A data-physical driving based converter full operating domain impedance prediction method

Abstract

The application discloses a converter full operation domain impedance prediction method based on data-physical driving, and belongs to the technical field of stability analysis and control of power electronic power systems. The method comprises the following steps: firstly, different disturbance signals with different disturbance frequencies are applied to the converter at different operation points, and wide-frequency impedance measured values of the converter at different operation points are obtained as a data set; only the voltage operation point and the disturbance frequency are input into a neural network to obtain internal parameters of the converter impedance, and then the current operation point is input into the neural network, and the internal function of the converter impedance is calculated in combination with the internal parameters of the converter impedance, so that the wide-frequency impedance prediction value of the converter is further calculated; and a loss function is calculated according to the impedance prediction value, and the neural network parameters are iteratively updated. In the method, the voltage and current operation points of the converter are split and input into the neural network, so that the number of impedance data required for impedance prediction is greatly reduced, and the extrapolation capacity is strong.

Description

Technical Field

[0001] This invention belongs to the field of stability analysis and control technology of power electronic power systems, specifically involving a data-physical driven method for predicting the impedance of a converter across the entire operating domain. Background Technology

[0002] With the continuous expansion of grid-connected new energy sources such as wind power and photovoltaics, the current power system exhibits a dual characteristic of "high proportion of new energy and high proportion of power electronic equipment." The dense connection of numerous power electronic devices, including new energy generating units and energy storage converters, to the grid, coupled with multi-timescale coupling in the control loop and the presence of low inertia and weak damping characteristics, easily triggers sub- / super-synchronous and mid-to-high frequency broadband oscillations, threatening the safe and stable operation of the power grid. Impedance analysis is an effective means of characterizing the interaction characteristics between power electronic equipment and the power grid, and analyzing the stability of the system under small disturbances. Since the impedance of power electronic devices such as converters is related to their steady-state operating point, and the impedance characteristics of converters differ at different steady-state operating points, it is crucial to quickly and accurately obtain the broadband impedance of grid-connected new energy sources and various power electronic devices across the entire operating domain.

[0003] The existing technical methods mainly include the following:

[0004] (1) Some studies have obtained the impedance values ​​of converters at different steady-state operating points by theoretically modeling the converter impedance. However, theoretical modeling requires knowledge of the specific control structure and control parameters of the converter, which are trade secrets in practice and will not be disclosed by the manufacturers. Therefore, theoretical modeling methods are not suitable for predicting the impedance of actual "black box" converters across the entire operating domain.

[0005] (2) The converter impedance is obtained by actual frequency sweep measurement. However, measuring the converter impedance at each steady-state operating point is time-consuming and laborious, and the impedance measurement operation at unstable operating points cannot be performed. Therefore, the actual impedance measurement method can only obtain the impedance of a few finite stable operating points and cannot conveniently obtain the impedance characteristics of the converter in the entire operating domain.

[0006] (3) The impedance of the converter is obtained based on artificial intelligence algorithms and data-driven methods, and the powerful nonlinear fitting function of neural networks is used to map the functional relationship between the converter impedance and the operating point.

[0007] However, existing data-driven methods generally use neural networks to fit the relationship between converter impedance and the voltage operating point V, current operating points Id, and Iq. This requires three-dimensional data (V, Id, Iq). Due to the limited extrapolation capability of neural networks, the dataset needs to cover as many operating point ranges as possible: voltage operating point V ranging from 0.9 to 1.1 pu, current operating point Id from 0 to 1.0 pu, and current operating point Iq from -1.0 to 1.0 pu. This results in existing neural network training methods requiring a large number of converter impedance values ​​from various operating points as training data. These impedance values ​​require repeated frequency sweep measurements, making data acquisition time-consuming and laborious. Furthermore, the large amount of training data increases computational load and training time, hindering the practical and rapid acquisition of converter impedance values ​​across the entire operating domain.

[0008] Therefore, there is an urgent need to propose a new method for rapidly predicting the impedance value of a converter across its entire operating range. Summary of the Invention

[0009] To address the problems of existing technologies that rely on a large amount of measurement data, involve large computational loads, and are time-consuming in calculating the impedance of a converter across its entire operating domain, this invention provides a data-physics driven method for predicting the impedance of a converter across its entire operating domain.

[0010] The specific technical solution is as follows:

[0011] S1, under multiple voltage and current operating points, by applying disturbance signals of different disturbance frequencies to the converter, the measured values ​​of positive sequence impedance, negative sequence impedance, positive sequence coupling impedance and negative sequence coupling impedance in the converter are obtained.

[0012] S2, input the voltage operating point and disturbance frequency into the neural network to obtain the internal parameters of the converter impedance, and realize data-driven operation;

[0013] S3, input the current operating point into the neural network, combine the converter impedance internal parameters with the current operating point, calculate the converter impedance internal function, and further obtain the predicted values ​​of positive sequence impedance, negative sequence impedance, positive sequence coupling impedance and negative sequence coupling impedance to realize physical drive;

[0014] S4. Based on the measured and predicted values, calculate the loss function and iteratively update the neural network parameters until convergence or the maximum number of iterations is reached, thus obtaining the trained neural network.

[0015] S5. Using the trained neural network, the predicted values ​​of the positive sequence impedance, negative sequence impedance, positive sequence coupling impedance, and negative sequence coupling impedance of the converter at each voltage operating point, current operating point, and disturbance frequency in the entire operating domain are obtained.

[0016] Furthermore, in S1, the measured wideband impedance of the converter is obtained by either the main circuit disturbance method or the secondary control loop disturbance method.

[0017] Furthermore, in S1, the selected voltage operating point is required to cover the entire voltage operating range of the converter.

[0018] Furthermore, in S1, the current operating point includes the d-axis current operating point and the q-axis current operating point.

[0019] Furthermore, in S2, the converter impedance internal parameters include a first internal parameter, a second internal parameter, a third internal parameter, and a fourth internal parameter, each internal parameter containing a real part and an imaginary part.

[0020] Furthermore, in S3, the converter impedance internal function includes a first internal function and a second internal function; the process of calculating the converter impedance internal function by combining the converter impedance internal parameters with the current operating point is as follows:

[0021] The real part of the first internal function is calculated by adding the real part of the second internal parameter to the product of the real part of the second internal parameter and the d-axis current operating point, and then subtracting the product of the imaginary part of the second internal parameter and the q-axis current.

[0022] The imaginary part of the first internal function is calculated by adding the imaginary part of the first internal parameter to the product of the real part of the second internal parameter and the q-axis current operating point, and then adding the imaginary part of the second internal parameter to the product of the d-axis current operating point.

[0023] The real part of the second internal function is calculated by adding the real part of the third internal parameter to the product of the real part of the fourth internal parameter and the d-axis current operating point, and then subtracting the product of the imaginary part of the fourth internal parameter and the q-axis current operating point.

[0024] The imaginary part of the second internal function is calculated by adding the imaginary part of the third internal parameter to the product of the real part of the fourth internal parameter and the q-axis current operating point, and then adding the imaginary part of the fourth internal parameter to the product of the d-axis current operating point.

[0025] Furthermore, in step S3, the formula for calculating the impedance prediction value is as follows:

[0026] ;

[0027] Where real represents the real part and imag represents the imaginary part. The imaginary unit, and This is a term that is only related to the parameters of the main circuit filter. When calculating different impedances, it is necessary to... and Replace with different values; , These are the real and imaginary parts of the first internal function, respectively. , These are the real and imaginary parts of the second inner function, respectively. , denoted as the real and imaginary parts of the predicted converter impedance, respectively, and s is the Laplace operator.

[0028] Furthermore, when calculating different impedances, and The possible values ​​are:

[0029] For positive sequence impedance , ;

[0030] For negative sequence impedance , ;

[0031] For positive sequence coupling impedance , ;

[0032] For negative sequence coupling impedance , ;

[0033] in, As the reference frequency, The frequency is the disturbance frequency.

[0034] Furthermore, in S3, the loss function is calculated by averaging the sum of the squares of the differences between the real and imaginary parts of all predicted and measured values ​​of impedances.

[0035] Furthermore, in step S4, after obtaining the trained neural network, the trained neural network and the validation dataset are used to obtain the predicted values ​​of the positive-sequence impedance, negative-sequence impedance, positive-sequence coupling impedance, and negative-sequence coupling impedance in the validation dataset; the validation dataset includes the voltage operating point, the current operating point, the disturbance frequency, and the corresponding measured values ​​of the positive-sequence impedance, negative-sequence impedance, positive-sequence coupling impedance, and negative-sequence coupling impedance.

[0036] The determination coefficient and root mean square error are calculated by combining the measured and predicted values ​​of all impedances. If the determination coefficient is higher than the determination threshold and the root mean square error is lower than the root mean square threshold, the training is terminated and the process proceeds to S5. Otherwise, the training is restarted.

[0037] The beneficial effects of this invention are:

[0038] (1) The training dataset only requires the voltage operating point V to cover its voltage operating range, and does not require the d-axis current operating point Id and the q-axis current operating point Iq to cover the entire current operating range. Therefore, it is not necessary for the voltage operating point V, the current operating point Id and Iq to cover the entire operating domain, which greatly reduces the amount of impedance data required for neural network training, and also reduces the amount of computation and training time in the training process.

[0039] (2) Since the internal parameters of the four converter impedances predicted during the training of the neural network in this invention are only related to the voltage operating point V and the disturbance frequency fp, the obtained data-physical dual-drive neural network model can predict the impedance values ​​corresponding to the current operating points of Id and Iq in the non-training dataset, and has a strong extrapolation capability.

[0040] (3) The data-physical dual-drive neural network model proposed in this invention is universal for converters with different control structures and different control parameters, and is applicable to "black box" converters with unknown control structure parameters.

[0041] In summary, this invention, based on the physical characteristics between converter impedance and operating point and a data-driven method, enables the rapid prediction and generation of converter impedance values ​​across the entire operating domain using a small amount of impedance measurement data. This allows for further analysis, judgment, and prediction of the stability of grid-connected converters at different operating points. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the converter topology and its impedance measurement method.

[0043] Figure 2 This is a flowchart of the data-physical dual-drive neural network model proposed in this invention.

[0044] Figure 3 This is a diagram illustrating the effect of converter impedance prediction in this invention, wherein... Figure 3 (a) in the figure shows the results of the predicted amplitude and predicted phase of the positive sequence impedance as a function of the disturbance frequency at different operating points. Figure 3 Figure (b) shows the predicted amplitude and predicted phase of the negative sequence coupling impedance as a function of the disturbance frequency at different operating points. Figure 3 (c) in the figure shows the results of the predicted amplitude and predicted phase of the positive sequence coupling impedance as a function of the disturbance frequency at different operating points. Figure 3 (d) in the figure shows the results of the predicted amplitude and predicted phase of the negative sequence impedance as a function of the disturbance frequency at different operating points. Detailed Implementation

[0045] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. Technical features in the various embodiments of the present invention can be combined accordingly without mutual conflict.

[0046] like Figure 2 As shown, the present invention proposes a data-physical driven method for predicting the impedance of a converter across its entire operating domain, which specifically includes the following steps:

[0047] Step 1: Obtain the broadband impedance values ​​of the converter at several different operating points using existing converter impedance measurement techniques.

[0048] Existing converter impedance measurement techniques mainly include two methods: injecting disturbance power into the main circuit through main circuit disturbance and injecting disturbance signals into the sampling circuit through secondary side control loop disturbance (see CN118641837A). Using either method, the measured broadband impedance value of the converter is obtained at different voltage operating points, d-axis current operating points, and q-axis current operating points. The broadband impedance includes four different impedances: positive sequence impedance Zpp, negative sequence impedance Znn, positive sequence coupling impedance Zcp, and negative sequence coupling impedance Zcn.

[0049] Figure 1 This is a schematic diagram of two converter impedance measurement techniques. Figure 1 In the middle, V dc V is the DC side voltage of the converter. g Z is the grid voltage. g The equivalent impedance of the power grid. The perturbation signal is the sampled value of the injected current. The voltage sampling value disturbance signal is injected through the secondary control loop disturbance method. For the converter output current, L is the output voltage of the converter. 1、 C f、 L2 together form a third-order LCL filter, representing the inverter-side filter inductance, filter capacitor parameters, and grid-side filter inductance. Z pp For the positive sequence impedance of the converter, Z nn Negative sequence impedance, Z cp For positive sequence coupling impedance, Z cn is the negative-sequence coupling impedance, and s is the Laplace operator.

[0050] Step 2: Using the converter broadband impedance values ​​at different operating points obtained in Step 1 as the training dataset, construct the data-physical dual-drive neural network proposed in this invention. The inputs are the converter voltage operating point V, current operating points Id and Iq, and disturbance frequency fp. The output is the converter impedance value at that operating point. Then, the neural network is trained so that it can predict the output converter broadband impedance value based on the input operating point. After training, a validation set test and evaluation are performed.

[0051] Specifically, it includes the following sub-steps:

[0052] (2.1) Construct the training dataset and validation set.

[0053] Since the wideband impedance value is related to the converter voltage operating point V, d-axis current operating point Id, q-axis current operating point Iq, disturbance frequency fp, as well as control parameters and main circuit filter parameters, this invention uses a training dataset composed of the real and imaginary parts of V, Id, Iq, fp, and the measured wideband impedance value.

[0054] For each group of converters, the voltage operating point V, the current operating points Id and Iq, and k different disturbance frequencies fp{fp1, fp2, fp3, ... The dataset consists of the real part Z_re and the imaginary part Z_im of the corresponding wideband impedance value of the converter under fpk. The voltage operating point V needs to cover the voltage operating range of the converter grid connection point (PCC), which is generally 0.9~1.1 pu. The current operating points Id and Iq can be selected from any number of sets, and do not need to cover the entire current operating range of the converter.

[0055] The training dataset is uniformly normalized and divided into training dataset and validation set according to the ratio.

[0056] (2.2) Input the normalized converter voltage operating point V and different disturbance frequencies fp into the neural network, and train the neural network to generate four converter impedance internal parameters that are only related to the voltage operating point V and the disturbance frequency fp, denoted as a0, a1, b0, and b1. These four parameters are all complex numbers, and their real and imaginary parts are denoted as a0_re, a0_im, a1_re, a1_im, b0_re, b0_im, b1_re, and b1_im, respectively.

[0057] (2.3) Input the current operating points Id and Iq into the neural network, and let them, along with the above four converter impedance internal parameters, calculate two converter impedance internal functions using physical formulas (which are obtained by analyzing the general laws of existing converter impedance theoretical models). , In fact, the virtual parts are recorded separately as , , , As shown in the following formula:

[0058]

[0059] (2.4) Further, the internal function f of the two converter impedances is used. A f B The values ​​are calculated according to the physical formula (which is also obtained by analyzing the general laws of existing converter impedance theoretical models), yielding the real part Zpre_re and the imaginary part Zpre_im of the predicted converter impedance, as shown in the following equation:

[0060]

[0061] in, and For terms that are only related to the parameters of the main circuit filter, the specific expression can be obtained from the existing converter impedance theory model; real indicates taking the real part, and imag indicates taking the imaginary part. is the imaginary unit, and s is the Laplace operator.

[0062] When predicting four different impedances in a broadband impedance value and The two values ​​are different: for the positive sequence impedance Zpp, , This refers to the positive-sequence quantity of the main circuit filter; for the negative-sequence impedance Znn, , , which is the negative sequence quantity of the main circuit filter, where The reference frequency is typically 50Hz; for the positive sequence coupling impedance Zcp, , For negative sequence coupling impedance Zcn, , .

[0063] Substitute different perturbation frequencies fp into the same... and From the expression, the impedance corresponding to the desired impedance can be obtained. and The value of is used to further obtain the predicted value of the impedance.

[0064] (2.5) Based on the mean square error between the real part Zpre_re and the imaginary part Zpre_im of the calculated converter impedance prediction value and the real part Z_re and the imaginary part Z_im of the impedance value in the actual dataset, the loss function Loss is calculated as follows:

[0065]

[0066] In the formula, N is the number of running points in the training set, and k is the number of frequency points.

[0067] By training the neural network parameters in (2.2) with the aforementioned loss function Loss until the loss function converges or reaches the preset training rounds, the predicted real and imaginary impedance values ​​gradually match the true values ​​in the dataset.

[0068] (2.6) Evaluate the trained neural network model.

[0069] Based on the partitioned validation set data, the input converter voltage operating point V, current operating point Id, and Iq are used to calculate the converter impedance value at that operating point, and the determination coefficient R is calculated. 2 The two metrics, root mean square error (RMSE) and the coefficient of determination (R²) of the trained neural network on the validation set, are used to measure the performance of the neural network. 2 A value close to 1, while maintaining a low RMSE, indicates that the trained model has met the requirements for impedance prediction.

[0070] Step 3: Input the voltage operating point V, current operating points Id and Iq, and disturbance frequency fp in the entire operating domain of the converter. Then, use the trained data-physical dual-drive neural network to predict and generate the broadband impedance values ​​of the converter in the entire operating domain, including positive sequence impedance Zpp, negative sequence impedance Znn, positive sequence coupling impedance Zcp, and negative sequence coupling impedance Zcn. This provides a model basis for subsequent small-signal stability analysis.

[0071] To verify the effectiveness of the method proposed in this invention, further simulation verification was carried out.

[0072] This verification uses the secondary control loop disturbance method to obtain the measured values ​​of the converter's wideband impedance. Specific operating points were set as follows: Id = 5A, 10A, 15A (3 points, not needing to cover the entire operating range); Iq = -5A, 0A, 5A (3 points, not needing to cover the entire operating range); V = 0.8 pu, 0.9 pu, 1.0 pu, 1.1 pu, 1.2 pu (5 points, only V covers the entire operating range). A total of 45 sets of converter impedance values ​​were measured at these operating points. The disturbance frequency fp ranged from 10-300Hz, with a frequency interval of 10Hz.

[0073] The training dataset and validation dataset ratio is 4:1. The training dataset is input into the proposed data-physical dual-driven neural network model. The inputs are the converter voltage operating point V, current operating points Id and Iq, and disturbance frequency fp. The output is the impedance value of the converter at that operating point. Then, the neural network is trained so that it can predict the wideband impedance value of the output converter based on the input operating points V, Id, and Iq. After training, a validation set is used for testing and evaluation. The evaluation result is the coefficient of determination R. 2 =0.999901, and the root mean square error (RMSE) = 0.267020, indicating that the neural network model constructed in this invention has been well trained.

[0074] The trained neural network was tested on datasets other than the non-training and validation datasets, and the results are as follows: Figure 3 As shown, (a) represents the theoretical model of the predicted amplitude and predicted phase of the positive sequence impedance Zpp at different operating points as a function of the disturbance frequency; (b) represents the theoretical model of the predicted amplitude and predicted phase of the negative sequence coupling impedance Zcn at different operating points as a function of the disturbance frequency; (c) represents the theoretical model of the predicted amplitude and predicted phase of the positive sequence coupling impedance Zcp at different operating points as a function of the disturbance frequency; and (d) represents the theoretical model of the predicted amplitude and predicted phase of the negative sequence impedance Znn at different operating points as a function of the disturbance frequency.

[0075] Figure 3 Each curve represents the theoretical model of the predicted four impedance values ​​(including amplitude and phase) of the converter's wideband impedance at different operating points, varying with the disturbance frequency. The types of dots represent the converter impedance values ​​predicted by the data-physical dual-drive neural network: blue dots represent the wideband impedance prediction results at operating points Id=11, Iq=-7, V=1.06; green triangles represent the wideband impedance prediction results at operating points Id=25, Iq=-7, V=1.04; and purple triangles represent the wideband impedance prediction results at operating points Id=25, Iq=-7, V=1.04. The inverted triangle represents the broadband impedance prediction result at the operating point Id=18.2, Iq=3.5, V=0.95; the orange positive direction represents the broadband impedance prediction result at the operating point Id=18.2, Iq=-7, V=1.05; the red diamond represents the broadband impedance prediction result at the operating point Id=11, Iq=3.5, V=0.96; and the brown star represents the broadband impedance prediction result at the operating point Id=25, Iq=3.5, V=0.93.

[0076] The results show that the trained data-physical dual-drive neural network in this invention can accurately predict the impedance value of the converter at the operating point in both the non-training dataset and the validation dataset.

[0077] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A data-physics driven method for predicting the impedance of a converter across its entire operating domain, characterized in that, include: S1, under multiple voltage and current operating points, by applying disturbance signals of different disturbance frequencies to the converter, the measured values ​​of positive sequence impedance, negative sequence impedance, positive sequence coupling impedance and negative sequence coupling impedance in the converter are obtained. S2, input the voltage operating point and disturbance frequency into the neural network to obtain the converter impedance internal parameters and realize data driving; the converter impedance internal parameters include a first internal parameter, a second internal parameter, a third internal parameter and a fourth internal parameter, and each internal parameter includes a real part and an imaginary part; S3, input the current operating point into the neural network, combine the converter impedance internal parameters with the current operating point, calculate the converter impedance internal function, and further obtain the predicted values ​​of positive sequence impedance, negative sequence impedance, positive sequence coupling impedance and negative sequence coupling impedance to realize physical drive; In S3, the converter impedance internal function includes a first internal function and a second internal function; the process of calculating the converter impedance internal function by combining the converter impedance internal parameters with the current operating point is as follows: The real part of the first internal function is calculated by adding the real part of the second internal parameter to the product of the real part of the second internal parameter and the d-axis current operating point, and then subtracting the product of the imaginary part of the second internal parameter and the q-axis current. The imaginary part of the first internal function is calculated by adding the imaginary part of the first internal parameter to the product of the real part of the second internal parameter and the q-axis current operating point, and then adding the imaginary part of the second internal parameter to the product of the d-axis current operating point. The real part of the second internal function is calculated by adding the real part of the third internal parameter to the product of the real part of the fourth internal parameter and the d-axis current operating point, and then subtracting the product of the imaginary part of the fourth internal parameter and the q-axis current operating point. The imaginary part of the second internal function is calculated by adding the imaginary part of the third internal parameter to the product of the real part of the fourth internal parameter and the q-axis current operating point, and then adding the imaginary part of the fourth internal parameter to the product of the d-axis current operating point. In S3, the formula for calculating the impedance prediction value is as follows: ； Where real represents the real part and imag represents the imaginary part. The imaginary unit, and This is a term that is only related to the parameters of the main circuit filter. When calculating different impedances, it is necessary to... and Replace with different values; , These are the real and imaginary parts of the first internal function, respectively. , These are the real and imaginary parts of the second inner function, respectively. , , respectively, are the real and imaginary parts of the predicted converter impedance, and s is the Laplace operator; S4. Based on the measured and predicted values, calculate the loss function and iteratively update the neural network parameters until convergence or the maximum number of iterations is reached, thus obtaining the trained neural network. S5 uses the trained neural network to obtain the predicted values ​​of the positive sequence impedance, negative sequence impedance, positive sequence coupling impedance, and negative sequence coupling impedance of the converter at each voltage operating point, current operating point, and disturbance frequency in the entire operating domain.

2. The method for predicting the impedance of a converter across its entire operating domain based on data-physics driven technology according to claim 1, characterized in that, In S1, the measured wideband impedance of the converter is obtained by either the main circuit disturbance method or the secondary control loop disturbance method.

3. The method for predicting the impedance of a converter across its entire operating domain based on data-physics driven technology as described in claim 1, characterized in that, In S1, the selected voltage operating point is required to cover the complete voltage operating range of the converter.

4. The method for predicting the impedance of a converter across its entire operating domain based on data-physics driven technology according to claim 1, characterized in that, In S1, the current operating point includes the d-axis current operating point and the q-axis current operating point.

5. The method for predicting the impedance of a converter across its entire operating domain based on data-physics driven technology according to claim 1, characterized in that, When calculating different impedances and The possible values ​​are: For positive sequence impedance , ; For negative sequence impedance , ; For positive sequence coupling impedance , ; For negative sequence coupling impedance , ; in, As the reference frequency, The frequency is the disturbance frequency.

6. The method for predicting the impedance of a converter across its entire operating domain based on data-physics driven technology according to claim 1, characterized in that, In S3, the loss function is calculated by averaging the sum of the squares of the differences between the real and imaginary parts of all predicted and measured values ​​of impedances.

7. The method for predicting the impedance of a converter across its entire operating domain based on data-physics driven technology according to claim 1, characterized in that, In step S4, after obtaining the trained neural network, the trained neural network and the validation dataset are used to obtain the predicted values ​​of the positive-sequence impedance, negative-sequence impedance, positive-sequence coupling impedance, and negative-sequence coupling impedance in the validation dataset. The validation dataset includes the voltage operating point, the current operating point, the disturbance frequency, and the corresponding measured values ​​of the positive-sequence impedance, negative-sequence impedance, positive-sequence coupling impedance, and negative-sequence coupling impedance. The determination coefficient and root mean square error are calculated by combining the measured and predicted values ​​of all impedances. If the determination coefficient is higher than the determination threshold and the root mean square error is lower than the root mean square threshold, the training is terminated and the process proceeds to S5. Otherwise, the training is restarted.

Images