entropy loss of the process quantity.

[0021] Preferably, the broadband impedance identification model adopts the KAN neural network model.

[0022] Preferably, the learning dataset for the broadband impedance identification model is denoted as {U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq 、f;Z' dq (freq)},U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq and Z' dq (freq) represents U sd I sd U sq I sq U ld I ld U lq I lq and Z dq The normalized value of (freq);

[0023] U sd and I sd U represents the voltage and current on the d-axis of the M3C's power frequency side. sq and I sq U represents the voltage and current on the q-axis of the M3C's power frequency side. ld and I ld U represents the voltage and current on the d-axis of the low-frequency side of the M3C. lq and I lq This represents the voltage and current on the d-axis of the M3C low-frequency side; Z dq (freq) is the wideband impedance of the M3C; Loss a The normalized value of the predicted impedance output by the model and Z' dq Cross-entropy loss of (freq).

[0024] This invention proposes a broadband impedance identification method for flexible low-frequency transmission systems. First, a broadband impedance identification model is trained, and then the model is combined with a dataset {U}. sd I sd U sq I sq U ld I ld U lq I lq f; Zdq (freq)} normalizes the operating condition samples to construct the operating condition input sample {U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq The operating condition input samples are combined with the wideband impedance identification model with different scanning frequencies f0. The obtained impedance normalization values ​​are inversely normalized to obtain the predicted impedance at each scanning frequency.

[0025] This invention proposes a wideband impedance control method, which obtains the target impedance by scanning at different frequencies f under a specified steady-state operating point; designs the current and voltage data of the M3C power frequency side and low frequency side ports in a synchronous rotating coordinate system, and constructs a design scheme {U}. sd I sd U sq I sq U ld I ld U lq I lq The design scheme is normalized by combining the training dataset of the broadband impedance identification model, and then input into the broadband impedance identification model to obtain the predicted impedance corresponding to the design scheme; the current and voltage data in the design scheme are adjusted until the predicted impedance is consistent with the target impedance.

[0026] This invention proposes a broadband impedance identification system for flexible low-frequency power transmission systems, comprising:

[0027] The data acquisition module is used to collect current and voltage data of the M3C power frequency side and low frequency side ports in a synchronous rotating coordinate system at various steady-state operating points.

[0028] The data processing module preprocesses the collected current and voltage data, filters data samples, and calculates the corresponding broadband impedance through simulation based on the filtered data samples to construct the initial sample {U}. sd I sd U sq I sq U ld I ld U lq I lq f; Z dq (freq)},U sd and I sd U represents the voltage and current on the d-axis of the M3C's power frequency side. sq and I sqU represents the voltage and current on the q-axis of the M3C's power frequency side. ld and I ld U represents the voltage and current on the d-axis of the low-frequency side of the M3C. lq and I lq This represents the voltage and current on the d-axis of the M3C low-frequency side; Z dq (freq) is the wideband impedance of the M3C; f is the scanning frequency;

[0029] The model training module normalizes the initial samples to obtain the dataset {U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq 、f;Z' dq (freq)},U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq and Z' dq (freq) represents U sd I sd U sq I sq U ld I ld U lq I lq and Z dq The normalized value of (freq); train a neural network model on the dataset as a broadband impedance identification model;

[0030] The impedance identification module acquires the current and voltage data {U} of the flexible low-frequency transmission system to be identified at a specified steady-state operating point and frequency f. sd I sd U sq I sq U ld I ld U lq I lq Then perform normalization processing, and then normalize the value {U'} sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I'lq} and frequency f are used as inputs to a wideband impedance identification model to obtain the corresponding wideband impedance Z. dq (freq).

[0031] The advantages of this invention are:

[0032] (1) The present invention proposes a training method for a broadband impedance identification model of a flexible low-frequency transmission system. During model training, a power conservation loss is introduced, establishing a broadband impedance identification model of the flexible low-frequency transmission system considering the low-frequency coupling effect. This model can accurately analyze the broadband oscillation generation mechanism of the flexible low-frequency transmission system considering the low-frequency coupling effect of the M3C (Multi-channel Cryogenic) system. The present invention solves the problem that due to the complex topology of the M3C system, the existence of low-frequency coupling, and the restriction of commercial technology confidentiality, it is difficult to accurately obtain relevant parameters and establish a broadband impedance identification model through theoretical derivation.

[0033] (2) This invention incorporates the energy conservation formulas of the M3C power frequency side / low frequency side and DC side as regularization of physical constraints into the loss function of the neural network to construct a physical information neural network model. By adding physical information constraints, this invention considers the coupling effect of the M3C power frequency side and low frequency side, as well as the influence of inter-frequency energy interaction on the system impedance frequency characteristics. This enables the acquisition of more accurate broadband impedance data for flexible low-frequency transmission systems and improves the interpretability of the broadband impedance identification model. It is used to analyze the broadband oscillation characteristics of flexible low-frequency transmission systems, which is beneficial to the study of the broadband oscillation generation mechanism of flexible low-frequency transmission systems. It also has important guiding significance for the design of control strategies and the suppression of broadband oscillations in practical flexible low-frequency transmission projects.

[0034] (3) This invention establishes a broadband impedance identification model for flexible low-frequency transmission systems based on the KAN neural network method. It does not require knowledge of the topology and controller parameters of the M3C in the flexible low-frequency transmission system. This solves the problem that the system topology is complex and the relevant parameters cannot be accurately obtained for theoretical derivation due to the restriction of commercial technology confidentiality.

[0035] (4) The broadband impedance identification method and control method of the flexible low-frequency power transmission system proposed in this invention can quickly and accurately predict and adjust the broadband impedance under different steady-state operating points through the broadband impedance identification model, so as to ensure the working reliability of the flexible low-frequency power transmission system. Attached Figure Description

[0036] Figure 1 This is a flowchart of the dataset construction method used in this invention;

[0037] Figure 2 This is a flowchart of the model training method proposed in this invention;

[0038] Figure 3This is a flowchart of a broadband impedance identification method for a flexible low-frequency power transmission system proposed in this invention.

[0039] Figure 4 This is a schematic diagram of the flexible low-frequency power transmission system module in the embodiment;

[0040] Figure 5 This is a schematic diagram of the operation process for an example embodiment;

[0041] Figure 6 This is a partial impedance effect diagram shown in the embodiments;

[0042] Figure 7 This is a diagram illustrating the impedance effect of the remaining portion in the embodiment. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0044] The symbols defined in this application are as follows:

[0045] d and q represent the d-axis and q-axis, respectively; s and l represent the power frequency side and low frequency side of the modular multilevel matrix converter (M3C), respectively.

[0046] The M3C power frequency side output port d-axis is abbreviated as M3C power frequency side d-axis, denoted by the subscript sd;

[0047] The d-axis of the M3C low-frequency side output port is abbreviated as M3C low-frequency side d-axis, and is denoted by the subscript ld;

[0048] The M3C power frequency side output port q-axis is abbreviated as M3C power frequency side q-axis, denoted by the subscript sq;

[0049] The q-axis of the M3C low-frequency side output port is abbreviated as M3C low-frequency side q-axis, denoted by the subscript lq;

[0050] Zkj represents the impedance generated by the influence of the k-axis on the j-axis, or simply the mutual impedance between the k-axis and the j-axis; Zjj represents the impedance generated by the influence of the j-axis on itself, or simply the j-axis self-impedance; k∈{sd,ld,sq,lq}, j∈{sd,ld,sq,lq}; sd, ld, sq, and lq represent the d-axis on the power frequency side, the d-axis on the low frequency side, the q-axis on the power frequency side, and the q-axis on the low frequency side, respectively.

[0051] This embodiment proposes a broadband impedance identification model for flexible low-frequency power transmission systems, which is used to identify the broadband impedance of a modular multilevel matrix converter (M3C) based on its operating current, voltage, and frequency f.

[0052] The operating voltage of the M3C includes the d-axis voltage U on the power frequency side. sd q-axis voltage U on the power frequency side sq Low-frequency side d-axis voltage U ld and low-frequency side q-axis voltage U lq ; s represents the power frequency side, l represents the low frequency side;

[0053] The operating current of the M3C includes the d-axis current I on the power frequency side. sd q-axis current I on the power frequency side sq Low-frequency side d-axis current I ld and low-frequency side q-axis current I lq ;

[0054] M3C's wideband impedance Z dq (freq) is written as:

[0055]

[0056] Where Zsdsd and Zsqsq represent the d-axis and q-axis self-impedances of the M3C on the power frequency side, respectively; Zldld and Zlqlq represent the d-axis and q-axis self-impedances of the M3C on the low frequency side, respectively; Zsdsq and Zsqsd represent the mutual impedances of the d-axis and q-axis of the M3C on the power frequency side, respectively; Zldlq and Zlqld represent the mutual impedances of the d-axis and q-axis of the M3C on the low frequency side, respectively; and Zsdld is the mutual impedance between the power frequency side d-axis and the low frequency side d-axis. Zsdlq is the mutual impedance between the power frequency side d-axis and the low frequency side q-axis; Zsqld is the mutual impedance between the power frequency side q-axis and the low frequency side d-axis; Zsqlq is the mutual impedance between the power frequency side q-axis and the low frequency side q-axis; Zldsd is the mutual impedance between the low frequency side d-axis and the power frequency side d-axis; Zldsq is the mutual impedance between the low frequency side d-axis and the power frequency side q-axis; Zlqsd is the mutual impedance between the low frequency side q-axis and the power frequency side d-axis; Zlqsq is the mutual impedance between the low frequency side q-axis and the power frequency side q-axis.

[0057] Reference Figure 1 The method for constructing the dataset for training a broadband impedance identification model for flexible low-frequency transmission systems includes the following steps:

[0058] Step S1: Build an electromagnetic transient time-domain simulation model of the flexible low-frequency transmission system based on electromagnetic transient simulation software or build a hardware-in-the-loop simulation model of the flexible low-frequency transmission system based on real-time simulation.

[0059] Step S2: Set the output active power of the flexible low-frequency power transmission system simulation model, that is, set different steady-state operating points. The output active power range is 0 to 1 pu, representing per-unit value. Specifically, in implementation, the steady-state operating points can be set at equal intervals within the output active power range, and the interval is n. In the embodiment, n = 0.005 is set.

[0060] Step S3: Collect the voltages and currents {U sd , I sd , U sq , I sq , U ld , I ld , U lq , I lq} of the M3C power frequency side and low-frequency side in the synchronous rotating coordinate system (dq axis) at each steady-state operating point.

[0061] Step S4: Use the frequency scanning method to obtain the broadband impedance at different steady-state operating points, thereby constructing a data set {U sd , I sd , U sq , I sq [[ID=​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​

[0063] In determining the working condition {U sd I sd U sq I sq U ld I ld U lq I lq Given the sweep frequency f, frequency sweep obtains the wideband impedance Z. dq The (freq) method is as follows:

[0064] At the steady-state operating point, the amplitude m of the disturbance source is set, and a disturbance voltage of amplitude m is injected into a specified location of the M3C. The simulation model is then scanned at a scanning frequency f to obtain the voltage response matrix U of the M3C. dq (freq) and voltage response matrix I dq (freq); Calculate the wideband impedance Z dq (freq) = U dq (freq)·I dq (freq) -1 .

[0065] It is worth noting that the steady-state operating point and the operating condition {U} sd I sd U sq I sq U ld I ld U lq I lq Correspondingly, at the same steady-state operating point, the impedance Z obtained by scanning at different frequencies f is... dq The frequencies (freq) are different; therefore, multiple frequencies can be used to scan at the same steady-state operating point, thereby obtaining multiple dataset samples at the same steady-state operating point.

[0066] In this embodiment, m is 5% to 10% of the rated voltage of the system port, the scanning frequency f takes a value in the range (0-1000) Hz, and the scanning frequency interval g is 10 Hz.

[0067]

[0068] Where Usj1 and Isj1 represent the voltage disturbance m = aU injected into the power frequency side of M3C when the q-axis injected voltage is 0. sd Then, the voltage and current responses generated by the j-axis of the M3C; 'a' is the rated coefficient, 5% ≤ a ≤ 10%;

[0069] Ulj1 and Ilj1 represent the voltage disturbance m = aU injected into the low-frequency side of the M3C along the d-axis when the q-axis injection voltage is 0. ldThen, the voltage and current responses generated by the j-axis of the M3C;

[0070] Usj2 and Isj2 represent the voltage disturbance m = aU injected into the power frequency side of M3C when the d-axis injected voltage is 0. sq Then, the voltage and current responses generated by the j-axis of the M3C;

[0071] Ulj1 and Ilj1 represent the voltage disturbance m = aU injected into the low-frequency side q-axis of the M3C when the d-axis injected voltage is 0. lq Then, the voltage and current responses generated by the j-axis of the M3C;

[0072] j∈{sd,ld,sq,lq}, where sd, ld, sq and lq represent the d-axis on the power frequency side, the d-axis on the low frequency side, the q-axis on the power frequency side, and the q-axis on the low frequency side, respectively.

[0073] Right now:

[0074] Ussd1 and Issd1 represent the voltage and current of the d-axis on the power frequency side of the M3C after injecting a power frequency side d-axis voltage disturbance into the M3C when the q-axis voltage is 0.

[0075] Ussq1 and Issq1 represent the voltage and current of the q-axis on the power frequency side of the M3C after injecting a power frequency side d-axis voltage disturbance into the M3C when the q-axis voltage is 0.

[0076] Usld1 and Isld1 represent the voltage and current of the low-frequency side d-axis of M3C after injecting a power frequency side d-axis voltage disturbance into M3C when the q-axis voltage is 0.

[0077] Uslq1 and Islq1 represent the voltage and current of the low-frequency side q-axis of the M3C after injecting a power frequency side d-axis voltage disturbance into the M3C when the q-axis voltage is 0.

[0078] Ulsd1 and Ilsd1 represent the voltage and current of the d-axis on the power frequency side of the M3C after injecting a low-frequency side d-axis voltage disturbance into the M3C when the q-axis voltage is 0.

[0079] Ulsq1 and Ilsq1 represent the voltage and current of the q-axis on the power frequency side of the M3C after injecting a low-frequency side d-axis voltage disturbance into the M3C when the q-axis voltage is 0.

[0080] Ulld1 and Illd1 represent the voltage and current of the low-frequency side d-axis of M3C after injecting a low-frequency side d-axis voltage disturbance into M3C when the q-axis voltage is 0.

[0081] Ullq1 and Illq1 represent the voltage and current of the low-frequency side q-axis of M3C after injecting a low-frequency side d-axis voltage disturbance into M3C when the q-axis voltage is 0.

[0082] Ussd2 and Issd2 represent the voltage and current on the d-axis of the M3C on the power frequency side after injecting a power frequency side q-axis voltage disturbance into the M3C when the d-axis voltage is 0.

[0083] Ussq2 and Issq2 represent the voltage and current of the q-axis on the power frequency side of M3C after injecting a power frequency side q-axis voltage disturbance into M3C when the d-axis voltage is 0.

[0084] Usld2 and Isld2 represent the voltage and current on the low-frequency side of the M3C after injecting a power frequency side q-axis voltage disturbance into the M3C when the d-axis voltage is 0.

[0085] Uslq2 and Islq2 represent the voltage and current of the low-frequency side q-axis of M3C after injecting a power frequency side q-axis voltage disturbance into M3C when the d-axis voltage is 0.

[0086] Ulsd2 and Ilsd2 represent the voltage and current of the d-axis on the power frequency side of the M3C after injecting a low-frequency side q-axis voltage disturbance into the M3C when the d-axis voltage is 0.

[0087] Ulsq2 and Ilsq2 represent the voltage and current of the q-axis on the power frequency side of the M3C after injecting a low-frequency q-axis voltage disturbance into the M3C when the d-axis voltage is 0.

[0088] Ulld2 and Illd2 represent the voltage and current of the low-frequency side d-axis of M3C after injecting a low-frequency side q-axis voltage disturbance into M3C when the d-axis voltage is 0.

[0089] Ullq2 and Illq2 represent the voltage and current of the low-frequency side q-axis of the M3C after injecting a low-frequency side q-axis voltage disturbance into the M3C when the d-axis voltage is 0.

[0090] Reference Figure 2 The training method for a broadband impedance identification model of a flexible low-frequency transmission system proposed in this embodiment includes the following steps:

[0091] SA1, Transfer the dataset {U sd I sd U sq I sq U ld I ld U lq I lq f; Z dq The normalization process (freq) is applied to obtain the normalized dataset, denoted as the normalized dataset {U'}. sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq、I' lq 、f;Z' dq (freq)}. The normalization method specifically adopts the minimum-maximum normalization method; U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq and Z' dq (freq) represents U sd I sd U sq I sq U ld I ld U lq I lq and Z dq The normalized value of (freq).

[0092] SA2. Construct and initialize the basic model. The input of the basic model is {U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq ,f}, the output is Z' dq (freq);

[0093] The base model can specifically adopt the KAN neural network model, which includes an inner layer (hidden layer) and an outer layer (output layer). The input data undergoes forward propagation computation within the base model. The specific forward propagation computation process is as follows:

[0094] Input vector x = {U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq The input variables f are passed to each neuron in the hidden layer, and the univariate function output of all received input variables is calculated and summed. A nonlinear transformation is performed using a univariate activation function to obtain the hidden layer output. The outputs of all hidden layer neurons are weighted and summed to obtain the final output y = Z'. dq (freq) is the normalized value of the amplitude and phase angle of the broadband impedance data.

[0095]

[0096] In the formula, x n h' represents the nth feature of the input vector x, ψ represents a hidden layer neuron, φ represents a univariate activation function; 2N+1 is the total number of neurons, N is the total number of features of x, and m is the neuron ordinal number; m As a transition parameter, φ m ψ represents the activation function corresponding to neuron m; m,n For the m-th neuron, x n The processing function, ψ m,n (x n ) represents the pair of x from the m-th neuron. n The processing results.

[0097]

[0098] in, The output of the hidden layer of the model represents the capacitor voltage U of the submodule of M3C. dc The process quantity, The hidden layer output of the model represents the low-frequency side d-axis common-mode voltage U of M3C. m3cd The process quantity, The hidden layer output of the model represents the low-frequency side q-axis common-mode voltage U of M3C. m3cq The process quantity, The hidden layer output of the model represents the d-axis common-mode voltage U on the M3C power frequency side. comd The process quantity, The hidden layer output of the model represents the q-axis common-mode voltage U on the M3C power frequency side. comq The process quantity.

[0099] SA3: Extract learning samples from the normalized dataset and substitute them into the base model, calculate the loss function Loss, and update the base model in reverse based on the loss function.

[0100] The loss function is:

[0101] Loss = Loss a +Loss phy +Loss bc_ld +Loss bc_lq +Loss bc_sd +Loss bc_sq (2)

[0102] Among them, Loss a The model prediction loss is calculated by combining the output label y and the true label Z' of the base model. dq The loss for (freq) can be calculated using the root mean square loss.

[0103]

[0104] In the formula, NE represents the number of samples in a single batch, and y i Z' represents the output of the base model for the i-th training sample. dq (freq) i Let represent the true label of the i-th learning sample; ||·||2 represents the 2-norm formula. Specifically, the 2-norm of the amplitude and the 2-norm of the phase angle are calculated at each frequency. Then, the results at each frequency are summed, and the mean is calculated as the loss. a .

[0105] Loss phy This is due to power conservation losses;

[0106]

[0107] Among them, U dci U represents the submodule capacitor voltage of the i-th learning sample. dc The submodule capacitor voltage U in a flexible low-frequency power transmission system dc The following power conservation constraints must be met:

[0108]

[0109] C dc Indicates the equivalent capacitance value of the submodule;

[0110] Loss bc_ld Loss is the d-axis loss on the low-frequency side. bc_lq This represents the q-axis loss on the low-frequency side.

[0111]

[0112] Among them, U m3cdi U m3cqi U represents the d-axis common-mode voltage U of the M3C low-frequency side of the i-th training sample. m3cd and q-axis common-mode voltage U m3cq ;

[0113] The calculation of the low-frequency side common-mode voltage of M3C in a flexible low-frequency transmission system satisfies the following low-frequency side constraints:

[0114]

[0115]

[0116] L represents the equivalent inductance of the bridge arm in a flexible low-frequency transmission system, w l This indicates the low-frequency side angle frequency of the M3C.

[0117] Loss bc_sdLoss is the d-axis loss on the power frequency side. bc_sq This refers to the q-axis loss on the power frequency side;

[0118]

[0119] Among them, U comdi U comqi Let U represent the d-axis common-mode voltage U of the M3C power frequency side of the i-th training sample. comd and q-axis common-mode voltage U comq ;

[0120] The calculation of the common-mode voltage on the power frequency side of the M3C in the flexible low-frequency transmission system satisfies the following power frequency side constraints:

[0121]

[0122] Among them, w s Indicates the power frequency side angular frequency of M3C

[0123] and Both are outputs of the hidden layer of the basic model, which are process quantities for impedance identification.

[0124] In step SA3, after calculating the loss function Loss, for each hidden layer neuron m and input variable x... n An adaptive optimization algorithm is used to perform gradient backpropagation and parameter updates on the base model. The specific calculation formula is as follows:

[0125]

[0126] Where, ψ m,n For the m-th neuron, x n The processing function, h' m For transition parameters, y is the output of the basic model. h m θ represents the activation value of the m-th hidden layer neuron; θ represents the weight parameters of the updated univariate function, i.e., the parameters to be updated in the base model; and μ represents the parameter update learning rate.

[0127] SA4. Determine whether the training of the basic model has reached the convergence condition. The convergence condition is set as follows: the number of basic model updates reaches a set value, or the difference of the loss function in H consecutive adjacent rounds is less than a set threshold.

[0128] No, then return to step SA3;

[0129] Yes, then the fixed foundation model is used as the broadband impedance identification model for flexible low-frequency transmission systems.

[0130] Reference Figure 3This embodiment proposes a broadband impedance identification method for flexible low-frequency transmission systems, comprising the following steps:

[0131] St1, Obtain operating condition samples of the flexible low-frequency transmission system at a specified steady-state operating point {U sd I sd U sq I sq U ld I ld U lq I lq};

[0132] St2, combining dataset {U sd I sd U sq I sq U ld I ld U lq I lq f; Z dq (freq)} normalizes the operating condition samples to construct the operating condition input sample {U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq 、f0};

[0133] St3, input the working condition input sample into the trained broadband impedance identification model, and output the predicted impedance normalization value corresponding to different scanning frequencies f0.

[0134] St4. Perform inverse normalization on the normalized value of the predicted impedance to obtain the predicted impedance, which includes the predicted amplitude and the predicted phase angle.

[0135] St5. Summarize the predicted impedances at different scanning frequencies to obtain the frequency response curve of the broadband impedance required for the flexible low-frequency transmission system.

[0136] The above broadband impedance identification model is verified in conjunction with specific embodiments below.

[0137] In this embodiment, the KAN neural network is used as the basic model to train the impedance identification model.

[0138] Reference Figure 5 In this embodiment, the construction Figure 4 The flexible low-frequency power transmission system shown has multiple steady-state operating points and corresponding operating condition data {U}. sd I sdU sq I sq U ld I ld U lq I lq The broadband impedance frequency characteristics of a flexible low-frequency transmission system were obtained through simulation and frequency sweeping in the PSCAD / EMTP environment. The sweeping frequency f was scanned at 10Hz intervals in the interval (0, 1000] Hz, that is, at each steady-state operating point, the sweeping frequency was 10Hz, 20Hz, 30Hz, ..., 990Hz, 1000Hz, respectively, to obtain the broadband impedance at each frequency, thereby constructing a dataset {U}. sd I sd U sq I sq U ld I ld U lq I lq f; Z dq (freq)}, and transfer the data from the dataset to U sd I sd U sq I sq U ld I ld U lq I lq Z dq After normalization of (freq), we obtain the normalized dataset {U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq 、f;Z' dq (freq)}.

[0139] In this embodiment, the normalized dataset is divided into a training set and a test set. On the training set, the KAN neural network is trained as a broadband impedance identification model using steps SA2-SA4 described above. Then, on the test set, the normalized samples {U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lqThe wideband impedance identification model is input by f, and the output value is inversely normalized to obtain the corresponding predicted impedance. During the test, by continuously changing the frequency f, the trend of the wideband impedance frequency characteristic with frequency at the steady-state operating point can be obtained.

[0140] In this embodiment, the simulated frequency sweep data on the test set is compared with the predicted data of the broadband impedance identification model. Figure 6 , Figure 7 As shown, the red dots represent the broadband impedance frequency characteristics obtained by frequency sweep simulation in the PSCAD / EMTP environment, i.e., the broadband impedance marked on the test sample; the blue curve represents the real-time broadband impedance frequency characteristics calculated by the broadband impedance identification model. The two have a high degree of agreement, proving the correctness of the online broadband impedance identification model for flexible low-frequency transmission systems.

[0141] Of course, those skilled in the art will recognize that the present invention is not limited to the details of the exemplary embodiments described above, but also includes the same or similar structures that can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0142] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

[0143] The technologies, shapes, and structures not described in detail in this invention are all known technologies.

Claims

1. A training method for a broadband impedance identification model of a flexible low-frequency transmission system, characterized in that, The wideband impedance identification model identifies the wideband impedance of the modular multilevel matrix converter M3C based on its operating current, voltage, and frequency f. The loss function during the training process of the broadband impedance identification model is: Loss=Loss a +Loss phy +Loss bc_ld +Loss bc_lq +Loss bc_sd +Loss bc_sq Among them, Loss a Loss is the model's prediction loss. phy Loss is due to power conservation. bc_ld Loss is the d-axis loss on the low-frequency side. bc_lq For low-frequency q-axis loss; Loss bc_sd Loss is the d-axis loss on the power frequency side. bc_sq This refers to the q-axis loss on the power frequency side; Power conservation loss, Loss phy In the M3C submodule capacitor voltage U dc Above calculation, U dc The power conservation constraint condition is satisfied: Among them, C dc Indicates the equivalent capacitance value of the M3C submodule; U sd and I sd U represents the voltage and current on the d-axis of the M3C's power frequency side. sq and I sq U represents the voltage and current on the q-axis of the M3C's power frequency side. ld and I ld U represents the voltage and current on the d-axis of the low-frequency side of the M3C. lq and I lq This represents the voltage and current on the d-axis of the low-frequency side of the M3C.

2. The training method for the broadband impedance identification model of the flexible low-frequency transmission system as described in claim 1, characterized in that, Loss bc_ld and Loss bc_lq The d-axis common-mode voltage U on the low-frequency side of M3C m3cd and q-axis common-mode voltage U m3cq Above calculation; U m3cd and U m3cq Satisfy the low-frequency side constraint conditions: Where L represents the equivalent inductance of the bridge arm of the flexible low-frequency transmission system, w l Indicates the low-frequency side angle frequency of the M3C; U ld and I ld U represents the voltage and current on the d-axis of the low-frequency side of the M3C. lq and I lq This represents the voltage and current on the d-axis of the low-frequency side of the M3C.

3. The training method for the broadband impedance identification model of the flexible low-frequency transmission system as described in claim 2, characterized in that, Loss bc_sd and Loss bc_sq The common-mode voltage U on the d-axis of the M3C power frequency side is respectively comd and q-axis common-mode voltage U comq Above calculation, U comd and U comq The following power frequency side constraints must be met: Among them, w s U represents the power frequency side angular frequency of the M3C. sd and I sd U represents the voltage and current on the d-axis of the M3C's power frequency side. sq and I sq This indicates the voltage and current on the q-axis of the M3C's power frequency side.

4. The training method for the broadband impedance identification model of the flexible low-frequency transmission system as described in claim 3, characterized in that, Loss bc_ld Using the M3C low-frequency side d-axis common-mode voltage U m3cd The low-frequency side d-axis common-mode voltage U represented by the hidden layer output of the model. m3cd Cross-entropy loss of process quantities; Loss bc_lq Using the low-frequency side q-axis common-mode voltage U of M3C m3cq The low-frequency side q-axis common-mode voltage U represented by the hidden layer output of the model. m3cq Cross-entropy loss of process quantities; Loss bc_sd Using M3C power frequency side d-axis common mode voltage U comd The hidden layer output of the model represents the d-axis common-mode voltage U on the power frequency side. comd Cross-entropy loss of process quantities; Loss bc_sq Using M3C power frequency side q-axis common mode voltage U comq The hidden layer output of the model represents the q-axis common-mode voltage U on the power frequency side. comq The cross-entropy loss of the process quantity.

5. The training method for the broadband impedance identification model of the flexible low-frequency transmission system as described in claim 1, characterized in that, Loss phy The submodule capacitor voltage U of M3C dc The capacitance voltage U of the representation submodule output by the hidden layer of the model dc The cross-entropy loss of the process quantity.

6. The training method for the broadband impedance identification model of the flexible low-frequency transmission system as described in claim 1, characterized in that, The wideband impedance identification model uses the KAN neural network model.

7. The training method for the broadband impedance identification model of the flexible low-frequency transmission system as described in any one of claims 1-6, characterized in that, The learning dataset for the broadband impedance identification model is denoted as {U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq 、f;Z' dq (freq)},U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq and Z' dq (freq) represents U sd I sd U sq I sq U ld I ld U lq I lq and Z dq The normalized value of (freq); U sd and I sd U represents the voltage and current on the d-axis of the M3C's power frequency side. sq and I sq U represents the voltage and current on the q-axis of the M3C's power frequency side. ld and I ld U represents the voltage and current on the d-axis of the low-frequency side of the M3C. lq and I lq This represents the voltage and current on the d-axis of the M3C low-frequency side; Z dq (freq) is the wideband impedance of the M3C; Loss a The normalized value of the predicted impedance output by the model and Z' dq Cross-entropy loss of (freq).

8. A broadband impedance identification method for a flexible low-frequency transmission system, employing the training method for a broadband impedance identification model of a flexible low-frequency transmission system as described in claim 7, characterized in that, First, a wideband impedance identification model is trained, and then combined with the dataset {U sd I sd U sq I sq U ld I ld U lq I lq f; Z dq (freq)} normalizes the operating condition samples to construct the operating condition input sample {U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq The operating condition input samples are combined with the wideband impedance identification model with different scanning frequencies f0. The obtained impedance normalization values ​​are inversely normalized to obtain the predicted impedance at each scanning frequency.

9. A broadband impedance control method employing the broadband impedance identification method for flexible low-frequency transmission systems as described in claim 8, characterized in that, Obtain the target impedance by scanning at different frequencies f under a specified steady-state operating point; Design the current and voltage data of the M3C power frequency side and low frequency side ports in a synchronous rotating coordinate system, and construct the design scheme {U sd I sd U sq I sq U ld I ld U lq I lq The design scheme is normalized by combining the training dataset of the broadband impedance identification model, and then input into the broadband impedance identification model to obtain the predicted impedance corresponding to the design scheme. Adjust the current and voltage data in the design scheme until the predicted impedance matches the target impedance.

10. A broadband impedance identification system for a flexible low-frequency transmission system, characterized in that, include: The data acquisition module is used to collect current and voltage data of the M3C power frequency side and low frequency side ports in a synchronous rotating coordinate system at various steady-state operating points. The data processing module preprocesses the acquired current and voltage data, filters data samples, and calculates the corresponding broadband impedance through simulation based on the filtered data samples to construct the initial sample {U}. sd I sd U sq I sq U ld I ld U lq I lq f; Z dq (freq)},U sd and I sd U represents the voltage and current on the d-axis of the M3C's power frequency side. sq and I sq U represents the voltage and current on the q-axis of the M3C's power frequency side. ld and I ld U represents the voltage and current on the d-axis of the low-frequency side of the M3C. lq and I lq This represents the voltage and current on the d-axis of the M3C low-frequency side; Z dq (freq) is the wideband impedance of the M3C; f is the scanning frequency; The model training module normalizes the initial samples to obtain the dataset {U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq 、f;Z' dq (freq)},U' sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq and Z' dq (freq) represents U sd I sd U sq I sq U ld I ld U lq I lq and Z dq The normalized value of (freq); train a neural network model on the dataset as a broadband impedance identification model; The impedance identification module acquires the current and voltage data {U} of the flexible low-frequency transmission system to be identified at a specified steady-state operating point and frequency f. sd I sd U sq I sq U ld I ld U lq I lq Then perform normalization processing, and then normalize the value {U'} sd 、I' sd 、U' sq 、I' sq 、U' ld 、I' ld 、U' lq 、I' lq } and frequency f are used as inputs to a wideband impedance identification model to obtain the corresponding wideband impedance Z. dq (freq).

Images