Method and device for constructing transient stability domain of grid-following type converter

Abstract

The invention discloses a method and device for constructing a transient stability domain of a grid-following converter, and the method comprises the steps: constructing a target model based on a topological structure of a grid-connected system of the grid-following converter; for each variable combination, utilizing the target model to solve a quiescent working point of the grid-following type converter, marking a stability label for the variable combination, and storing the variable combination, the corresponding stability label and the solved quiescent working point into a data pool; engineering features are added to variable combinations in the data pool; constructing a multi-task learning model based on the stability label and the engineering features, and carrying out mixed training on the multi-task learning model; obtaining a stability judgment result of each combination, and further obtaining a corresponding static working point prediction value; and generating a parameter feasible region based on the stability judgment result of all the variable combinations and the static working point predicted value thereof. The construction of the transient stability domain can provide a quantitative basis for parameter setting.

Description

Technical Field

[0001] This application relates to the field of power system technology, and in particular to a method and apparatus for constructing the transient stability domain of a grid-connected converter. Background Technology

[0002] Existing transient stability assessment methods for grid-following converters (GFLCs) are mostly based on linearization analysis, voltage and current margin criteria, or large-scale time-domain simulations. These methods cannot directly determine the existence of the steady-state point (SSOP) under a given operating condition, and relying on simplified power flow or steady-state equations to solve for the SSOP results in poor accuracy. On the one hand, existing methods are computationally intensive and struggle to efficiently construct transient stability domains within high-dimensional operating condition-parameter space. On the other hand, they often treat transient stability assessment and SSOP existence determination and solving as independent tasks, modeling and training them separately without sharing features within a unified framework, leading to limited data utilization efficiency and model generalization ability.

[0003] In recent years, deep learning and physical constraint modeling have begun to be applied to converter stability analysis. However, most existing works still remain in the single-task learning paradigm, lacking a mechanism for joint modeling and collaborative optimization of stability classification and static operating point regression. Furthermore, the weight design between different loss terms depends on experience and cannot be quantified, resulting in insufficient stability and engineering applicability of the training process. Summary of the Invention

[0004] This disclosure provides a method and apparatus for constructing the transient stability domain of a grid-connected converter, thereby at least solving the above-mentioned technical problems existing in the prior art.

[0005] According to a first aspect of this application, a method for constructing the transient stability domain of a grid-connected converter is provided, comprising: Based on the topology of a grid-connected system with a grid-connected converter, a target model is constructed. Using at least one operating condition and at least two control parameters as variables, sampling combinations are performed within the value range of the variables to obtain multiple variable combinations; For each of the variable combinations, the static operating point of the grid-type converter is solved using the target model. Based on whether the solution is successful or not, a stability label is marked for the variable combination. The variable combination, its corresponding stability label, and the obtained static operating point are stored in the data pool. Among them, the successful solution is marked as stable, and the unsuccessful solution is marked as unstable. Add engineering features to the variable combinations in the data pool; the engineering features are features generated based on prior knowledge. Based on the stability label and the engineering features, the variable combinations in the data pool are divided into a first subset for training classification tasks and a second subset for training regression tasks. Construct a multi-task learning model, which includes a classification task head and a regression task head; The classification task head is trained using the first subset of the dataset, and the regression task head is trained using the second subset of the dataset. During the training process, an adaptive loss balancing mechanism is used to coordinate the classification loss and the regression loss. A physical constraint loss based on the target model is introduced for the training of the regression task within a preset time period to complete the hybrid training of the multi-task learning model. Different combinations of operating condition variables and control parameters are input into the trained multi-task learning model. The stability determination results of each combination are obtained through the classification task head, and the corresponding static operating point prediction values ​​are obtained through the regression task head. Based on the stability determination results of all variable combinations and their static operating point prediction values, a parameter feasible domain characterizing the stable operating range of the grid-connected converter system is generated.

[0006] In one possible implementation, the grid-connected converter system includes: a main circuit and a control loop; the construction of a target model based on the topology of the grid-connected converter system includes: Based on the topology of the main circuit, a first mathematical model is constructed; Based on the topology of the control loop, a second mathematical model is constructed; By combining the first mathematical model and the second mathematical model, the target model is obtained.

[0007] In one embodiment, the grid-connected converter circuit includes: a DC-side DC voltage source, a three-phase converter, a filter inductor, a PWM modulator, a filter capacitor, and a grid impedance. The control loop includes: a phase-locked loop unit, a coordinate transformation unit, a power calculation unit, a power outer loop control unit, and a current inner loop control unit.

[0008] In one possible implementation, the operating condition variable is the transient grid voltage amplitude; The control parameters include the proportional coefficient and integral coefficient of the phase-locked loop; The variables also include grid-connected inductance parameters.

[0009] In one embodiment, the engineering feature is a physical quantity or dimensionless number that is calculated based on the combination of variables and can reflect the dynamic characteristics of the system.

[0010] In one possible implementation, the step of dividing the variable combinations in the data pool into a first subset for training a classification task and a second subset for training a regression task based on the stability label and the engineering features includes: The first subset of data is defined as all combinations of variables marked as unstable, and the subset of variables marked as stable that are determined to be close to the stability boundary based on the engineering characteristics. The remaining portion of the variable combinations marked as stable, which is determined to be far from the stable boundary based on the engineering characteristics, is divided into the second subset.

[0011] In one embodiment, the adaptive loss balancing mechanism is used to dynamically adjust the relative weights of the classification task loss and the regression task loss during training.

[0012] In one possible implementation, the physical constraint loss is the residual loss calculated from the system of differential-algebraic equations corresponding to the target model.

[0013] In one possible implementation, the feasible domain of the parameters includes at least a two-dimensional stable region boundary map with the transient grid voltage amplitude as the abscissa and the phase-locked loop proportional coefficient, integral coefficient, or grid-connected inductance as the ordinate.

[0014] According to a second aspect of this application, a device for constructing a transient stability domain for a grid-connected converter is provided, comprising: The first building module is used to construct the target model based on the topology of the grid-connected system with a grid-connected converter. The acquisition module is used to sample and combine at least one operating condition and at least two control parameters as variables within the range of values ​​of the variables to obtain multiple variable combinations; The solution module is used to solve the static operating point of the grid-type converter for each variable combination using the target model, and to mark the variable combination with a stability label according to whether the solution is successful or not, and to store the variable combination, its corresponding stability label and the obtained static operating point in the data pool; wherein, the successful solution is marked as stable, and the unsuccessful solution is marked as unstable. The combination module is used to add engineering features to combinations of variables in the data pool; the engineering features are features generated based on prior knowledge. The partitioning module is used to divide the combination of variables in the data pool into a first subset for training a classification task and a second subset for training a regression task, based on the stability label and the engineering features. The second building module is used to build a multi-task learning model, which includes a classification task head and a regression task head. The training module is used to train the classification task head using the first subset of the dataset and the regression task head using the second subset of the dataset. During the training process, an adaptive loss balancing mechanism is used to coordinate the classification loss and the regression loss. In a preset time period, a physical constraint loss based on the target model is introduced for the training of the regression task to complete the hybrid training of the multi-task learning model. The determination module is used to input different combinations of operating condition variables and control parameters into the trained multi-task learning model, obtain the stability determination results of each combination through the classification task head, and obtain the corresponding static operating point prediction value through the regression task head. The generation module is used to generate a parameter feasible domain that characterizes the stable operating range of the grid-connected converter system based on the stability determination results of all variable combinations and their static operating point prediction values.

[0015] According to a third aspect of this application, an electronic device is provided, comprising: A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in any of the above embodiments.

[0016] According to a fourth aspect of this application, a non-transitory computer-readable storage medium is provided storing computer instructions for causing the computer to perform the methods described in this application.

[0017] According to a fifth aspect of this application, a computer program product is provided, comprising a computer program or instructions that, when executed by a processor, implement the method described in this application.

[0018] The technical solution of this application can solve the problems of difficulty and inefficiency in stability judgment during transient processes, help improve the flexibility of the system under variable operating conditions, and the construction of the transient stability domain can provide a quantitative basis for parameter tuning.

[0019] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0020] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of this application are illustrated in the drawings by way of example and not limitation, in which: In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

[0021] Figure 1 A schematic diagram illustrating the implementation process of the transient stability domain construction method for a grid-type converter in an embodiment of this application is shown. Figure 2 The diagram shows the main circuit topology of the grid converter in an embodiment of this application. Figure 3 The diagram shows the phase-locked loop control circuit of the grid converter in an embodiment of this application; Figure 4 The diagram shows a non-phase-locked loop control circuit diagram of a grid-type converter in an embodiment of this application; Figure 5 This application shows a schematic diagram of the grid-connected converter, the grid-synchronized rotating coordinate system, and the three-phase stationary coordinate system in an embodiment of the present application. Figure 6 The simulation diagram of the mathematical model verification in the embodiment of this application is shown; Figure 7 The diagram shows the training loss of the neural network model in an embodiment of this application; Figure 8 The voltage amplitude-PLL scaling factor stability domain diagram of the embodiment of this application is shown; Figure 9 The voltage amplitude-PLL integral coefficient stability domain diagram of the embodiment of this application is shown; Figure 10 The voltage amplitude-grid inductor stability domain diagram of an embodiment of this application is shown; Figure 11 This paper shows a schematic diagram of the structure of the transient stability domain construction device for the grid converter in an embodiment of this application; Figure 12 A schematic diagram of the composition structure of the electronic device in an embodiment of this application is shown. Detailed Implementation

[0022] To make the objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] In the following description, the terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first" and "second" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0025] The following description, in conjunction with the accompanying drawings, introduces a method and apparatus for constructing the transient stability domain of a grid-connected converter provided in this application.

[0026] like Figure 1 As shown, this application provides a method for constructing the transient stability domain of a grid-connected converter, including: S101, based on the topology of the grid-connected system of the grid-connected converter, construct the target model.

[0027] In some embodiments, the grid-connected converter system includes: a main circuit and a control loop; the construction of the target model based on the topology of the grid-connected converter system includes: Based on the topology of the main circuit, a first mathematical model is constructed; Based on the topology of the control loop, a second mathematical model is constructed; By combining the first mathematical model and the second mathematical model, the target model is obtained.

[0028] The grid-connected converter circuit includes: a DC-side DC voltage source, a three-phase converter, a filter inductor, a PWM modulator, a filter capacitor, and a grid impedance. The control loop includes: a phase-locked loop unit, a coordinate transformation unit, a power calculation unit, a power outer loop control unit, and a current inner loop control unit.

[0029] like Figure 2 , Figure 3 and Figure 4 The diagram shows the main circuit and control loop of a grid-connected converter, including the phase-locked loop (PLL) control loop and non-PLL control loop. This application simplifies the main circuit and control loop during modeling to construct the target model. One aspect is dynamic decoupling on the DC side: due to the large-capacity capacitor on the DC bus side... The DC voltage closed-loop (PI) control of the preceding stage can quickly suppress voltage disturbances, so the DC voltage source can be regarded as an ideal constant voltage source, and its dynamics are decoupled from the inverter's downstream stage and AC side circuits in the time domain. Secondly, PWM and power switching are simplified: in small-signal modeling, if the switching frequency is much higher than the system's dynamic bandwidth, the switch and PWM circuit can be considered as a unit static gain, avoiding the introduction of high-frequency components in the time-domain simulation. This application adopts an averaged model for the power switching devices in the three-phase converter and approximates the overall voltage gain of the PWM modulator and the converter composed of switches as a constant. .

[0030] In this application, the modeling can be constructed by modeling the main circuit and control loop of the GFLC grid-connected system. A first mathematical model is constructed based on the topology of the grid-connected converter circuit. A second mathematical model is constructed based on the topology of the control loop. The first and second mathematical models are then combined to obtain the target model.

[0031] Specifically, the details for each unit in the control loop are as follows: The power calculation unit is used to calculate the active and reactive power output of the grid-connected converter; specifically, the power calculation unit is represented in the following way.

[0032] in, , These represent the active power and reactive power output by the converter to the power grid, respectively. The measured values ​​of the d-axis and q-axis components of the voltage at point PCC; The d-axis and q-axis components of the current injected into the grid by the converter are measured.

[0033] A phase-locked loop (PLL) unit is used to automatically synchronize the output signal with a reference input signal in both frequency and phase. Specifically, a PLL unit is represented as follows:

[0034]

[0035] in, express The difference between it and its reference value (usually 0, to facilitate power decoupling); The angular velocity generated by the PLL is used to rotate the grid converter's rotating coordinate system at that angular velocity. This is the system's rated angular frequency; These represent the integral gain and proportional gain of the PI loop in the PLL, respectively.

[0036] The power outer loop control unit is used to enable the grid-connected converter to accurately track a given power command and generate a current reference value; specifically, the power outer loop control unit is represented as follows.

[0037]

[0038] in, These represent the errors between the reference values ​​and measured values ​​of the active and reactive power outputs of the grid-connected converter, respectively. These represent the d-axis and q-axis components of the reference value of the current generated by the outer power loop, respectively. These represent the integral gain and proportional gain of the power outer loop, respectively. , They represent respectively.

[0039] The current inner loop control unit is used to ensure that the output voltage and filter inductor current of the grid-connected converter accurately track the given values, and to provide reactive power support to the grid connection point voltage based on the voltage drop during transients. Specifically, the current inner loop control unit is represented as follows:

[0040]

[0041] in, The d and q-axis components represent the error between the reference and measured values ​​of the inductor current, respectively. These represent the d-axis and q-axis components of the converter output voltage reference value generated by the inner current loop, respectively. This is the value of the filter inductance; These represent the integral gain and proportional gain of the current control inner loop, respectively.

[0042] The filter unit on the output side of a grid-connected converter includes a filter inductor, the parasitic resistance of the filter inductor, and a filter capacitor. The filter unit is represented as follows.

[0043]

[0044] in, , These respectively represent the power grid Voltage at point PCC in a synchronously rotating coordinate system Axial components; , These represent the current injected into the power grid by the filter inductor and the filter, respectively. Axial components; , These represent the current injected into the grid by the converter. Axial component (i.e., current flowing through the grid equivalent impedance); This is the value of the filter capacitor; This is the parasitic resistance value of the filter inductor; This is the value of the filter inductance; This refers to the angular frequency of the power grid operation.

[0045] A power grid unit comprises an infinite power grid and its equivalent line impedance. Specifically, a power grid unit is represented as follows:

[0046] in, Indicates the equivalent transmission line impedance. This represents the angular frequency of the power grid.

[0047] The coordinate transformation unit is used to convert the output electrical quantities of the converter from the converter rotating coordinate system to the grid synchronous rotating coordinate system. Specifically, the coordinate transformation unit is represented as follows: Because the rotating coordinate system of the GFLC rotates at different speeds than the rotating coordinate system of the power grid, the three-phase stationary coordinate system and the power grid... The relationship between the synchronous rotating coordinate system and the GFLC rotating coordinate system is shown in the appendix. Figure 5 As shown. Figure 5 middle: The composite voltage vector at the PCC point of the GFLC; This is the composite vector of grid voltage. The rotational speed of the synchronous rotating reference coordinate system is The phase of Ug in the three-phase stationary coordinate system is And it always coincides with the x-axis. In steady state, With PSC The d-axis of the coordinate system coincides. The converter control system is established in its own power synchronous coordinate system, while the external AC system is established in the grid synchronous selection coordinate system. Therefore, it is necessary to transform all physical quantities into a reference coordinate system rotating at synchronous speed, i.e. The calculations are performed in a unified coordinate system based on synchronous rotation, and the coordinate transformation is as follows:

[0048]

[0049] in, This represents the phase difference between two synchronously rotating coordinate systems with different rotational speeds.

[0050] S102, taking at least one operating condition and at least two control parameters as variables, sampling and combining them within the range of values ​​of the variables to obtain multiple variable combinations.

[0051] In this application, the operating condition variable is the transient grid voltage amplitude; The control parameters include the proportional coefficient and integral coefficient of the phase-locked loop; The variables also include grid-connected inductance parameters.

[0052] Specifically, this application considers the entire operating range of the GFLC transient operating condition variables and the phase-locked loop control parameters, sampling and combining them within the range of their values.

[0053] S103, for each of the variable combinations, the static operating point of the grid-type converter is solved using the target model, and a stability label is marked for the variable combination according to whether the solution is successful or not. The variable combination, its corresponding stability label and the obtained static operating point are stored in the data pool; wherein, the solution is successfully marked as stable, and the solution fails as unstable.

[0054] This application attempts to solve for the static operating point of the converter using a target model, obtaining a data pool containing information mapping the system stability and static operating point solution results to combinations of operating conditions and control parameters. Specifically, it considers one variable (all physical quantities mentioned in this application are per-unit values) for the transient voltage drop condition of the GFLC, namely the transient voltage range of the grid voltage amplitude. and the phase-locked loop control parameters of the GFLC, namely , Other control parameters (power outer loop PI control parameters) , Current control loop PI control parameters , Hardware parameters (filter inductor) Parasitic resistance of filter inductor Filter capacitor According to the traditional parameter tuning method, the initial value and value range are tuned; within the value range of 1 operating condition parameter, 2 adjustable control parameters, and 1 circuit parameter, the operating condition and parameters are uniformly sampled and combined. Using the GFLC mathematical model described in steps 1 and 2, the SSOP of the converter is solved. The operating condition and parameter combination that is successfully solved is recorded as a stable combination and the stability label is marked as 1. The operating condition and parameter combination that fails to be solved is recorded as an unstable combination and the stability label is marked as 0. A data pool containing the solution results of the operating condition and control parameter combination to the static operating point and the combination stability mapping information is obtained.

[0055] S104, add engineering features to the combination of variables in the data pool; the engineering features are features generated based on prior knowledge.

[0056] In some embodiments, the engineering features are physical quantities or dimensionless numbers that are calculated based on the combination of variables and can reflect the dynamic characteristics of the system.

[0057] S105, based on the stability label and the engineering features, the combination of variables in the data pool is divided into a first subset for training the classification task and a second subset for training the regression task.

[0058] In some embodiments, the step of dividing the variable combinations in the data pool into a first subset for training a classification task and a second subset for training a regression task based on the stability label and the engineering features includes: The first subset of data is defined as all combinations of variables marked as unstable, and the subset of variables marked as stable that are determined to be close to the stability boundary based on the engineering characteristics. The remaining portion of the variable combinations marked as stable, which is determined to be far from the stable boundary based on the engineering characteristics, is divided into the second subset.

[0059] In this application, based on operating condition variables and control parameters, engineering features are introduced into the data pool as an attention mechanism. The data pool is then divided into a first subset and a second subset according to the requirements of two different tasks (classification and regression) in the multi-task learning model. It can also be further divided into training and validation sets. Based on the main engineering features, combinations labeled as stable are further divided into internally stable combinations and boundary-stable combinations. Boundary-stable combinations are assigned to the first subset of the dataset. Boundary stable combinations are assigned to the second subset; combinations marked as unstable are assigned entirely to the first subset. The internal stable combinations are assigned to the first subset of the dataset. The internal stable combinations are allocated to the second subset. Then, the first and second subsets are used for training and validation of the two neural networks in an 8:2 ratio.

[0060] The attention mechanism can be expressed as follows: several engineering features that differ significantly between stable and unstable samples, such as power margin, voltage deviation, and frequency deviation, at a certain moment during converter operation, are used as the attention mechanism. The specific formula is as follows: The power margin of the converter at a certain moment is

[0061] Grid voltage deviation

[0062] Frequency deviation is

[0063] Control gain ratio is

[0064] Because the output of the dataset contains steady-state errors of magnitude [missing information] The physical quantity also includes the standardized inductor current, which is on the order of magnitude. To eliminate dimensional differences and avoid the model completely ignoring or overemphasizing a certain output, the physical quantities are normalized. The formula for the scaling operation is:

[0065] in, This represents the original value of the j-th feature of the i-th sample; Let be the mean of the j-th feature on the training set; Let be the standard deviation of the j-th feature on the training set.

[0066] S106, Constructing a multi-task learning model, wherein the multi-task learning model includes a classification task head and a regression task head.

[0067] The multi-task learning model employs a multi-task learning neural network.

[0068] S107, the classification task head is trained using the first subset of the dataset, and the regression task head is trained using the second subset of the dataset. During the training process, an adaptive loss balancing mechanism is used to coordinate the classification loss and the regression loss. In a preset time period, a physical constraint loss based on the target model is introduced for the training of the regression task, thereby completing the hybrid training of the multi-task learning model.

[0069] In some embodiments, the adaptive loss balancing mechanism is used to dynamically adjust the relative weights of the classification task loss and the regression task loss during training.

[0070] The physical constraint loss is the residual loss calculated from the system of differential-algebraic equations corresponding to the target model.

[0071] Specifically, this application addresses the problem that multi-task learning models need to simultaneously optimize three objectives: classification (stability determination of the corresponding classifier head), regression (SSOP solution of the corresponding regression head), and physical constraints (PDE residuals), and that the significant differences in the loss magnitudes of each objective can easily lead to training instability. To solve this problem, an adaptive loss balancing mechanism is introduced. This mechanism is... Each loss item is set separately. A learnable log-variance parameter and the total loss function Defined as:

[0072] in For the first The loss for each task; in this application, there are 3 tasks, and the loss items are categorized as follows: Loss for each task. Regression loss Physical constraint loss This application utilizes learnable parameters. Automatically adjust the weights of various losses when the uncertainty (i.e., variance) of a certain task increases. When the weight increases, its weight automatically decreases, thereby achieving a dynamic balance of multi-task loss, avoiding tedious manual weight tuning, and significantly improving the stability of model convergence.

[0073] In addition, regarding physical constraint loss Internally, it consists of multiple (e.g., 12 in this application) partial differential equation (PDE) residuals To avoid excessively large residual gradients from individual equations dominating training, a robust normalized PDE loss calculation method is introduced. This method first calculates all... The running variance of the residuals of each equation And use its reciprocal as an adaptive scaling factor:

[0074] in, To prevent division by zero, take a small value. Then, the Huber loss, which is robust to outliers, is used. (A smoothed L1 loss) replaces the traditional mean squared error, resulting in the final physical constraint loss. for:

[0075] Through the above improvements, this application ensures a reasonable balance between the loss contributions of different tasks and between different equations within the physical constraints, enabling the neural network to learn stably and efficiently.

[0076] The multi-task learning model constructed in this application includes a shared feature encoder, which is responsible for extracting deep features from the input operating parameters and control parameters. This encoder is followed by two independent heads: a multilayer perceptron for the classification head that handles the stability determination task and needs to complete the binary classification task, and a multilayer perceptron for the regression head that handles the static operating point (SSOP) solution task and needs to complete the SSOP solution task.

[0077] As a specific embodiment, training set 1 and training set 2 are used to train the multi-task learning model. A batch is sampled from training set 1 and input into the shared encoder and classification head. The classification loss is calculated using the binary cross-entropy loss function. A batch is sampled from training set 2 and input into the shared encoder and regression head. The regression loss is calculated using the mean squared error loss function. To calculate the physical constraint loss Due to the output of the regression head The value, which has been normalized as described in S102, must first be denormalized to its physical value. The inverse normalization formula is as follows:

[0078] Then, Substituting into the target model and using the normalized PDE loss calculation method, we obtain...

[0079] Will , and The adaptive loss balancing mechanism described in S103, which inputs three loss factors, yields a weighted single total loss. .right Perform a backpropagation to collaboratively update all parameters of the shared encoder, classification head, and regression head.

[0080] The training loss process is as follows Figure 6 As shown, at 1000 epochs, the classification and regression losses have converged. At this point, adding the PDE loss has a slight impact on the regression loss, but it can recover to the previous level relatively quickly. This indicates that the multi-task learning model can satisfy the physical constraints represented by the PDE while fitting the dataset, and has good generalization ability. After 2500 epochs, the three losses converge, and all converge to close to 0, indicating that the multi-task learning model in this application performs well.

[0081] Furthermore, this application can also utilize a validation set to validate the trained multi-task learning model. The validation set 1 described in S102 is input into the model's shared encoder and classifier head, and the output predicted labels are compared with the true labels. Four metrics with complementary characteristics in evaluating classification performance—accuracy, precision, recall, and F1 score—are calculated to assess the model's accuracy in determining transient stability. For the binary classification task of this application, if all four metrics are above 90%, it indicates good classification performance. When calculating the classification metrics, four basic variables are defined: TP (True Positive), FP (False Positive), TN (True Negative), and FN (False Negative). In the binary classification problem of this application, a positive example (SSOP) is defined as the existence of a positive example (SSOP), and a negative example (SSOP) is defined as the absence of a negative example (SSOP).

[0082] Precision:

[0083] Precision represents the proportion of truly positive samples (i.e., SSOPs actually exist) among all samples predicted as positive by the model (i.e., the model believes SSOPs exist).

[0084] Recall:

[0085] High recall indicates that the model has basically not missed any samples where SSOP exists.

[0086] Accuracy:

[0087] Accuracy represents the proportion of samples in which the model correctly predicts the correct F1 score.

[0088]

[0089] A high F1 score is achieved when both Precision and Recall are high.

[0090] SSOP solution performance verification: Input the validation set 2 described in S102 into the shared encoder and regression head of the model, perform the inverse normalization of the normalized value output by the network as described in step S104, and compare it with the true SSOP solution in the validation set. Calculate the mean square error index to evaluate the accuracy of the model in solving the static operating point; the smaller the mean square error, the better the regression performance.

[0091]

[0092] in, Indicates the total number of samples; Indicates the first The true value of each sample; Indicates the first The model prediction value for each sample.

[0093] S108, different combinations of operating condition variables and control parameters are input into the trained multi-task learning model, and the stability determination results of each combination are obtained through the classification task head, and the corresponding static operating point prediction values ​​are obtained through the regression task head.

[0094] The stability determination and solution performance of the multi-task learning model are shown in Table 1 below. It can be seen that the multi-task learning model performs well in both classification and regression tasks. It can predict the stability of GFLC and the specific SSOP value under various operating conditions of grid voltage drop, and obtain the operating condition-parameter stability domain, thus providing a basis for parameter tuning.

[0095] Table 1 Classification / Solution Performance of Multi-Task Learning Model

[0096] S109, Based on the stability determination results and static operating point prediction values ​​of all variable combinations, generate a parameter feasible domain that characterizes the stable operating range of the grid-connected converter system.

[0097] The feasible domain of the parameters includes at least a two-dimensional stable region boundary map with the transient grid voltage amplitude as the abscissa and the phase-locked loop proportional coefficient, integral coefficient or grid-connected inductance as the ordinate.

[0098] This application addresses one operating condition parameter (grid voltage amplitude). ), and 2 adjustable control parameters (PLL proportional coefficient) PLL integral coefficient ) or 1 circuit parameter (grid inductor) Dense grid sampling is performed within the range of its values. The parameter combination of each sampling point is input into the trained multi-task learning model. The classification head output of the multi-task learning model is used to determine whether the parameter combination is stable; if stable, it outputs 1, and if unstable, it outputs 0. If it is determined to be stable, the regression head output of the model is further used to obtain the SSOP solution for this working condition. Visualizing all parameter points determined to be stable by the classification head yields the following result: Figure 7 , Figure 8 , Figure 9 The feasible regions of operating condition parameters such as voltage amplitude-PLL proportional coefficient, voltage amplitude-PLL integral coefficient, and voltage amplitude-grid inductance shown provide an intuitive basis for GFLC parameter tuning. Figure 8 , Figure 9 , Figure 10 In the middle, the instability of the GFLC grid-connected system outside the stability domain corresponds to the instability of the system; by Figure 8 It can be seen that, with the PLL scaling factor As the fault depth increases, the system is more likely to oscillate due to the coupling introduced by the PLL, thus increasing the risk of transient instability. Figure 9 Similarly. From Figure 10 It can be seen that, with grid-connected inductance Increased power grid strength, or weakened grid strength, leads to a higher risk of transient instability in the system as the fault depth increases.

[0099] This application's technical solution aims to accurately construct and rapidly determine the transient stability domain of a grid-connected converter circuit under transient operating conditions. By introducing a multi-task learning framework, transient stability classification and static operating point regression are simultaneously performed within the multi-task learning model, thereby improving the accuracy and efficiency of stability domain characterization across a wide range of operating conditions and parameter spaces. Transient stability domain construction refers to determining whether a static operating point satisfying constraints exists for a given operating condition variable and control parameters, considering transient disturbances such as voltage dips, frequency shifts, and grid impedance changes, and delineating the stability region boundary of the GFLC grid-connected system within the operating condition and parameter space.

[0100] This application includes mathematical model construction and sample set generation, multi-task learning model design, model training and transient stability domain construction, and performance verification. Specifically, the target model construction focuses on the main circuit and control loop of a GFLC grid-connected system, establishing a closed-loop nonlinear mathematical model including an outer voltage loop, an inner current loop, and a grid-connected side filtering stage. A sample dataset with stability labels and static operating point solutions is generated through operating condition disturbances and control parameter scanning. In the multi-task learning model design stage, a multi-task neural network structure consisting of a shared feature encoder, a classification task head, and a regression task head is constructed. Engineering features and physical constraint losses are introduced to improve the model's ability to represent the operating condition-parameter-stability relationship. In the model training and transient stability domain construction stage, an adaptive loss balancing mechanism is used to jointly optimize the classification loss, regression loss, and physical constraint loss, obtaining a multi-task learning model that can simultaneously output transient stability judgment results and static operating points. This model is then used to quickly scan and construct the GFLC transient stability domain in a high-dimensional operating condition-parameter space.

[0101] The transient stability domain construction method for grid-connected converters proposed in this application achieves coordinated optimization of stability determination and static operating point solution under complex transient operating conditions by establishing a unified modeling and inference mechanism that combines a multi-task learning model with a grid-connected converter mathematical model. This significantly improves the accuracy and computational efficiency of transient stability domain construction.

[0102] like Figure 11 As shown, this application provides a device for constructing the transient stability domain of a grid-connected converter, comprising: The first construction module 1101 is used to construct the target model based on the topology of the grid-connected system of the grid-connected converter. The acquisition module 1102 is used to sample and combine at least one working condition and at least two control parameters as variables within the range of values ​​of the variables to obtain multiple variable combinations; The solution module 1103 is used to solve the static operating point of the grid-type converter for each variable combination using the target model, and to mark the variable combination with a stability label according to whether the solution is successful or not, and to store the variable combination, its corresponding stability label and the obtained static operating point in the data pool; wherein, the successful solution is marked as stable, and the unsuccessful solution is marked as unstable. The combination module 1104 is used to add engineering features to the combination of variables in the data pool; the engineering features are features generated based on prior knowledge. The partitioning module 1105 is used to partition the combination of variables in the data pool into a first subset for training a classification task and a second subset for training a regression task based on the stability label and the engineering features. The second construction module 1106 is used to construct a multi-task learning model, which includes a classification task head and a regression task head. Training module 1107 is used to train the classification task head using the first subset of dataset and the regression task head using the second subset of dataset. During the training process, an adaptive loss balancing mechanism is used to coordinate the classification loss and regression loss. Physical constraint loss based on the target model is introduced for the training of the regression task within a preset time period to complete the hybrid training of the multi-task learning model. The determination module 1108 is used to input different combinations of operating condition variables and control parameters into the trained multi-task learning model, obtain the stability determination result of each combination through the classification task head, and obtain the corresponding static operating point prediction value through the regression task head. The generation module 1109 is used to generate a parameter feasible domain that characterizes the stable operating range of the grid-connected converter system based on the stability determination results of all variable combinations and their static operating point prediction values.

[0103] According to embodiments of this application, this application also provides an electronic device and a readable storage medium.

[0104] The electronic device includes at least one processor and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which, when executed, enable the at least one processor to perform the grid-connected converter transient stability domain construction method described in this application. The computer instructions are used to cause the computer to execute the grid-connected converter transient stability domain construction method described in this application.

[0105] This application also provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the grid-connected converter transient stability domain construction method of this application.

[0106] Figure 12 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of this application is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the application described and / or claimed herein.

[0107] like Figure 12 As shown, device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0108] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of monitors, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0109] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as the grid-connected converter transient stability domain construction method. For example, in some embodiments, the grid-connected converter transient stability domain construction method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the grid-connected converter transient stability domain construction method described above can be performed. Alternatively, in other embodiments, the computing unit 801 may be configured by any other suitable means (e.g., by means of firmware) to perform a method for constructing a transient stability domain for a grid-type converter.

[0110] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for constructing the transient stability domain of a grid-connected converter, characterized in that, include: Based on the topology of a grid-connected system with a grid-connected converter, a target model is constructed. Using at least one operating condition and at least two control parameters as variables, sampling combinations are performed within the value range of the variables to obtain multiple variable combinations; For each of the variable combinations, the static operating point of the grid-type converter is solved using the target model. Based on whether the solution is successful or not, a stability label is marked for the variable combination. The variable combination, its corresponding stability label, and the obtained static operating point are stored in the data pool. Among them, the successful solution is marked as stable, and the unsuccessful solution is marked as unstable. Add engineering features to the variable combinations in the data pool; the engineering features are features generated based on prior knowledge. Based on the stability label and the engineering features, the variable combinations in the data pool are divided into a first subset for training classification tasks and a second subset for training regression tasks. Construct a multi-task learning model, which includes a classification task head and a regression task head; The classification task head is trained using the first subset of the dataset, and the regression task head is trained using the second subset of the dataset. During the training process, an adaptive loss balancing mechanism is used to coordinate the classification loss and the regression loss. A physical constraint loss based on the target model is introduced for the training of the regression task within a preset time period to complete the hybrid training of the multi-task learning model. Different combinations of operating condition variables and control parameters are input into the trained multi-task learning model. The stability determination results of each combination are obtained through the classification task head, and the corresponding static operating point prediction values ​​are obtained through the regression task head. Based on the stability determination results of all variable combinations and their static operating point prediction values, a parameter feasible domain characterizing the stable operating range of the grid-connected converter system is generated.

2. The method according to claim 1, characterized in that, The grid-connected converter system includes: a main circuit and a control loop; the target model constructed based on the topology of the grid-connected converter system includes: Based on the topology of the main circuit, a first mathematical model is constructed; Based on the topology of the control loop, a second mathematical model is constructed; By combining the first mathematical model and the second mathematical model, the target model is obtained.

3. The method according to claim 1, characterized in that, The grid-connected converter circuit includes: a DC-side DC voltage source, a three-phase converter, a filter inductor, a PWM modulator, a filter capacitor, and grid impedance. The control loop includes: a phase-locked loop unit, a coordinate transformation unit, a power calculation unit, a power outer loop control unit, and a current inner loop control unit.

4. The method according to claim 1, characterized in that, The operating condition variable is the transient grid voltage amplitude; The control parameters include the proportional coefficient and integral coefficient of the phase-locked loop; The variables also include grid-connected inductance parameters.

5. The method according to claim 1, characterized in that, The engineering features are physical quantities or dimensionless numbers that are calculated based on the combination of the variables and can reflect the dynamic characteristics of the system.

6. The method according to claim 1, characterized in that, Based on the stability label and the engineering features, the variable combinations in the data pool are divided into a first subset for training a classification task and a second subset for training a regression task, including: The first subset of data is defined as all combinations of variables marked as unstable, and the subset of variables marked as stable that are determined to be close to the stability boundary based on the engineering characteristics. The remaining portion of the variable combinations marked as stable, which is determined to be far from the stable boundary based on the engineering characteristics, is divided into the second subset.

7. The method according to claim 1, characterized in that, The adaptive loss balancing mechanism is used to dynamically adjust the relative weights of the classification task loss and the regression task loss during training.

8. The method according to claim 1, characterized in that, The physical constraint loss is the residual loss calculated from the system of differential-algebraic equations corresponding to the target model.

9. The method according to claim 1, characterized in that, The feasible domain of the parameters includes at least a two-dimensional stable region boundary map with the transient grid voltage amplitude as the abscissa and the phase-locked loop proportional coefficient, integral coefficient or grid-connected inductance as the ordinate.

10. A device for constructing a transient stability domain for a grid-connected converter, characterized in that, include: The first building module is used to construct the target model based on the topology of the grid-connected system with a grid-connected converter. The acquisition module is used to sample and combine at least one operating condition and at least two control parameters as variables within the range of values ​​of the variables to obtain multiple variable combinations; The solution module is used to solve the static operating point of the grid-type converter for each variable combination using the target model, and to mark the variable combination with a stability label according to whether the solution is successful or not, and to store the variable combination, its corresponding stability label and the obtained static operating point in the data pool; wherein, the successful solution is marked as stable, and the unsuccessful solution is marked as unstable. The combination module is used to add engineering features to combinations of variables in the data pool; the engineering features are features generated based on prior knowledge. The partitioning module is used to divide the combination of variables in the data pool into a first subset for training a classification task and a second subset for training a regression task, based on the stability label and the engineering features. The second building module is used to build a multi-task learning model, which includes a classification task head and a regression task head. The training module is used to train the classification task head using the first subset of the dataset and the regression task head using the second subset of the dataset. During the training process, an adaptive loss balancing mechanism is used to coordinate the classification loss and the regression loss. In a preset time period, a physical constraint loss based on the target model is introduced for the training of the regression task to complete the hybrid training of the multi-task learning model. The determination module is used to input different combinations of operating condition variables and control parameters into the trained multi-task learning model, obtain the stability determination results of each combination through the classification task head, and obtain the corresponding static operating point prediction value through the regression task head. The generation module is used to generate a parameter feasible domain that characterizes the stable operating range of the grid-connected converter system based on the stability determination results of all variable combinations and their static operating point prediction values.

Images