Virtual synchronous machine inertia and damping parameter identification method based on pinn

Abstract

The application belongs to the technical field of power system analysis and control, and specifically discloses a virtual synchronous machine inertia and damping parameter identification method based on PINN. The application first carries out time alignment and dimensionless processing on the angular velocity and active power imbalance of the virtual synchronous machine in the power disturbance period to obtain a dimensionless value; then a neural network is constructed with time as input, the dimensionless predicted value of the angular velocity and active power imbalance as output, and the inertia time constant and damping coefficient as trainable parameters; a physical loss is constructed based on the residual of the dimensionless swing equation; a data loss and an initial condition loss are respectively constructed based on the difference between the predicted value and the dimensionless value; a total loss function is obtained by weighted summation of the losses; the network weight, bias and trainable parameters are updated by minimizing the total loss function until convergence; and the inertia time constant and damping coefficient after convergence are converted into named values. Thus, the accurate identification of the inertia time constant and damping coefficient of the virtual synchronous machine is realized.

Description

Technical Field

[0001] This application belongs to the field of power system analysis and control technology, and more specifically, relates to a method for identifying the inertia and damping parameters of a virtual synchronous machine based on PINN. Background Technology

[0002] With the large-scale integration of renewable energy sources such as wind power and photovoltaics, as well as energy storage systems, into the power grid via power electronic converters, the traditional power system dominated by synchronous generators is gradually evolving towards power electronics. Power electronic interface power supplies are decoupled from the grid and do not provide the inherent inertial response of rotating mass, leading to a continuous decrease in the system's equivalent inertia. This results in an increased rate of frequency change and a lower frequency minimum after fault disturbances, posing a serious challenge to system frequency stability. To alleviate these problems, Virtual Synchronous Generator (VSG) control technology is widely used in converter control. This technology introduces a swing equation into the converter control loop, making the converter exhibit inertia and damping characteristics similar to those of a traditional synchronous generator, thereby providing active frequency support for the grid. The inertia time constant H and damping coefficient D of the VSG are two key parameters characterizing its frequency support capability. Their values ​​directly determine the system's dynamic response characteristics and stability margin; therefore, accurately obtaining these two parameters is of great significance for grid stability analysis, parameter tuning, and coordinated control.

[0003] However, in practical engineering, the inertia time constant and damping coefficient of a virtual synchronous machine are often difficult to obtain directly. On the one hand, converter control parameters are usually encapsulated within the manufacturer's controller and cannot be directly read externally. On the other hand, in complex topologies such as those containing uncontrolled rectifiers or multiple converters in parallel, the system's equivalent inertia and damping are not equal to the setpoint of a single controller and must be identified through measured operating data. Currently, commonly used parameter identification methods mainly include the least squares method and its improved methods. These methods require numerical differentiation of the measured angular velocity signal to obtain angular acceleration. However, the measured signal inevitably contains measurement noise, which is significantly amplified by numerical differentiation, leading to large deviations in the identification results. Furthermore, when the disturbance is relatively gentle and the rate of change of angular velocity is small, the inertia term in the swing equation is several orders of magnitude smaller than the damping and power imbalance terms. The information carried by the inertia term is submerged by noise, causing the identification of inertia parameters to exhibit ill-conditioned characteristics. Inertia and damping compensate for each other and are difficult to separate accurately at the same time. Therefore, there is an urgent need for a method that has strong noise resistance and can stably identify the inertia time constant and damping coefficient under weak excitation conditions. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this application is to provide a method for identifying the inertia and damping parameters of a virtual synchronizer based on PINN, aiming to solve the problem of the difficulty in accurately identifying the inertia time constant and damping coefficient of a virtual synchronizer.

[0005] The first aspect of this application relates to a method for identifying the inertia and damping parameters of a virtual synchronizer based on PINN, comprising: Time alignment and per-unit processing were performed on the time-series data of angular velocity and active power imbalance of the virtual synchronous machine during the power disturbance period to obtain the per-unit values ​​of angular velocity and active power imbalance. A neural network is constructed with time as input, the normalized predicted values ​​of the angular velocity and active power imbalance as output, and the inertia time constant and damping coefficient of the virtual synchronous machine as trainable parameters. The physical loss is constructed based on the residuals of the standardized swing equation of the virtual synchronous machine during the power disturbance period; wherein the angular velocity derivative in the residual is obtained by the neural network through automatic differentiation of the output with respect to the input time; the data loss is constructed based on the difference between the standardized predicted value and the standard value of the angular velocity and the difference between the standardized predicted value and the standard value of the active power imbalance during the power disturbance period; the initial condition loss is constructed based on the difference between the standardized predicted value and the standard value of the angular velocity and the difference between the standardized predicted value and the standard value of the active power imbalance at the start of the disturbance. The weighted summation of the physical loss, data loss, and initial condition loss constitutes the total loss function. By minimizing the total loss function, the weights and biases of the neural network are updated, and the trainable parameters are also updated, until convergence. The converged inertia time constant and damping coefficient are converted into named values ​​to obtain the identified inertia time constant and damping coefficient.

[0006] Preferably, the time alignment process specifically involves: using the timestamp of the time series data with a lower sampling rate as the reference timestamp, downsampling the time series data with a higher sampling rate by an integer multiple, or selecting the data point closest to the reference timestamp for matching sampling, so that the timestamps of the two time series data are aligned.

[0007] Preferably, the per-unit processing specifically involves: dividing the angular velocity by the rated angular velocity to obtain the per-unit value of the angular velocity, and dividing the active power imbalance by the system power reference value to obtain the per-unit value of the active power imbalance.

[0008] Preferably, the neural network is a fully connected neural network, including an input layer, at least one hidden layer and an output layer, and the hidden layer uses a smooth and differentiable activation function to ensure that the output of the neural network is highly differentiable with respect to the input time, thereby supporting the automatic differentiation.

[0009] Preferably, the activation function is one of the Swish function, hyperbolic tangent function, Sigmoid function, or Softplus function.

[0010] Preferably, the physical loss is specifically the mean square value of the residual of the per-unit swing equation during the power disturbance period; The data loss specifically refers to: the root mean square error of the normalized predicted value and the normalized value of the angular velocity during the power disturbance period, plus the root mean square error of the normalized predicted value and the normalized value of the active power imbalance. The initial condition loss is specifically: at the start of the disturbance, the square of the difference between the normalized predicted value and the normalized value of the angular velocity plus the square of the difference between the normalized predicted value and the normalized value of the active power imbalance.

[0011] Preferably, the residual of the standardized rocking equation is specifically:

[0012] in, For residuals, The inertia time constant, The damping coefficient is... The derivative of the angular velocity obtained by automatic differentiation of the neural network. This is the per-unit predicted value of angular velocity. This is the per-unit predicted value of the active power imbalance.

[0013] Preferably, the weighting coefficients of physical loss, data loss, and initial condition loss in the total loss function are set according to the order-of-magnitude differences of each item, so that the inertia time constant and damping coefficient can obtain sufficient gradient information from the physical loss.

[0014] Preferably, the total loss function is minimized using a gradient optimization algorithm, wherein the gradient optimization algorithm is one of Adam, stochastic gradient descent, or L-BFGS.

[0015] In a second aspect, this application provides an electronic device, comprising: at least one memory for storing a program; and at least one processor for executing the program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to execute the method described in the first aspect or any possible implementation thereof.

[0016] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art: (1) In this application, the inertia time constant and damping coefficient are directly set as trainable parameters of the neural network, so that the automatic differentiation mechanism of the neural network can be used to obtain the derivative of angular velocity with respect to time during the training process, which replaces the numerical differentiation of the measured signal in the traditional least squares method. This fundamentally avoids the amplification effect of numerical differentiation on measurement noise and significantly improves the accuracy and robustness of inertia time constant and damping coefficient identification under noisy measured data.

[0017] (2) In this application, the inertia time constant and damping coefficient are directly set as trainable parameters of the neural network, so that it can obtain gradient information from the physical loss and automatically converge to the true value during the training process without relying on the internal parameters of the manufacturer's controller. At the same time, by weighting the physical loss, data loss and initial condition loss, the problem of small inertia information and ill-conditioned identification under weak excitation conditions can be addressed by adjusting the weights to enhance the gradient information of the inertia parameters, thereby achieving stable separation and identification of the inertia time constant and damping coefficient.

[0018] (3) This application embeds the swing equation as a physical constraint into the loss function, so that the identification results simultaneously satisfy the measured data and physical laws. Compared with the pure data-driven method, the introduction of physical constraints reduces the search space of the solution and improves the physical interpretability and reliability of the identification results.

[0019] (4) In the data preprocessing stage, the measured data is normalized to convert each physical quantity into a dimensionless per-unit value, which eliminates the gradient imbalance problem caused by the huge difference in magnitude between the nominal values, and ensures that the magnitude of each component of the physical loss is matched and effectively trained; at the same time, it makes the identification results under different capacity and voltage level equipment uniform and comparable.

[0020] (5) This application only requires two measured signals, namely angular velocity and active power imbalance, to complete the identification. It has low requirements for measurement signals and is suitable for complex topologies such as uncontrolled rectifier devices and multiple converters in parallel. It has good engineering applicability. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method for identifying the inertia and damping parameters of a virtual synchronizer based on PINN, provided in an embodiment of this application.

[0022] Figure 2 This is a trajectory diagram showing the convergence of the inertia time constant and damping coefficient during the neural network training process, as provided in the embodiments of this application.

[0023] Figure 3 This is a schematic diagram comparing the predicted and actual angular velocity values ​​provided in the embodiments of this application.

[0024] Figure 4This is a schematic diagram of the composition structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0026] In this application, the term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A existing alone, A and B existing simultaneously, and B existing alone. In this application, the symbol " / " indicates that the related objects are in an "or" relationship, for example, A / B means A or B.

[0027] In this application, the terms "first" and "second," etc., are used to distinguish different objects, not to describe a specific order of objects. For example, "first response message" and "second response message," etc., are used to distinguish different response messages, not to describe a specific order of response messages.

[0028] In this application, the term "electrical connection" can refer to a direct circuit connection or a signal transmission via a communication protocol.

[0029] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0030] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more, for example, multiple processing units means two or more processing units, multiple elements means two or more elements, etc.

[0031] For ease of understanding, the English abbreviations and related technical terms used in the embodiments of this application will be explained and described below.

[0032] The method in this application embeds the swing equation of the virtual synchronizer as a physical constraint into the loss function of the neural network to form a Physical Information Neural Network (PINN). Then, the automatic differentiation mechanism of the neural network is used to replace the numerical differentiation of the measured signal. In this way, while suppressing the influence of measurement noise, the inertia time constant and damping coefficient are used as trainable parameters. Through training, they automatically converge to the true value, thereby achieving accurate identification of the inertia time constant and damping coefficient of the virtual synchronizer.

[0033] The present application will be further described below with reference to the accompanying drawings and embodiments, such as... Figure 1 The illustration shows an embodiment of a method for identifying the inertia and damping parameters of a virtual synchronizer based on PINN, which specifically includes the following steps: S1. Perform time alignment and per-unit processing on the time-series data of angular velocity and active power imbalance of the virtual synchronous machine during the power disturbance period to obtain the per-unit values ​​of angular velocity and active power imbalance. Specifically: During the grid-connected operation of the virtual synchronous machine, the angular velocity of the virtual synchronous machine is synchronously collected during periods of power disturbance in the power grid. imbalance with active power Time-series data, including active power imbalance. Defined as mechanical power (or power reference value). With electromagnetic power The difference, that is .

[0034] Due to angular velocity imbalance with active power The sampling rates are often different. It is necessary to first use the timestamp of the one with the lower sampling rate as the base timestamp, and then downsample the time series data with the higher sampling rate by an integer multiple, or to sample the data by matching the nearest base timestamp. This will achieve timestamp alignment of the two time series data and obtain time series data with equal time intervals and consistent timestamps.

[0035] Furthermore, a time window containing the complete disturbance process is extracted, and the data is normalized by dividing the angular velocity by the rated angular velocity. To perform per-unit calculations, divide the active power by the system power reference value. The values ​​are standardized per unit to obtain the per-unit values ​​of angular velocity and active power imbalance.

[0036] S2. Construct a neural network that takes time as input, outputs the standardized predicted values ​​of the angular velocity and active power imbalance, and uses the inertia time constant and damping coefficient of the virtual synchronizer as trainable parameters. Specifically: Construct a fully connected neural network whose input is time. The outputs are standardized predicted values ​​of active power imbalance and angular velocity. The network consists of an input layer, several hidden layers, and an output layer. The hidden layers employ smooth, differentiable activation functions to ensure high-order differentiability of the network output with respect to the input time, thus supporting subsequent automatic differentiation operations. The activation function can be one of the following: Swish function, hyperbolic tangent function, sigmoid function, or Softplus function.

[0037] inertia time constant With damping coefficient Set as a trainable parameter of the neural network, participating in optimization along with the network's weights and biases. and The initial value is used as the starting point for optimization iteration.

[0038] S3. Based on the residuals of the standardized swing equation of the virtual synchronous machine during the power disturbance period, construct the physical loss; wherein, the angular velocity derivative in the residual is obtained by the neural network through automatic differentiation of the output with respect to the input time; based on the difference between the standardized predicted value and the standard value of the angular velocity and the difference between the standardized predicted value and the standard value of the active power imbalance during the power disturbance period, construct the data loss; based on the difference between the standardized predicted value and the standard value of the angular velocity and the difference between the standardized predicted value and the standard value of the active power imbalance at the start of the disturbance, construct the initial condition loss; including: S31. Physical loss.

[0039] The per-unit swing equation for the virtual synchronizer is:

[0040] Then there is the residual of the standardized rocking equation. for:

[0041] in, The inertia time constant, The damping coefficient is... The angular velocity derivative is obtained by automatic differentiation of the neural network, which eliminates the need for numerical differentiation of the measured signal, thus avoiding the amplification effect of measurement noise caused by numerical differentiation. This is the per-unit predicted value of angular velocity. This is the per-unit predicted value of the active power imbalance.

[0042] Several points are arranged during the power disturbance period, and the residuals are calculated at each point. And take its mean square value as the physical loss. As shown below:

[0043] in, The number of observation points can be flexibly selected independently of the observation points during the disturbance period.

[0044] S32. Data loss.

[0045] Using the per-unit values ​​of angular velocity and active power imbalance as observed values, the per-unit predicted values ​​output by the neural network are made to approximate the observed values ​​at each time step, thus constructing a data loss mechanism. As shown below:

[0046] in, This represents the number of observation points during the power disturbance period. , These are the standardized predicted values ​​output by the neural network at each observation point. , These are the observed values ​​at the corresponding observation points and times.

[0047] S33. Initial condition loss.

[0048] Using the per-unit values ​​of angular velocity and active power imbalance as observed values, the per-unit predicted value output by the neural network is made to approximate the observed value at the start of the disturbance, thus constructing the initial conditional loss. :

[0049] , The per-unit predicted value output by the neural network at the initial time of the perturbation. , These are the observations corresponding to the start time of the disturbance.

[0050] S4. The physical loss, data loss, and initial condition loss are weighted and summed to form the total loss function. By minimizing the total loss function, the weights and biases of the neural network are updated, and the trainable parameters are also updated, until convergence.

[0051] The physical loss, data loss, and initial condition loss are linearly combined according to their respective weights to form the total loss function. :

[0052] in, , and These are the weighting coefficients for physical loss, data loss, and initial condition loss, respectively. Since the numerical magnitudes of physical loss, data loss, and initial condition loss often differ significantly, by appropriately setting the above weighting coefficients, the inertia time constant and damping coefficient can obtain sufficient gradient information from the physical loss, preventing their gradients from being overwhelmed by the large magnitude of observation loss.

[0053] A gradient-based optimization algorithm is used to minimize the total loss function, updating the neural network weights while simultaneously optimizing the trainable parameters. and Synchronous updates are performed. As iterations proceed, the neural network output gradually satisfies the physical constraints of the rocking equation while satisfying the actual values, and the trainable parameters... and Then it converges, as Figure 2 The image shows the trajectory of iterative convergence, as follows: Figure 3 The diagram shows a comparison between the predicted and actual angular velocity values ​​after convergence, demonstrating that the neural network model has high prediction accuracy after convergence.

[0054] The gradient optimization algorithm is one of Adam, stochastic gradient descent, or L-BFGS.

[0055] S5. Convert the converged inertia time constant and damping coefficient into named values ​​to obtain the identified inertia time constant and damping coefficient.

[0056] When the total loss converges, the trainable parameters and After stabilization, the converged and The identification result is output as a per-unit value. This is based on the power reference value used. With rated angular velocity Convert the values ​​to nominal values ​​using the following formula to obtain the actual inertia time constant and damping coefficient of the virtual synchronizer:

[0057]

[0058] in, For the moment of inertia to have a given value, This is a nominal value for the damping coefficient. When using different power reference values, the identified per-unit values ​​can be converted according to the ratio of the reference values ​​to obtain parameter results under a unified reference.

[0059] Based on the methods in the above embodiments, this application provides an electronic device, such as... Figure 4As shown, the electronic device may include a processor, a communications interface, a memory, and a communication bus, wherein the processor, communications interface, and memory communicate with each other via the communication bus. The processor can invoke logical instructions stored in the memory to execute the methods described in the above embodiments.

[0060] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0061] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0062] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0063] It is understood that the processor in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0064] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0065] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0066] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.

[0067] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for identifying the inertia and damping parameters of a virtual synchronous machine based on PINN, characterized in that, include: Time alignment and per-unit processing were performed on the time-series data of angular velocity and active power imbalance of the virtual synchronous machine during the power disturbance period to obtain the per-unit values ​​of angular velocity and active power imbalance. A neural network is constructed with time as input, the normalized predicted values ​​of the angular velocity and active power imbalance as output, and the inertia time constant and damping coefficient of the virtual synchronous machine as trainable parameters. The physical loss is constructed based on the residuals of the standardized swing equation of the virtual synchronous machine during the power disturbance period; wherein the angular velocity derivative in the residual is obtained by the neural network through automatic differentiation of the output with respect to the input time; the data loss is constructed based on the difference between the standardized predicted value and the standard value of the angular velocity and the difference between the standardized predicted value and the standard value of the active power imbalance during the power disturbance period; the initial condition loss is constructed based on the difference between the standardized predicted value and the standard value of the angular velocity and the difference between the standardized predicted value and the standard value of the active power imbalance at the start of the disturbance. The weighted summation of the physical loss, data loss, and initial condition loss constitutes the total loss function. By minimizing the total loss function, the weights and biases of the neural network are updated, and the trainable parameters are also updated, until convergence. The converged inertia time constant and damping coefficient are converted into named values ​​to obtain the identified inertia time constant and damping coefficient.

2. The method for identifying the inertia and damping parameters of a virtual synchronizer according to claim 1, characterized in that, The time alignment process specifically involves: using the timestamp of the time series data with a lower sampling rate as the base timestamp, downsampling the time series data with a higher sampling rate by an integer multiple, or selecting the data point closest to the base timestamp for matching sampling, so that the timestamps of the two time series data are aligned.

3. The method for identifying the inertia and damping parameters of a virtual synchronizer according to claim 1, characterized in that, The per-unit processing specifically involves: dividing the angular velocity by the rated angular velocity to obtain the per-unit value of the angular velocity, and dividing the active power imbalance by the system power reference value to obtain the per-unit value of the active power imbalance.

4. The method for identifying the inertia and damping parameters of a virtual synchronizer according to claim 1, characterized in that, The neural network is specifically a fully connected neural network, including an input layer, at least one hidden layer, and an output layer. The hidden layer uses a smooth and differentiable activation function to ensure that the output of the neural network is highly differentiable with respect to the input time, thereby supporting the automatic differentiation.

5. The method for identifying the inertia and damping parameters of a virtual synchronizer according to claim 4, characterized in that, The activation function is one of the Swish function, hyperbolic tangent function, Sigmoid function, or Softplus function.

6. The method for identifying the inertia and damping parameters of a virtual synchronizer according to claim 1, characterized in that, The physical loss specifically refers to the mean square value of the residuals of the per-unit swing equation during the power disturbance period. The data loss specifically refers to: the root mean square error of the normalized predicted value and the normalized value of the angular velocity during the power disturbance period, plus the root mean square error of the normalized predicted value and the normalized value of the active power imbalance. The initial condition loss is specifically: at the start of the disturbance, the square of the difference between the normalized predicted value and the normalized value of the angular velocity plus the square of the difference between the normalized predicted value and the normalized value of the active power imbalance.

7. The method for identifying the inertia and damping parameters of a virtual synchronizer according to claim 1, characterized in that, The residual of the standardized rocking equation is specifically: in, For residuals, The inertia time constant, The damping coefficient is... The derivative of the angular velocity obtained by automatic differentiation of the neural network. This is the per-unit predicted value of angular velocity. This is the per-unit predicted value of the active power imbalance.

8. The method for identifying the inertia and damping parameters of a virtual synchronizer according to claim 1, characterized in that, The weighting coefficients of physical loss, data loss, and initial condition loss in the total loss function are set according to the order-of-magnitude differences of each item, so that the inertia time constant and damping coefficient can obtain sufficient gradient information from the physical loss.

9. The method for identifying the inertia and damping parameters of a virtual synchronizer according to claim 1, characterized in that, The total loss function is minimized using a gradient optimization algorithm, which is one of Adam, stochastic gradient descent, or L-BFGS.

10. An electronic device, characterized in that, Includes memory and one or more processors; The memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions; The one or more processors invoke the computer instructions to cause the electronic device to perform the method as described in any one of claims 1-9.

Images