Impedance Identification Method and System Based on Physics-Informed Neural Network

Abstract

The present invention discloses an impedance identification method and system based on a physics-informed neural network. First, the working point information of the MMC after preprocessing and the original impedance data are obtained; then, based on the working point information, a corresponding control physics information is selected according to the purpose of impedance identification to train a first neural network, obtaining an initial physics-informed neural network model and outputting training control physics information; then, based on the working point information and the training control physics information, a second neural network is trained to output training impedance data; the training impedance data is compared with the original impedance data to obtain a training error, and the second neural network is adjusted according to the error to obtain an impedance identification model; finally, impedance identification is completed using the impedance identification model. The present invention can improve the training accuracy of the neural network impedance modeling method, significantly reducing the error of impedance identification.

Description

Technical Field

[0001] The present invention relates to the field of power electronics technology, and in particular to an impedance identification method, system, device and medium based on a physics-informed neural network. Background Art

[0002] With the rapid development of the new power system, while power electronic equipment is widely used in the power system, oscillation events dominated by power electronic equipment occur frequently, seriously affecting the safe and stable operation of the power system. The stability analysis method based on the impedance transfer function can be used to solve the oscillation problem of the power system with a high proportion of power electronic equipment. Among them, the impedance modeling method, as an important method for the stability analysis of the power grid and the converter, has the characteristics of intuitive physical meaning and simple analysis.

[0003] At present, a few existing technologies carry out impedance identification of power electronic equipment based on the generalization ability of artificial neural networks. Among them, powerful non-linear processing tools such as neural networks can be used to train according to existing data, and then a neural network impedance identification model under multiple operating points can be obtained. However, these traditional neural network impedance identification methods have large errors at impedance mutation points for impedance identification of modular multilevel converters (MMCs) with many resonance peaks, resulting in low accuracy of MMC impedance identification. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an impedance identification method, system, device and medium based on a physics-informed neural network, which can enhance the connection between the input operating point information and the output impedance information and improve the accuracy of impedance identification.

[0005] To solve the above technical problems, an embodiment of the present invention provides an impedance identification method based on a physics-informed neural network, including:

[0006] Obtain a preprocessed training data set; wherein, the training data set includes the operating point information of the modular multilevel converter and the original impedance data;

[0007] Based on the operating point information, select corresponding control physical information according to the purpose of impedance identification to train a first neural network, obtain an initial physical information neural network model, and output training control physical information;

[0008] Based on the operating point information and the training control physical information, train a second neural network, output training impedance data, compare the training impedance data with the original impedance data to obtain a second training error. If the second training error is greater than or equal to a set threshold, adjust the second neural network until the error is less than the set threshold to obtain an impedance identification model;

[0009] Input the operating point information and the selected control physical information into the impedance identification model to obtain the impedance output at the corresponding operating point, thereby completing the impedance identification of the modular multilevel converter.

[0010] Optionally, before obtaining the preprocessed training data set, it further includes:

[0011] Obtain the operating point information of the modular multilevel converter and the original impedance data at different frequencies through a simulation model or physical measurement to obtain a training data set.

[0012] Optionally, the preprocessing includes phase extension and normalization of the training data set.

[0013] Furthermore, the control physical information includes differential-mode side control physical information and common-mode side control physical information;

[0014] The differential-mode side control physical information includes the physical information C of differential-mode control and current dmi_p and the physical information C of differential-mode control and voltage dmu_p ;

[0015] The common-mode side control physical information includes the physical information C of common-mode control and current cmi_p and the physical information C of common-mode control and voltage cmu_p ;

[0016] Among them, the definition formula of the physical information of differential-mode control and current is:

[0017]

[0018] The definition formula of the physical information of differential-mode control and voltage is:

[0019]

[0020] The definition formula of the physical information of common-mode control and current is:

[0021]

[0022] The definition formula of the physical information of common-mode control and voltage is:

[0023]

[0024] In each definition formula, M dm_p,f＝fp and M cm_p,f＝fp respectively represent the perturbations obtained at the differential-mode side control and the common-mode side control when a three-phase positive-sequence perturbation with frequency f p is injected from the AC side of the MMC to the port; I p,f＝fp represents the current perturbation.

[0025] Further, selecting corresponding control physical information according to the purpose of impedance identification includes:

[0026] If identifying the AC impedance of a modular multilevel converter, select differential-mode side control physical information;

[0027] If identifying the DC impedance of a modular multilevel converter, select common-mode side control physical information.

[0028] Further, based on the operating point information, training a first neural network by selecting corresponding control physical information according to the purpose of impedance identification to obtain an initial physical information neural network model and outputting training control physical information includes:

[0029] Calculating original control physical information according to the operating point information;

[0030] Training a first neural network by selecting corresponding control physical information according to the purpose of impedance identification and outputting training control physical information;

[0031] Comparing the training control physical information with the original control physical information to obtain a first training error; if the first training error is greater than or equal to a set threshold, adjust the first neural network until the error is less than the set threshold to obtain an initial physical information neural network model and output the final training control physical information.

[0032] Preferably, for the impedance identification method based on a physical information neural network, after completing the impedance identification of the modular multilevel converter, it further includes:

[0033] Analyzing the stability of the interaction between the modular multilevel converter and the power grid using the impedance identification result.

[0034] To solve the above technical problems, an embodiment of the present invention further provides an impedance identification system based on a physical information neural network, including:

[0035] A training data acquisition module for acquiring a preprocessed training data set; wherein, the training data set includes the operating point information of a modular multilevel converter and original impedance data;

[0036] A first training module for training a first neural network by selecting corresponding control physical information according to the purpose of impedance identification based on the operating point information to obtain an initial physical information neural network model and outputting training control physical information;

[0037] A second training module, configured to train a second neural network based on the operating point information and the training control physical information, output training impedance data, compare the training impedance data with the original impedance data to obtain a second training error, and if the second training error is greater than or equal to a set threshold, adjust the second neural network until the error is less than the set threshold, thereby obtaining an impedance identification model;

[0038] An impedance identification module, configured to input the operating point information and the selected control physical information into the impedance identification model to obtain an impedance output at the corresponding operating point, thereby completing the impedance identification of the modular multilevel converter.

[0039] An embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the impedance identification method based on a neural network of physical information as described in any one of the above is implemented.

[0040] An embodiment of the present invention further provides a computer-readable storage medium, which includes a stored computer program. When the computer program runs, the device where the computer-readable storage medium is located is controlled to execute the impedance identification method based on a neural network of physical information as described in any one of the above.

[0041] Compared with the prior art, an impedance identification method, system, device, and medium based on a neural network of physical information provided by an embodiment of the present invention first obtain the operating point information and the original impedance data of a preprocessed modular multilevel converter; then, according to the purpose of impedance identification, select corresponding control physical information, train a first neural network based on the operating point information and the selected control physical information to obtain an initial physical information neural network model, and output training control physical information; then train a second neural network based on the operating point information and the training control physical information, and output training impedance data; compare the training impedance data with the original impedance data to obtain a training error, and adjust the second neural network according to the error to obtain an impedance identification model; finally, use the impedance identification model to complete impedance identification. By combining the control physical information of a power electronic device, the present invention can enrich the input data set of the neural network, thereby enhancing the relationship between the input data set and the output data volume of the neural network, obtaining a better impedance fitting effect, and solving the problem that in the impedance identification method based on a conventional neural network, there are obvious errors in both the impedance amplitude and phase near the resonance peak when dealing with power electronic devices with multi-resonance peak characteristics; through two trainings of the physical information neural network, compared with the conventional neural network training method, the training accuracy of the neural network impedance modeling method is improved, and the error of impedance identification is significantly reduced. Description of the Drawings

[0042] To more clearly illustrate the technical solution of the present invention, the accompanying drawings to be used in the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0043] Figure 1 is a flowchart of an impedance identification method based on a physics-informed neural network provided by an embodiment of the present invention;

[0044] Figure 2 is another flowchart of an impedance identification method based on a physics-informed neural network provided by an embodiment of the present invention;

[0045] Figure 3 is the training process of the physics-informed neural network of an impedance identification method based on a physics-informed neural network provided by an embodiment of the present invention;

[0046] Figure 4 is a schematic diagram of the structure of the physics-informed neural network of an impedance identification method based on a physics-informed neural network provided by an embodiment of the present invention;

[0047] Figure 5 is the training result and error graph of the physics-informed neural network of an impedance identification method based on a physics-informed neural network provided by an embodiment of the present invention;

[0048] Figure 6 is a comparison graph of the results of the method of the embodiment of the present invention and the neural network impedance identification using the traditional method provided by an embodiment of the present invention;

[0049] Figure 7 The error comparison graph of the method of the embodiment of the present invention and the neural network impedance identification using the traditional method provided by an embodiment of the present invention;

[0050] Figure 8 The Nyquist curve and the simulation time-domain waveform graph made from the impedance identification results of an impedance identification method based on a physics-informed neural network provided by an embodiment of the present invention;

[0051] Figure 9 is a structural block diagram of an impedance identification system based on a physics-informed neural network provided by an embodiment of the present invention;

[0052] Figure 10 is a structural block diagram of a terminal device provided by an embodiment of the present invention. Specific embodiments

[0053] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0054] See Figure 1 , which is a flowchart of an impedance identification method based on a physical information neural network provided by an embodiment of the present invention. The method includes steps S1 to S5:

[0055] S1. Obtain a preprocessed training data set; wherein, the training data set includes the operating point information of the modular multilevel converter and the original impedance data;

[0056] Specifically, before obtaining the preprocessed training data set, it further includes:

[0057] Obtain the operating point information of the modular multilevel converter and the original impedance data at different frequencies through a simulation model or physical measurement to obtain a training data set.

[0058] Furthermore, the preprocessing includes phase expansion and normalization of the training data set.

[0059] It should be noted that the impedance, operating point information (generally including the d-axis voltage v d , q-axis voltage v q , d-axis current i d , q-axis current i q , frequency f, etc.) of the MMC (modular multilevel converter) at different frequencies can be obtained through a simulation model or physical measurement, so as to preprocess the data. Mainly, the input data is preprocessed through phase expansion and normalization. Among them, phase expansion mainly aims at the problem that the impedance phase angle is limited to (-π, π) after using the angle function, and the impedance phase is offset by 2π to expand the phase; normalization is to solve the dimensional difference between the input data and ensure that the data is within the range of (-1, 1).

[0060] It is worth noting that the operating point information describes the performance characteristics of the converter under different operating states. These data can provide detailed information about the behavior and response of the converter. Once these data are obtained, the training data set can be constructed, the phase of the training data set can be expanded to increase the dimension of the data; normalization processing is performed to make the data comparable under different scales, which helps to ensure that the model can learn and generalize better during the training process.

[0061] S2. Based on the working point information, select the corresponding control physical information according to the purpose of impedance identification to train the first neural network, obtain the initial physical information neural network model, and output the training control physical information;

[0062] Specifically, the control physical information includes differential-mode side control physical information and common-mode side control physical information;

[0063] The differential-mode side control physical information includes differential-mode control and current physical information C dmi_p and differential-mode control and voltage physical information C dmu_p ;

[0064] The common-mode side control physical information includes common-mode control and current physical information C cmi_p and common-mode control and voltage physical information C cmu_p ;

[0065] Among them, the definition formula of the differential-mode control and current physical information is:

[0066]

[0067] The definition formula of the differential-mode control and voltage physical information is:

[0068]

[0069] The definition formula of the common-mode control and current physical information is:

[0070]

[0071] The definition formula of the common-mode control and voltage physical information is:

[0072]

[0073] In each definition formula, M dm_p,f＝fp and M cm_p,f＝fp respectively represent the perturbations obtained at the differential-mode side control and the common-mode side control when a three-phase positive-sequence perturbation with frequency f p is injected from the MMC AC side to the port; I p,f＝fp represents the current perturbation.

[0074] It should be noted that according to the characteristics of the MMC, the physical information can be divided into differential-mode side control physical information and common-mode side control physical information:

[0075] For the differential-mode side of the MMC, when a three-phase positive-sequence perturbation with a specific frequency f p is injected from the MMC AC side to the port, the differential-mode control can obtain the perturbation M dm_p,f＝fp , combined with the current perturbation I p,f＝fp, by analogy with the definition of impedance, the relationship between differential-mode control and current can be defined as C dmi_p , as shown in Equation (1), representing the physical information of differential-mode control and current, C dmi_p also includes amplitude and phase angle data (using bold variables (e.g., M dm_p , C dmi_p etc.) to represent phasors, which have both amplitude and phase angle information); the relationship between differential-mode control and voltage is defined as C dmu_p , as shown in Equation (2), representing the physical information of differential-mode control and voltage, C dmu_p also includes amplitude and phase angle data.

[0076] For the common-mode side of the MMC, when a single-phase disturbance with a specific frequency f p is injected from the MMC DC bus to the port, the common-mode control can obtain the disturbance M cm_p,f＝fp , combined with the current disturbance I p,f＝fp , by analogy with the definition of impedance, the relationship between common-mode control and current can be defined as C cmi_p , as shown in Equation (3), representing the physical information C of common-mode control and current cmi_p also includes amplitude and phase angle data; the relationship between common-mode control and voltage is defined as C cmu_p , as shown in Equation (4), representing the physical information of common-mode control and voltage, C cmu_p also includes amplitude and phase angle data.

[0077] Further, the corresponding control physical information is selected according to the purpose of impedance identification, including:

[0078] If identifying the AC impedance of the modular multilevel converter by impedance identification, select the differential-mode side control physical information;

[0079] If identifying the DC impedance of the modular multilevel converter by impedance identification, select the common-mode side control physical information.

[0080] It should be noted that the special feature of the MMC is that its AC and DC ports have two control degrees of freedom, namely differential-mode and common-mode. It is necessary to fully consider the AC and DC control characteristics and AC and DC impedance characteristics of the MMC. Therefore, according to the equivalent AC and DC loop model of the MMC, since the differential-mode side controls the AC side, when choosing to identify the AC impedance of the MMC, the control physical information C dmi_p and C dmu_p of the differential-mode side can be selected, while ignoring the control physical information of the common-mode side. Similarly, since the common-mode side controls the DC side, when choosing to identify the DC impedance of the MMC, the control physical information C cmi_p and C cmu_p of the common-mode side can be selected, while ignoring the control physical information of the differential-mode side.

[0081] S3. Train a second neural network based on the working point information and the training control physical information, output training impedance data, compare the training impedance data with the original impedance data to obtain a second training error. If the second training error is greater than or equal to the set threshold, adjust the second neural network until the error is less than the set threshold to obtain an impedance identification model.

[0082] S5. Input the working point information and the selected control physical information into the impedance identification model to obtain the impedance output at the corresponding working point, and complete the impedance identification of the modular multilevel converter.

[0083] Compared with the prior art, the embodiment of the present invention provides an impedance identification method based on a physics-informed neural network. By combining the control physical information of power electronic equipment, it can enrich the input data set of the neural network, thereby enhancing the connection between the input data set and the output data volume of the neural network, obtaining a better impedance fitting effect, and solving the problem that there are obvious errors in both the impedance amplitude and phase near the resonance peak when using the impedance identification method based on a conventional neural network to process power electronic equipment with multi-resonant peak characteristics; and through two trainings of the physics-informed neural network, compared with the conventional neural network training method, it improves the training accuracy of the neural network impedance modeling method, significantly reducing the error of impedance identification.

[0084] In an alternative embodiment, based on the working point information, select the corresponding control physical information according to the purpose of impedance identification to train a first neural network to obtain an initial physics-informed neural network model and output training control physical information, including:

[0085] Calculate the original control physical information according to the working point information;

[0086] Select the corresponding control physical information according to the purpose of impedance identification to train the first neural network and output training control physical information;

[0087] Compare the training control physical information with the original control physical information to obtain a first training error; if the first training error is greater than or equal to the set threshold, adjust the first neural network until the error is less than the set threshold to obtain an initial physics-informed neural network model and output the final training control physical information.

[0088] Based on any of the above embodiments, after completing the impedance identification of the modular multilevel converter, the impedance identification method based on a physics-informed neural network further includes:

[0089] Use the impedance identification result to analyze the stability of the interaction between the modular multilevel converter and the power grid.

[0090] It should be noted that in the embodiments of the present invention, by adding the physical information of the MMC controller as the operating point, compared with the conventional neural network training method, the error of MMC impedance identification can be greatly reduced, thus providing a basis for stability analysis.

[0091] To describe the technical solutions in the embodiments of the present invention more clearly, refer to Figure 2 , which is another flowchart of a method for impedance identification based on a physics-informed neural network provided by the embodiments of the present invention. As Figure 2 shown, the method includes the following steps:

[0092] Step 1: Obtain data. Specifically, the sequence impedance and operating point information of the MMC at different frequencies can be obtained by frequency sweeping (which can be used for subsequent calculation of the differential-mode side and common-mode side control physical information of the MMC).

[0093] Step 2: Preprocess the data (mainly phase expansion and normalization processing) to obtain a training data set.

[0094] Step 3: Select control physical information. According to the characteristics of the MMC, the physical information can be divided into differential-mode side control physical information and common-mode side control physical information. According to the purpose of impedance identification, if the AC impedance of the MMC is to be identified, the differential-mode side control physical information C dmi_p and C dmu_p are selected. If the DC impedance of the MMC is to be identified, the common-mode side control physical information C cmi_p and C cmu_p are selected.

[0095] Step 4: First training. Use the operating point information as the input and the selected control physical information as the output to train the neural network to obtain a control information neural network model. After the training is completed, it is necessary to determine whether the error meets the requirements (specifically, defined as the error being less than 5%). When the error meets the requirements, the required initial control physical information neural network model is obtained. Otherwise, adjust the neural network and repeat the above steps.

[0096] Step 5: Second training. Conduct impedance identification on the AC side or DC side of the MMC. Compare the training impedance data generated by the neural network with the original impedance data to obtain a training error (specifically, defined as the error being less than 5%). When the error meets the requirements, the required neural network model (impedance identification model) is obtained, thus completing impedance identification. Otherwise, adjust the neural network and repeat the above steps.

[0097] Exemplarily, the training process of the physics-informed neural network is divided into two trainings. The specific process is as Figure 3As shown. The input of the first training of the neural network is the operating point information, and the output is the amplitude and phase angle of the control physical information selected in the above step three. Taking the neural network input as the operating point (d-axis voltage v d , q-axis voltage v q , d-axis current i d , q-axis current i q , frequency f), and the output is the amplitude and phase angle of the control physical information C dmi_p and C dmu_p as an example, the neural network established in the first training is a 5-input 4-output system. The second training is based on the control information generated after the first training. The input of the neural network is the operating point information and the control information generated in the first training, and the output of the neural network is the amplitude and phase angle of the sequence impedance. Taking the neural network input as the operating point (d-axis voltage v d , q-axis voltage v q , d-axis current i d , q-axis current i q , frequency f), the amplitude and phase angle of the differential-mode side control physical information C dmi_p and C dmu_p as an example, and the output is the amplitude and phase angle of the sequence impedance Z p , the neural network established in the second training is a 9-input 2-output system.

[0098] In specific implementation, the embodiment of the present invention also verifies the above impedance identification method based on the physical information neural network by building a simulation model in Matlab / Simulink. Taking the AC side of the MMC as an example, a three-phase positive-sequence voltage disturbance with a frequency range of 1 to 1000 Hz and an interval of 2 Hz is injected into the AC side of the MMC, and the operating point is the active power P = 30 MW to P = 50 MW, with an interval of 2.5 MW. When training, the BP neural network is used, which includes an input layer, an output layer and multiple hidden layers. The layers are fully connected, and each neuron in each layer is not connected to each other. The structure is as Figure 4 shown. For the physical information neural network mentioned in any of the above embodiments, when the input and output data are given, a definite neural network model can be obtained through the setting of the number of layers m, the number of neurons in each layer n, and the neurons. At the same time, the embodiment of the present invention also uses a traditional neural network for impedance identification for comparison. In the embodiment of the present invention, to ensure the consistency of the neural network scale, the hidden layers of both neural networks are 3, and the corresponding number of neurons in each layer is 10\12\15 respectively.

[0099] As Figures 5 - 7 shown, it is the training result and error of the physical information neural network in the embodiment of the present invention. Among them, the control physical information serves as an intermediate bridge for training the impedance, and after the training of the first layer of the neural network, the output has a good fitting result. Taking the identification of the AC impedance of the MMC as an example, asFigure 5 As shown, the control physical information C on the differential-mode side of the output dmi_p The maximum amplitude error is 0.2230 dB, and the maximum phase angle error is 0.2719°. C dmu_p The maximum amplitude error is 0.1442 dB, and the maximum phase angle error is 0.2308°. The error values are small, indicating that a good fitting effect can be achieved.

[0100] Furthermore, in the embodiment of the present invention, the fitted control physical information and the operating point information are used as the input of the second-layer neural network to train the system impedance. See Figure 6 , which are the results of the impedance identification of the neural network by the method of the present invention and the traditional method in the embodiment of the present invention; see Figure 7 , which are the errors of the impedance identification of the neural network by the method of the present invention and the traditional method in the embodiment of the present invention. As Figures 6 - 7 shown, at this time, the maximum impedance amplitude error of the method of the present invention is 0.2489 dB, and the maximum impedance phase angle error is 0.3280°; while using the traditional neural network training method, since the impedance of the MMC is prone to mutation in the low-frequency band, resulting in poor actual training effect, at this time, the maximum impedance amplitude error is 1.0794 dB, and the maximum impedance phase angle error is 5.0632°. At this time, the maximum amplitude error of the impedance identification result based on the joint drive of control physical information and data is 0.2489 dB, and the maximum impedance phase angle error is 0.3280°. The errors are all less than 0.5%, and the maximum errors of the amplitude and phase are reduced by 76.9% and 93.5% compared with the conventional neural network. Using the physical information neural network training method in the embodiment of the present invention significantly reduces the impedance training error, verifying the effectiveness of the method of the present invention.

[0101] Preferably, after completing the neural network impedance identification, the impedance theory can be used to judge the system stability. That is, when the grid short-circuit ratio is 1.85 and the MMC sends out power at the rated power, the stability of the interaction between the MMC and the grid is analyzed using the neural network impedance identification result. See Figure 8 , which are the Nyquist curves and simulation time-domain waveforms verified by the impedance results identified by the impedance identification method based on the physical information neural network provided in the embodiment of the present invention. As Figure 8 shown, at 3 s of the simulation time, when the system short-circuit ratio is switched from 2.09 to 1.85, the system starts to oscillate, which also verifies the effectiveness of the impedance identification method based on the physical information neural network of the present invention.

[0102] Based on the above method item embodiments, the present invention correspondingly provides system item embodiments.

[0103] See Figure 9, which is a structural block diagram of an impedance identification system based on a physics-informed neural network provided by an embodiment of the present invention. The system includes:

[0104] A training data acquisition module 21, configured to acquire a preprocessed training data set. Among them, the training data set includes the operating point information of the modular multilevel converter and the original impedance data.

[0105] A first training module 22, configured to select corresponding control physics information according to the purpose of impedance identification based on the operating point information to train a first neural network, obtain an initial physics-informed neural network model, and output training control physics information.

[0106] A second training module 23, configured to train a second neural network based on the operating point information and the training control physics information, output training impedance data, compare the training impedance data with the original impedance data to obtain a second training error. If the second training error is greater than or equal to a set threshold, adjust the second neural network until the error is less than the set threshold to obtain an impedance identification model.

[0107] An impedance identification module 24, configured to input the operating point information and the selected control physics information into the impedance identification model to obtain an impedance output at the corresponding operating point, and complete the impedance identification of the modular multilevel converter.

[0108] Preferably, the impedance identification system based on the physics-informed neural network further includes a data acquisition module, configured to obtain the operating point information of the modular multilevel converter and the original impedance data at different frequencies through a simulation model or physical measurement to obtain a training data set.

[0109] Preferably, the first training module 22 includes a control physics information selection unit, configured to:

[0110] If identifying the AC impedance of the modular multilevel converter, select differential-mode side control physics information.

[0111] If identifying the DC impedance of the modular multilevel converter, select common-mode side control physics information.

[0112] Preferably, the first training module 22 further includes a first training unit, configured to:

[0113] Calculate the original control physics information according to the operating point information.

[0114] Select corresponding control physics information according to the purpose of impedance identification to train the first neural network and output training control physics information.

[0115] Compare the training control physical information with the original control physical information to obtain a first training error; if the first training error is greater than or equal to a set threshold, adjust the first neural network until the error is less than the set threshold to obtain an initial physical information neural network model, and output the final training control physical information.

[0116] Preferably, the impedance identification system based on the physics-informed neural network further includes a stability analysis module for:

[0117] Analyze the stability of the interaction between the modular multilevel converter and the power grid by using the impedance identification result.

[0118] It should be noted that the impedance identification system based on the physics-informed neural network provided in the embodiments of the present invention is used to execute all the process steps of the impedance identification method based on the physics-informed neural network in the above embodiments. The working principles and beneficial effects of the two correspond one by one, so they will not be elaborated here.

[0119] The embodiments of the present invention also provide a terminal device, such as Figure 10 shown, which is a structural block diagram of a preferred embodiment of a terminal device provided by the present invention. The terminal device includes a processor 31, a memory 32, and a computer program stored in the memory 32 and configured to be executed by the processor 31. When the processor 31 executes the computer program, it implements the impedance identification method based on the physics-informed neural network described in any of the above embodiments.

[0120] In addition, the embodiments of the present invention also provide a computer-readable storage medium. The computer-readable storage medium includes a stored computer program. When the computer program runs, it controls the device where the computer-readable storage medium is located to execute the impedance identification method based on the physics-informed neural network described in any of the above embodiments.

[0121] When the processor 31 executes the computer program, it implements the steps in the embodiments of the above-mentioned impedance identification method based on the physics-informed neural network, such as Figure 1 all the steps of the impedance identification method based on the physics-informed neural network shown. Or, when the processor 31 executes the computer program, it implements the functions of each module in the embodiments of the above-mentioned impedance identification system based on the physics-informed neural network, such as Figure 9 the functions of each module of the impedance identification system based on the physics-informed neural network shown.

[0122] Preferably, the computer program may be divided into one or more modules / units, which are stored in the memory 32 and executed by the processor 31 to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program in the terminal device.

[0123] The processor 31 may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or the processor 31 may also be any conventional processor. The processor 31 is the control center of the terminal device and connects various parts of the terminal device through various interfaces and circuits.

[0124] The memory 32 mainly includes a program storage area and a data storage area. Among them, the program storage area may store an operating system, application programs required for at least one function, etc., and the data storage area may store relevant data, etc. In addition, the memory 32 may be a high-speed random access memory, or may also be a non-volatile memory, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc., or the memory 32 may also be other volatile solid-state storage devices.

[0125] It should be noted that the above terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that Figure 10 the structural block diagram shown is only an example of the structure of the above terminal device and does not constitute a limitation on the structure of the above terminal device. The above terminal device may include more or fewer components than shown, or combine certain components, or have different components.

[0126] In summary, for an impedance identification method, system, device and medium based on a physics-informed neural network provided by an embodiment of the present invention, the working point information of a modular multilevel converter and original impedance data after preprocessing are first obtained; then, according to the purpose of impedance identification, corresponding control physics information is selected, and a neural network is trained based on the working point information and the selected control physics information to obtain an initial physics-informed neural network model, and training control physics information is output; then, the initial physics-informed neural network model is trained based on the working point information and the training control physics information to output training impedance data; the training impedance data is compared with the original impedance data to obtain a training error, and the initial physics-informed neural network model is adjusted according to the error to obtain an impedance identification model; finally, impedance identification is completed using the impedance identification model. By combining the control physics information of power electronic equipment, the present invention can enrich the input data set of the neural network, thereby enhancing the connection between the input data set and the output data volume of the neural network, obtaining a better impedance fitting effect, and solving the problem that there are obvious errors in the impedance amplitude and phase near the resonance peak when the impedance identification method based on a conventional neural network processes power electronic equipment with multi-resonant peak characteristics; through two trainings of the physics-informed neural network, compared with the conventional neural network training method, the training accuracy of the neural network impedance modeling method is improved, and the error of impedance identification is greatly reduced.

[0127] The above is the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements are also regarded as the protection scope of the present invention.

Claims

1. An impedance identification method based on physical information neural network, characterized in that: include: Acquire a preprocessed training data set; wherein the training data set includes operating point information and original impedance data of the modular multilevel converter; Based on the working point information, corresponding control physical information is selected according to the purpose of impedance identification to train the first neural network, an initial physical information neural network model is obtained, and the training control physical information is output; Training a second neural network based on the working point information and the training control physical information, outputting training impedance data, comparing the training impedance data with the original impedance data to obtain a second training error, and if the second training error is greater than or equal to a set threshold, adjusting the second neural network until the error is less than the set threshold, thereby obtaining an impedance identification model; Inputting the operating point information and the selected control physical information into the impedance identification model, obtaining the impedance output at the corresponding operating point, and completing the impedance identification of the modular multilevel converter; Wherein, the control physical information includes differential mode side control physical information and common mode side control physical information; The differential mode side control physical information includes the physical information C of the differential mode control and the current dmi_p and differential mode control and voltage physical information C dmu_p ; The common mode side control physical information includes common mode control and current physical information C cmi_p and common mode control and voltage physical information C cmu_p ; The physical information definition of the differential mode control and current is as follows: The physical information definition of the differential mode control and voltage is: The common mode control and the physical information definition of the current are: The physical information definition of the common mode control and voltage is: In each definition, M dm_p,f＝fp and M cm_p,f＝fp They represent the frequency f injected from the MMC AC side port respectively. p When there is a three-phase positive sequence disturbance, the disturbances obtained at the differential mode side control and the common mode side control; I p,f＝fp Indicates current disturbance.

2. The impedance identification method based on physical information neural network according to claim 1, characterized in that: Before obtaining the preprocessed training data set, the method further includes: The operating point information of the modular multilevel converter and the original impedance data at different frequencies are obtained through simulation models or physical measurements to obtain a training data set.

3. The impedance identification method based on physical information neural network according to claim 1, characterized in that: The preprocessing includes phase expansion and normalization of the training data set.

4. The impedance identification method based on physical information neural network according to claim 1, characterized in that: The selecting corresponding control physical information according to the purpose of impedance identification includes: If the impedance identifies the AC impedance of the modular multilevel converter, the differential mode side control physical information is selected; If the impedance identifies the DC impedance of the modular multilevel converter, the common mode side control physical information is selected.

5. The impedance identification method based on physical information neural network according to claim 4, characterized in that: The method of selecting corresponding control physical information to train the first neural network based on the working point information according to the purpose of impedance identification, obtaining an initial physical information neural network model, and outputting training control physical information includes: Calculate original control physical information according to the working point information; Selecting corresponding control physical information according to the purpose of impedance identification to train the first neural network, and outputting the training control physical information; The training control physical information is compared with the original control physical information to obtain a first training error; if the first training error is greater than or equal to a set threshold, the first neural network is adjusted until the error is less than the set threshold, an initial physical information neural network model is obtained, and the final training control physical information is output.

6. The impedance identification method based on physical information neural network according to claim 1, characterized in that: After completing the impedance identification of the modular multilevel converter, the method further includes: The impedance identification result is used to analyze the stability of the interaction between the modular multilevel converter and the power grid.

7. An impedance identification system based on physical information neural network, characterized in that: include: A training data acquisition module, used to acquire a preprocessed training data set; wherein the training data set includes operating point information and original impedance data of the modular multi-level converter; A first training module is used to select corresponding control physical information to train a first neural network based on the working point information and the purpose of impedance identification, obtain an initial physical information neural network model, and output training control physical information; a second training module, for training a second neural network based on the working point information and the training control physical information, outputting training impedance data, comparing the training impedance data with the original impedance data to obtain a second training error, and if the second training error is greater than or equal to a set threshold, adjusting the second neural network until the error is less than the set threshold, thereby obtaining an impedance identification model; An impedance identification module, used for inputting the operating point information and the selected control physical information into the impedance identification model, obtaining the impedance output under the corresponding operating point, and completing the impedance identification of the modular multilevel converter; Wherein, the control physical information includes differential mode side control physical information and common mode side control physical information; The differential mode side control physical information includes the physical information C of the differential mode control and the current dmi_p and differential mode control and voltage physical information C dmu_p ; The common mode side control physical information includes common mode control and current physical information C cmi_p and common mode control and voltage physical information C cmu_p ; The physical information definition of the differential mode control and current is as follows: The physical information definition of the differential mode control and voltage is: The common mode control and the physical information definition of the current are: The physical information definition of the common mode control and voltage is: In each definition, M dm_p,f＝fp and M cm_p,f＝fp They represent the frequency f injected from the MMC AC side port respectively. p When there is a three-phase positive sequence disturbance, the disturbances obtained at the differential mode side control and the common mode side control; I p,f＝fp Indicates current disturbance.

8. A terminal device, characterized in that: It comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements the impedance identification method based on physical information neural network as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the impedance identification method based on physical information neural network as described in any one of claims 1 to 6.

Images