Method and device for identifying dominant instability mode of multi-circuit direct current sending-out system
By combining convolutional neural networks and ensemble regression models with Shapley additive interpretation and one-way ANOVA, the problem of identifying the dominant instability mode of a multi-circuit DC transmission system under complex transient instability was solved, achieving higher accuracy and stronger interpretation capabilities for stability assessment and control strategy support.
Patent Information
- Application Number
- CN202511094610.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-06
- Publication Date
- 2025-11-28
AI Technical Summary
Existing technologies struggle to accurately identify the dominant instability modes of multi-circuit DC transmission systems under complex transient instability conditions. Furthermore, traditional methods are inefficient and subjective, failing to meet the needs of online auxiliary decision-making and control parameter optimization for power grid operation.
A convolutional neural network-based model combined with an ensemble regression model and interpretive analysis methods is adopted. A sample set is generated through electromechanical transient time-domain simulation, and the model is trained to identify the dominant instability modes. The contribution and significance of control parameters are quantified by Shapley additive interpretation and one-way ANOVA, so as to achieve accurate identification of instability modes.
It improves the recognition accuracy and interpretation capability under various transient disturbance scenarios, provides stronger stability assessment and control strategy support, and has good engineering practical value.
Smart Images
Figure CN121030331A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of dynamic stability analysis and evaluation of power systems, and in particular to a method and apparatus for identifying the dominant instability mode of a multi-circuit DC transmission system. Background Technology
[0002] The geographical separation between my country's energy resources and power load centers necessitates the large-scale inter-regional transmission of renewable energy through HVDC (High Voltage Direct Current) systems. Against this backdrop, multi-circuit HVDC transmission systems have gradually become an important component of my country's inter-regional power grid.
[0003] In HVDC systems, transient faults such as commutation failure or DC blocking occur frequently. Limited by the reactive power regulation capability and response lag characteristics of the AC side, these faults often induce transient overvoltages in the sending-end grid, seriously threatening the system's voltage stability. Especially when faults occur in adjacent AC systems at the sending end, they can easily further excite power angle oscillations and frequency shifts, thus inducing composite transient instability phenomena of voltage, frequency, and power angle, significantly increasing the risk of system operation. Identifying the dominant instability mode in composite transient instability phenomena can help pinpoint the root cause of system instability and formulate effective stability improvement strategies. However, most existing studies rely on extensive operating condition simulations followed by manual or statistical methods to determine the dominant instability mode and key control parameters, resulting in low efficiency, strong subjectivity, and difficulty in interpretation, failing to meet the practical needs of online auxiliary decision-making and control parameter optimization in power grid operation. In recent years, the application of deep learning in power system stability assessment has gradually emerged, demonstrating excellent performance in pattern recognition and feature extraction; however, most models lack interpretability and are difficult to use for subsequent analysis.
[0004] Therefore, a new technical solution is urgently needed to address the technical problem of accurately identifying the dominant instability mode of a multi-circuit DC transmission system under complex transient instability conditions. Summary of the Invention
[0005] This invention provides a method and apparatus for identifying the dominant instability mode of a multi-circuit DC transmission system, in order to solve the technical problem of how to accurately identify the dominant instability mode of a multi-circuit DC transmission system under complex transient instability conditions.
[0006] To achieve the above objectives, the present invention provides a method for identifying the dominant instability mode of a multi-circuit DC transmission system, characterized by comprising:
[0007] Based on the pre-acquired set of power grid operation modes, transient fault sets, and control parameters, electromechanical transient time-domain simulation is performed to obtain a sample set; the sample set includes the time-series data of preset electrical quantities and the corresponding stable or unstable modes of the power grid;
[0008] A convolutional neural network-based model is trained based on the sample set to obtain the first model; the first model is used to predict the dominant instability mode of the unstable sample to obtain the dominant instability mode; the instability modes include voltage, power angle and frequency instability.
[0009] The contribution of each control parameter in the instability sample is obtained by combining the pre-selected integrated regression model with Shapley additive interpretation. The dominant control parameters under each instability mode are obtained based on the contribution. The significance of the control parameters in the instability sample is analyzed by one-way ANOVA to obtain the significance ranking of the control parameters. The key control parameters under each instability mode are obtained by comparing the dominant control parameters and the significance ranking of the control parameters.
[0010] Preferably, electromechanical transient time-domain simulation is performed based on the pre-acquired set of power grid operating modes, transient fault sets, and control parameters to obtain a sample set including:
[0011] Obtain the set of power grid operation modes, which includes preset combinations of power output from sending-end renewable energy sources and loads at receiving end; obtain the set of transient faults, which includes pre-selected typical power grid faults; obtain the control parameters, which include pre-selected control parameters.
[0012] A preset number of electromechanical transient time-domain simulations are performed on each transient fault of the power grid under various operating modes, combined with control parameters. The time-series response trajectories of voltage, current, active power, and reactive power of each monitoring node of the power grid at a preset number of time segments after encountering a transient fault are recorded in each simulation, and the corresponding stable or unstable mode of the power grid is also recorded. The time-series response trajectory and the corresponding stable or unstable mode of the power grid in one simulation are taken as a sample, and the samples of all simulations constitute a sample set.
[0013] Preferably, the first model obtained by training a convolutional neural network-based model based on the sample set includes:
[0014] The samples in the sample set are extracted according to three channels: voltage, frequency, and power angle, and then uniformly processed by Z-score standardization to transform them into a standard normal distribution with a mean of 0 and a standard deviation of 1, thus obtaining a standard sample set.
[0015] A model based on a two-dimensional convolutional neural network is constructed. The model includes an input layer, two sets of convolutional layers, pooling layers, flattening layers, fully connected layers, and a softmax classification output layer. The network input is a four-dimensional tensor that includes the number of samples, the number of channels, the number of time steps, and the number of nodes.
[0016] The standard sample set is divided into a training set and a test set according to a preset ratio; the model based on the two-dimensional convolutional neural network is trained for a preset number of rounds based on the training set, and supervised training is performed using the cross-entropy loss function. The Adam optimizer is used during the training process; after training, the test set is used for testing until the accuracy reaches a preset value, and the first model is obtained; the first model is used to distinguish voltage, power angle and frequency instability modes.
[0017] Preferably, the contribution of each control parameter in the instability sample is obtained by combining the pre-selected integrated regression model with Shapley's additive interpretation method, including:
[0018] The mapping relationship between the control parameters and the degree of instability of each unstable sample is established based on the pre-selected integrated regression model; the contribution of each control parameter is calculated based on the mapping relationship and the Shapley additive interpretation method.
[0019] The pre-selected ensemble regression models include the random forest regression model, the extreme gradient boosting regression model, the gradient boosting regression model, and the lightweight gradient boosting regression model.
[0020] Preferably, the dominant control parameters for each instability mode, based on their contribution, include:
[0021] Based on the contribution, the distribution characteristics of SHAP values are analyzed and a visualization is generated, including:
[0022]
[0023] Where, φ i denoted by , represents the marginal contribution of the i-th control parameter to the model output, which includes the dominant instability mode of the unstable sample; N is the set of all control parameters; S is a subset that does not include the i-th control parameter; f(·) is the output function of the regression model;
[0024] Based on the visualization analysis of the average contribution and positive / negative influence trends of each control parameter to each instability mode, the dominant control parameters under each instability mode are obtained.
[0025] Preferably, the significance of the control parameters in the unstable sample is analyzed using one-way ANOVA, and the ranking of the significance of the control parameters is as follows:
[0026] Under the instability category, the model output is the dependent variable, and the control parameters are the independent variables. Calculate the sum of squares:
[0027]
[0028] Among them, Y j This is the output of the j-th sample; The output mean of all samples; n is the total number of samples;
[0029] Based on the total sum of squares, the F-test statistic includes:
[0030]
[0031] Among them, SS A To handle the sum of squares, SS E This is the sum of squares of the error terms;
[0032] The p-value is calculated based on the F-test statistic, including:
[0033] p = P(F > F) calc )
[0034] Among them, F calc This refers to the actual calculated F-test statistic value;
[0035] The smaller the p-value of the control parameter, the more significant its impact on the current instability mode;
[0036] Calculate the p-value of each control parameter under each instability mode, and rank the control parameters by significance based on the p-value.
[0037] The present invention also provides a dominant instability mode identification device for a multi-circuit DC transmission system, used in the method of the present invention, characterized in that the device includes a first module, a second module, a third module and a fourth module;
[0038] The first module is used to perform electromechanical transient time-domain simulation based on the pre-acquired set of power grid operation modes, transient fault sets, and control parameters to obtain a sample set; the sample set includes the time-series data of preset electrical quantities and the corresponding stable or unstable modes of the power grid;
[0039] The second module is used to train a convolutional neural network-based model based on the sample set to obtain the first model; and to predict the dominant instability mode of the unstable sample based on the first model; the instability mode includes voltage, power angle and frequency instability.
[0040] The third module is used to obtain the contribution of each control parameter in the instability sample based on the pre-selected integrated regression model and Shapley additive interpretation method; to obtain the dominant control parameter in each instability mode based on the contribution; and to conduct a significant impact analysis on the control parameters of the instability sample using one-way ANOVA to obtain the significance ranking of the control parameters.
[0041] The fourth module is used for comparative analysis based on the dominant control parameters and the ranking of control parameters by significance, to obtain the key control parameters under each instability mode.
[0042] The present invention has the following beneficial effects:
[0043] The dominant instability mode identification method for multi-circuit DC transmission systems of this invention can accurately identify instability modes such as voltage, frequency, and power angle under various transient disturbance scenarios. It combines Shapley additive interpretation and one-way ANOVA to analyze and quantify the influence of each control parameter. Compared with traditional methods, this invention's method has higher classification accuracy, stronger interpretability, and a wider range of applications, providing effective support for stability assessment, parameter optimization, and control strategy formulation. It has good engineering practical value and promising prospects for widespread application.
[0044] The dominant instability mode identification device for the multi-DC transmission system of the present invention, when used in the method of the present invention, has the same beneficial effects as the method of the present invention.
[0045] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description
[0046] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:
[0047] Figure 1 This is a schematic diagram of the method flow of a preferred embodiment of the present invention. Detailed Implementation
[0048] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.
[0049] See Figure 1 In a preferred embodiment of the present invention, a method for identifying the dominant instability mode of a multi-circuit DC transmission system is provided, comprising:
[0050] S1. Perform electromechanical transient time-domain simulation based on the pre-acquired power grid operation mode set, transient fault set and control parameters to obtain a sample set; the sample set includes the time-series data of preset electrical quantities and the corresponding stable or unstable modes of the power grid.
[0051] In a preferred embodiment of the present invention, electromechanical transient time-domain simulation is performed based on a pre-acquired set of power grid operating modes, a set of transient faults, and control parameters to obtain a sample set including:
[0052] Obtain the set of power grid operation modes, which includes preset combinations of power output from sending-end renewable energy sources and loads at receiving end; obtain the set of transient faults, which includes pre-selected typical power grid faults, such as N-1 short circuits, DC commutation failures, and blocking; obtain the control parameters, which include pre-selected control parameters, such as low-voltage ride-through control and VDCOL control.
[0053] For each transient fault of the power grid under various operating modes, a preset number of electromechanical transient time-domain simulations are performed in conjunction with control parameters. The time-series response trajectories of voltage, current, active power, and reactive power of each monitoring node of the power grid (pre-selected sending system nodes) at a preset number of time sections after encountering a transient fault are recorded in each simulation, and the corresponding stable or unstable mode of the power grid is recorded. The time-series response trajectory and the corresponding stable or unstable mode of the power grid in one simulation are taken as a sample, and the samples of all simulations constitute a sample set.
[0054] S2. Train a convolutional neural network-based model based on the sample set to obtain the first model; predict the dominant instability mode of the unstable sample based on the first model; the instability mode includes voltage, power angle and frequency instability.
[0055] In a preferred embodiment of the present invention, training a convolutional neural network-based model based on a sample set to obtain a first model includes:
[0056] The samples in the sample set are extracted according to three channels: voltage, frequency, and power angle. Z-score standardization is then applied to transform them into a standard normal distribution with a mean of 0 and a standard deviation of 1, resulting in a standard sample set. Z-score standardization includes:
[0057]
[0058] Where x is the original eigenvalue, μ is the mean of the eigenvalue, and σ is the standard deviation of the eigenvalue.
[0059] A model based on a two-dimensional convolutional neural network is constructed. The model includes an input layer, two sets of convolutional layers, pooling layers, flattening layers, fully connected layers, and a softmax classification output layer. The network input is a four-dimensional tensor that includes the number of samples, the number of channels, the number of time steps, and the number of nodes.
[0060] The standard sample set is divided into a training set and a test set according to a preset ratio; the model based on the two-dimensional convolutional neural network is trained for a preset number of rounds based on the training set, and supervised training is performed using the cross-entropy loss function. The Adam optimizer is used during the training process; after training, the test set is used for testing until the accuracy reaches a preset value, and the first model is obtained; the first model is used to distinguish voltage, power angle and frequency instability modes.
[0061] S3. Based on the pre-selected integrated regression model and Shapley additive interpretation method, obtain the contribution of each control parameter in the instability sample; based on the contribution, obtain the dominant control parameter in each instability mode.
[0062] In a preferred embodiment of the present invention, the contribution of each control parameter in the instability sample is obtained by combining a pre-selected ensemble regression model with Shapley's additive interpretation method, including:
[0063] The mapping relationship between the control parameters and the degree of instability of each unstable sample is established based on the pre-selected integrated regression model; the contribution of each control parameter is calculated based on the mapping relationship and the Shapley additive interpretation method.
[0064] The pre-selected ensemble regression models include the random forest regression model, the extreme gradient boosting regression model, the gradient boosting regression model, and the lightweight gradient boosting regression model.
[0065] In a preferred embodiment of the present invention, the dominant control parameters for each instability mode, obtained based on the contribution degree, include:
[0066] Based on the contribution, the distribution characteristics of SHAP values are analyzed and a visualization is generated, including:
[0067]
[0068] Where, φ i denoted by , represents the marginal contribution of the i-th control parameter to the model output, which includes the dominant instability mode of the unstable sample; N is the set of all control parameters; S is a subset that does not include the i-th control parameter; f(·) is the output function of the regression model;
[0069] Based on the visualization analysis of the average contribution and positive / negative influence trends of each control parameter to each instability mode, the dominant control parameters under each instability mode are obtained.
[0070] S4. Using one-way ANOVA, a significance analysis of the control parameters in the unstable sample is conducted to obtain a ranking of the control parameters by significance. S4 specifically includes:
[0071] Under the instability category, the model output is the dependent variable, and the control parameters are the independent variables. Calculate the sum of squares:
[0072]
[0073] Among them, Y j This is the output of the j-th sample; The output mean of all samples; n is the total number of samples;
[0074] Based on the total sum of squares, the F-test statistic includes:
[0075]
[0076] Among them, SS A To handle the sum of squares, SS E This is the sum of squares of the error terms;
[0077] The p-value is calculated based on the F-test statistic, including:
[0078] p = P(F > F) calc )
[0079] Among them, F calc This refers to the actual calculated F-test statistic value;
[0080] The smaller the p-value of the control parameter, the more significant its impact on the current instability mode;
[0081] Calculate the p-value of each control parameter under each instability mode, and rank the control parameters by significance based on the p-value.
[0082] In a preferred embodiment of the present invention, the statistical process analyzes three types of modes: voltage instability, frequency instability, and power angle instability, and cross-validates them with the SHAP ranking results, thereby improving the reliability and robustness of the dominant parameter identification results.
[0083] S5. Based on the dominant control parameters and the ranking of the significance of the control parameters, a comparative analysis is conducted to obtain the key control parameters under each instability mode.
[0084] The dominant instability mode identification method for multi-circuit DC transmission systems of this invention can accurately identify instability modes such as voltage, frequency, and power angle under various transient disturbance scenarios. It combines Shapley additive interpretation and one-way ANOVA to analyze and quantify the influence of each control parameter. Compared with traditional methods, this invention's method has higher classification accuracy, stronger interpretability, and a wider range of applications, providing effective support for stability assessment, parameter optimization, and control strategy formulation. It has good engineering practical value and promising prospects for widespread application.
[0085] In a preferred embodiment of the present invention, a dominant instability mode identification device for a multi-circuit DC transmission system is also provided for use in the method of the present invention. The device includes a first module, a second module, a third module, and a fourth module.
[0086] The first module is used to perform electromechanical transient time-domain simulation based on the pre-acquired set of power grid operation modes, transient fault sets, and control parameters to obtain a sample set; the sample set includes the time-series data of preset electrical quantities and the corresponding stable or unstable modes of the power grid;
[0087] The second module is used to train a convolutional neural network-based model based on the sample set to obtain the first model; and to predict the dominant instability mode of the unstable sample based on the first model; the instability mode includes voltage, power angle and frequency instability.
[0088] The third module is used to obtain the contribution of each control parameter in the instability sample based on the pre-selected integrated regression model and Shapley additive interpretation method; to obtain the dominant control parameter in each instability mode based on the contribution; and to conduct a significant impact analysis on the control parameters of the instability sample using one-way ANOVA to obtain the significance ranking of the control parameters.
[0089] The fourth module is used for comparative analysis based on the dominant control parameters and the ranking of control parameters by significance, to obtain the key control parameters under each instability mode.
[0090] The dominant instability mode identification device for the multi-DC transmission system of the present invention, when used in the method of the present invention, has the same beneficial effects as the method of the present invention.
[0091] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for identifying the dominant instability mode of a multi-circuit DC transmission system, characterized in that, include: Based on the pre-acquired set of power grid operation modes, transient fault sets, and control parameters, electromechanical transient time-domain simulation is performed to obtain a sample set; The sample set includes time-series data of preset electrical quantities and the corresponding stable or unstable modes of the power grid. A first model is obtained by training a convolutional neural network-based model based on the sample set; the dominant instability mode of the unstable sample is obtained by predicting the sample set based on the first model; the instability mode includes voltage, power angle and frequency instability. The contribution of each control parameter in the instability sample is obtained by combining the pre-selected integrated regression model with Shapley additive interpretation; the dominant control parameter in each instability mode is obtained based on the contribution; the significance of the control parameters in the instability sample is analyzed by one-way ANOVA to obtain the significance ranking of the control parameters; the key control parameters in each instability mode are obtained by comparing the dominant control parameters and the significance ranking of the control parameters.
2. The method for identifying the dominant instability mode of a multi-circuit DC transmission system according to claim 1, characterized in that, The electromechanical transient time-domain simulation, based on the pre-acquired power grid operation mode set, transient fault set, and control parameters, yields a sample set including: Obtain the set of power grid operation modes, which includes preset combinations of power output from sending-end renewable energy sources and loads at receiving ends; obtain the set of transient faults, which includes pre-selected typical power grid faults; obtain the control parameters, which include pre-selected control parameters. For each transient fault of the power grid under various operating modes, a preset number of electromechanical transient time-domain simulations are performed in conjunction with the control parameters; the time-series response trajectories of voltage, current, active power, and reactive power of each monitoring node of the power grid at a preset number of time segments after encountering a transient fault are recorded in each simulation, and the corresponding stable or unstable mode of the power grid is recorded; the time-series response trajectory and the corresponding stable or unstable mode of the power grid in one simulation are taken as a sample, and the sample set is composed of all simulation samples.
3. The method for identifying the dominant instability mode of a multi-circuit DC transmission system according to claim 2, characterized in that, The first model, obtained by training a convolutional neural network-based model using the aforementioned sample set, includes: The samples in the sample set are extracted according to three channels: voltage, frequency, and power angle, and then uniformly processed by Z-score standardization to transform them into a standard normal distribution with a mean of 0 and a standard deviation of 1, thus obtaining a standard sample set. A model based on a two-dimensional convolutional neural network is constructed. The model includes an input layer, two sets of convolutional layers, a pooling layer, a flattening layer, a fully connected layer, and a softmax classification output layer. The network input is a four-dimensional tensor including the number of samples, the number of channels, the number of time steps, and the number of nodes. The standard sample set is divided into a training set and a test set according to a preset ratio; the model based on the two-dimensional convolutional neural network is trained for a preset number of rounds according to the training set, and supervised training is performed using the cross-entropy loss function. The Adam optimizer is used during the training process; after training, the model is tested using the test set until the accuracy reaches a preset value, and the first model is obtained; the first model is used to distinguish voltage, power angle and frequency instability modes.
4. The method for identifying the dominant instability mode of a multi-circuit DC transmission system according to claim 3, characterized in that, The contribution of each control parameter in the unstable sample, obtained by combining the pre-selected integrated regression model with Shapley's additive interpretation method, includes: A mapping relationship between the control parameters and the degree of instability of each unstable sample is established based on the pre-selected integrated regression model; the contribution of each control parameter is calculated based on the mapping relationship and the Shapley additive interpretation method. The pre-selected ensemble regression models include random forest regression model, extreme gradient boosting regression model, gradient boosting regression model, and lightweight gradient boosting regression model.
5. The method for identifying the dominant instability mode of a multi-circuit DC transmission system according to claim 4, characterized in that, The dominant control parameters for each instability mode, based on the aforementioned contribution, include: Based on the contribution level, SHAP value distribution characteristics are analyzed and a visualization is generated, including: Where, φ i denoted by , represents the marginal contribution of the i-th control parameter to the model output, which includes the dominant instability mode of the unstable sample; N is the set of all control parameters; S is a subset that does not include the i-th control parameter; f(·) is the output function of the regression model; Based on the visualization analysis, the average contribution and positive / negative influence trends of each control parameter on each instability mode are obtained, thus identifying the dominant control parameters under each instability mode.
6. The method for identifying the dominant instability mode of a multi-circuit DC transmission system according to claim 5, characterized in that, The significance analysis of the control parameters of the unstable sample based on one-way ANOVA, and the resulting ranking of the significance of the control parameters, include: Under the instability category, the model output is the dependent variable, and the control parameters are the independent variables. Calculate the sum of squares: Among them, Y j This is the output of the j-th sample; The output mean of all samples; n is the total number of samples; Based on the total sum of squares, the F-test statistic includes: Among them, SS A To handle the sum of squares, SS E This is the sum of squares of the error terms; The p-value is calculated based on the F-test statistic, including: p=P(F>F calc ) Among them, F calc This refers to the actual calculated F-test statistic value; The smaller the p-value of the control parameter, the more significant its impact on the current instability mode; Calculate the p-value of each control parameter under each instability mode, and rank the control parameters by significance based on the p-value.
7. A dominant instability mode identification device for a multi-circuit DC transmission system, used in the method described in any one of claims 1 to 6, characterized in that, The device includes a first module, a second module, a third module, and a fourth module; The first module is used to perform electromechanical transient time-domain simulation based on the pre-acquired set of power grid operation modes, transient fault sets, and control parameters to obtain a sample set; the sample set includes time-series data of preset electrical quantities and the corresponding stable or unstable modes of the power grid; The second module is used to train a convolutional neural network-based model based on the sample set to obtain a first model; and to predict the dominant instability mode of the unstable sample based on the first model; the instability mode includes voltage, power angle and frequency instability. The third module is used to obtain the contribution of each control parameter in the unstable sample based on the pre-selected integrated regression model and Shapley additive interpretation method; to obtain the dominant control parameter under each unstable mode based on the contribution; and to perform a significant impact analysis on the control parameters of the unstable sample using one-way ANOVA to obtain the significance ranking of the control parameters. The fourth module is used to perform comparative analysis based on the dominant control parameter and the significance ranking of the control parameters to obtain the key control parameters under each instability mode.