Convolutional neural network-based sending-end power grid power oscillation suppression effect evaluation method and system, and computer readable storage medium

Through a convolutional neural network-based method, combining steady-state and transient feature data, the power oscillation suppression effect of flexible DC converter stations is evaluated, which solves the problems of inaccurate evaluation and low simulation efficiency in the prior art, and realizes the send-in Efficiency, reliability and stability evaluation and control of the power grid.

CN119029927APending Publication Date: 2024-11-26HUNAN UNIV +3
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411114720.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The existing power oscillation suppression effect evaluation technology for different control methods of flexible DC converter stations is not accurate enough and has low simulation efficiency, making it difficult to achieve efficient and reliable power oscillation suppression in the power transmission terminal in high altitude areas.

Method used

Using a method based on convolutional neural network, the convolutional neural network model is constructed and trained, and the fault limit resection time is solved by dichotomy. Combining steady-state and transient feature data, the system stability margin is evaluated to provide power oscillation suppression effect. quantitative evaluation.

Benefits of technology

It significantly improves the accuracy and efficiency of evaluation, simplifies the system modeling process, reduces the computing resource requirements, improves the stability and reliability of the power grid at the end, and supports optimized operation and stable control.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119029927A_ABST
    Figure CN119029927A_ABST
Patent Text Reader

Abstract

The invention discloses a convolutional neural network-based sending-end power grid power oscillation suppression effect evaluation method and system and a computer readable storage medium, and the method comprises the following steps: inputting to-be-evaluated data into a trained convolutional neural network CNN model, and obtaining an evaluation result; the specific content of constructing and training the CNN model comprises the following steps: generating a stability calculation sample set Xm for stability analysis and power oscillation suppression effect evaluation; solving a fault limit clearing time CCT value of each sample in the Xm by using a bisection method, and comparing the fault limit clearing time CCT value with the actual fault clearing time Tac to obtain a stability margin Tm of the system; acquiring a transient feature Fm, integrating Xm and Fm into an input data set, and taking the stability margin Tm as an output data set; and constructing and training a convolutional neural network (CNN) model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method, system and computer-readable storage medium for evaluating the effect of power oscillation suppression of a sending-end power grid based on a convolutional neural network. Background Art

[0002] High-altitude areas with sufficient sunshine and strong solar radiation have natural advantages in developing photovoltaics. If there are abundant hydropower resources in the region at the same time, the prospects for the development and utilization of clean energy are broad. If the power grid in the high-altitude area is used as the sending-end power grid, if clean energy such as photovoltaics and hydropower are transmitted to areas with heavy power supply tasks through appropriate transmission channels, it will be expected to alleviate the power supply tension in the area. However, the current sending-end power grid structure in high-altitude areas is generally weak, the power transmission capacity of existing AC transmission channels is limited, and the collection and transmission of clean energy is very difficult, so flexible DC transmission technology is required. Since photovoltaic and other clean energy power generation units and flexible DC converter stations contain a large number of power electronic devices and complex control links, there is mutual coupling between the control links of different devices. When the sending-end power grid is severely disturbed, complex power oscillation phenomena may occur, causing the sending-end power grid to become unstable. How to implement efficient and reliable power oscillation suppression effect evaluation on the sending-end power grid in massive operation scenarios based on various possible control methods of flexible DC converter stations is one of the key challenges faced in the actual operation of the sending-end power grid.

[0003] The existing power oscillation suppression effect evaluation technology for different control modes of flexible DC converter stations mainly adopts the equivalent model method based on mathematical analysis and the time domain simulation method based on numerical analysis. The equivalent model method can simplify the modeling process of the system to a certain extent and has a certain theoretical basis, but it ignores some important nonlinear characteristics and dynamic changes, resulting in inaccurate evaluation results. With the continuous expansion of the system scale and the increasing complexity of the grid structure, the accurate establishment of the system equivalent model has become more difficult and complex, and the applicability and versatility of the model are limited. The time domain simulation method simulates the system in the time domain through numerical calculation means to obtain the dynamic response of the system under different control modes, thereby evaluating the power oscillation suppression effect. Although this method is relatively intuitive and can provide relatively reliable evaluation results, it usually requires a lot of computing resources and time, especially when dealing with large-scale power systems and complex operation scenarios, the simulation efficiency is low. Summary of the invention

[0004] In view of this, the present invention provides a method, system and computer-readable storage medium for evaluating the power oscillation suppression effect of a sending-end power grid based on a convolutional neural network, which are used to at least solve the problems of insufficient accuracy and low simulation efficiency of the existing power oscillation suppression effect evaluation technology for different control modes of flexible DC converter stations.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A method for evaluating the power oscillation suppression effect of a sending-end power grid based on a convolutional neural network comprises the following steps:

[0007] Input the data to be evaluated into the trained convolutional neural network (CNN) model to obtain the evaluation results. The specific contents of constructing and training the convolutional neural network (CNN) model include:

[0008] According to the various operation modes of the sending-end power grid R j , transient fault F k And different control modes of flexible DC converter stations at the sending end C n , generate a stable calculation sample set X for stability analysis and power oscillation suppression effect evaluation m ; where j = 1, 2, ..., M; k = 1, 2, ..., P; n = 1, 2, ..., Q; X m = {R j ,F k ,C n}, m = 1, 2, ..., M*P*Q;

[0009] Solve for X using the binary method m The fault limit removal time CCT value of each sample in the fault removal time is compared with the actual fault removal time T ac Compare and get the system stability margin T m ;

[0010] The time series data of voltage U, active power P, reactive power Q or frequency f of each node are sampled to form a transient response time window. The sampling interval [T0,T fault ] covers the three moments before the fault, the fault occurrence and the fault removal; the transient response time window is composed of the transient characteristics F of the sample m , X m As steady-state characteristics and transient characteristics F m Integrate the input data set into the stability margin T m As the output data set; wherein each node includes each load node and each generator node;

[0011] Construct a convolutional neural network (CNN) model, train the CNN model through input data sets and output data sets, and optimize the parameters of the CNN model to obtain a trained CNN model.

[0012] Preferably, according to various operation modes R of the sending-end power grid j , transient fault F k And different control modes of flexible DC converter stations at the sending end C n, generate a stable calculation sample set X for stability analysis and power oscillation suppression effect evaluation m The specific contents include:

[0013] Operation Mode R j Including setting different load levels, start and stop states of different generator sets or different grid topologies, j = 1, 2, ..., M;

[0014] Transient fault F k Including three-phase short circuit fault, single-phase short circuit fault, DC single-pole blocking or DC double-pole blocking, k = 1, 2, ..., P;

[0015] Different control modes of sending-end flexible DC converter station C n Including active power and reactive power control methods, n = 1, 2, ..., Q;

[0016] On the simulation platform, batch build stable calculation sample sets X, X including different operation modes, transient faults, and control modes of the sending-end flexible DC converter station m = {R j ,F k ,C n}, m=1,2,...,M*P*Q.

[0017] Preferably, the binary search method is used to solve X m The fault limit removal time CCT value of each sample in the fault removal time is compared with the actual fault removal time T ac Compare and get the system stability margin T m The specific contents include:

[0018] Set the solution range of batch calculation CCT value for all samples [T min ,T max ];

[0019] The fault removal time range is gradually reduced by iterative binary division: In each iterative calculation, the current intermediate value T is obtained. i =(T min +T max ) / 2, for each sample at the fault removal time T i After each simulation, the calculation result file is read to obtain the maximum power angle difference Δδ between any two generators within the simulation time interval. max And judge the stability of the system by the power angle stability criterion;

[0020] If the system is in T i If the current T i Update T min Otherwise, the current T i Update T max ;

[0021] Until the updated T min and T max When the difference between is less than the preset threshold ε, the current T i For X m The fault limit clearing time CCT i ;

[0022] CCT i The actual fault removal time T ac Compare, record the stability state of the system, and calculate the stability margin T of the system m =CCT i -T ac , T m As X m A quantitative evaluation index of the system power oscillation suppression effect.

[0023] Preferably, the content of judging the stability of the system by the power angle stability criterion specifically includes:

[0024] The power angle stability criterion is:

[0025]

[0026] When δ TSI When <0, it means that under the current disturbance, the system transient power angle is unstable, otherwise the system remains stable;

[0027] The stability state of the system is stored in the state information S in the form of a 0-1 variable, where S = 1 means that the system is stable at the current T i When S=0, it means the system is stable at the current T i Unstable down.

[0028] Preferably, the steady-state features X are analyzed before integration into the input dataset. m With transient characteristics F m Normalize the feature data to ensure that the data are compared on the same scale;

[0029] Both the input and output datasets are randomly divided into training and test sets.

[0030] Preferably, the specific contents of constructing the convolutional neural network model include:

[0031] The input layer is used to receive the input data set and pass the data to the convolutional layer;

[0032] The convolution layer includes at least one single convolution layer, each of which extracts features of different levels of data layer by layer. A batch normalization layer and a ReLU activation layer are set after each single convolution layer.

[0033] The batch normalization layer normalizes the final output of the convolution layer to reduce the difference between the features of different batches; the ReLU activation layer performs nonlinear transformation on the features output by the batch normalization layer;

[0034] The pooling layer downsamples the feature data after nonlinear transformation;

[0035] The Dropout layer randomly discards some neurons during training to ensure the robustness of the model;

[0036] The feature data after downsampling in the pooling layer is flattened into a one-dimensional vector through the fully connected layer, and the predicted value is output through the regression layer to achieve the regression task.

[0037] Preferably, after building the CNN model, the network architecture is set: the initial learning rate, batch size and maximum number of training times of the CNN model are set, the training set is shuffled in each training cycle and a training process curve is drawn.

[0038] Preferably, the trained CNN model is predicted using the test set, and the root mean square error RMSE and determination coefficient R of the CNN model on the test set are calculated. 2 ,in:

[0039]

[0040] Among them, y (i) is the true value of the sample, is the sample prediction value, is the sample mean, m is the number of samples;

[0041] The RMSE and R 2 To evaluate the generalization ability of the model, the hyperparameters of the model are further adjusted according to the evaluation results to obtain the final CNN model.

[0042] Preferably, a system for evaluating the effect of power oscillation suppression of a sending-end power grid based on a convolutional neural network comprises: an input module and a model building and training module;

[0043] The input module is used to input the data to be evaluated into the trained convolutional neural network (CNN) model to obtain the evaluation results;

[0044] The model building and training module is used to build and train the convolutional neural network (CNN) model, including:

[0045] According to the various operation modes of the sending-end power grid R j , transient fault F k And different control modes of flexible DC converter stations at the sending end C n , generate a stable calculation sample set X for stability analysis and power oscillation suppression effect evaluation m; where j = 1, 2, ..., M; k = 1, 2, ..., P; n = 1, 2, ..., Q; X m = {R j ,F k ,C n}, m = 1, 2, ..., M*P*Q;

[0046] Solve for X using the binary method m The fault limit removal time CCT value of each sample in the fault removal time is compared with the actual fault removal time T ac Compare and get the system stability margin T m ;

[0047] The time series data of voltage U, active power P, reactive power Q or frequency f of each node are sampled to form a transient response time window. The sampling interval [T0,T fault ] covers the three moments before the fault, the fault occurrence and the fault removal; the transient response time window is composed of the transient characteristics F of the sample m , X m As steady-state characteristics and transient characteristics F m Integrate the input data set into the stability margin T m As the output data set; wherein each node includes each load node and each generator node;

[0048] Construct a convolutional neural network (CNN) model, train the CNN model through input data sets and output data sets, and optimize the parameters of the CNN model to obtain a trained CNN model.

[0049] Preferably, a computer-readable storage medium stores a computer program thereon, characterized in that the program implements the steps of any of the above methods when executed by a processor.

[0050] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method, system and computer-readable storage medium for evaluating the power oscillation suppression effect of a sending-end power grid based on a convolutional neural network, which has the following beneficial effects:

[0051] From the perspective of data-driven, the massive amount of power grid operation data is fully utilized, and the key system power oscillation characteristics and laws are automatically extracted from it. Compared with the traditional evaluation method based on the equivalent model method and the time domain simulation method, the present invention does not require complex system modeling, greatly simplifies the workflow, and reduces the demand for computing resources. By training and analyzing large-scale data, the present invention can quickly and accurately evaluate the power oscillation suppression effect of different sending-end flexible DC converter station control modes on the system, and provide strong technical support for the optimized operation and stable control of the sending-end power grid. In addition, the convolutional neural network model can improve the accuracy and efficiency of the evaluation in a complex sending-end power grid environment by extracting deep features from the operation data, and significantly improve the stability and reliability of the system. Through the present invention, the sending-end power grid staff can more accurately evaluate the effects of various converter station control modes on the power oscillation suppression of the sending-end power grid, thereby selecting the optimal sending-end converter station control mode and improving the stability and reliability of the system. The present invention not only helps to enhance the stability and reliability of the power transmission of the sending-end power grid, but also provides important technical support for the safe and stable operation of the future cross-regional interconnected power system containing a high proportion of new energy, and has outstanding application potential and value. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0053] Figure 1 It is a flowchart of a method for evaluating the effect of power oscillation suppression of a sending-end power grid based on a convolutional neural network disclosed in the present invention;

[0054] Figure 2 It is a flowchart of solving CCT by binary division in an embodiment of the present invention;

[0055] Figure 3 This is a structural diagram of the convolutional neural network model built in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] The present invention provides a method for evaluating the power oscillation suppression effect of a sending-end power grid based on a convolutional neural network, comprising the following steps:

[0058] Input the data to be evaluated into the trained convolutional neural network (CNN) model to obtain the evaluation results. The specific contents of constructing and training the convolutional neural network (CNN) model include:

[0059] According to the various operation modes of the sending-end power grid R j , transient fault F k And different control modes of flexible DC converter stations at the sending end C n , generate a stable calculation sample set X for stability analysis and power oscillation suppression effect evaluation m ; where j = 1, 2, ..., M; k = 1, 2, ..., P; n = 1, 2, ..., Q; X m = {R j ,F k ,C n}, m = 1, 2, ..., M*P*Q;

[0060] Solve for X using the binary method m The fault limit removal time CCT value of each sample in the fault removal time is compared with the actual fault removal time T ac Compare and get the system stability margin T m ;

[0061] The time series data of voltage U, active power P, reactive power Q or frequency f of each node are sampled to form a transient response time window. The sampling interval [T0,T fault ] covers the three moments before the fault, the fault occurrence and the fault removal; the transient response time window is composed of the transient characteristics F of the sample m , X m As steady-state characteristics and transient characteristics F m Integrate the input data set into the stability margin T m As the output data set; wherein each node includes each load node and each generator node;

[0062] Construct a convolutional neural network (CNN) model, train the CNN model through input data sets and output data sets, and optimize the parameters of the CNN model to obtain a trained CNN model.

[0063] It should be noted that:

[0064] In this embodiment, 100 operation modes are constructed based on the PSD-BPA platform, and a three-phase AC short circuit fault is simulated at [10%, 30%, 50%, 70%, 90%] of the front-end node on 6 AC lines. A total of 8 control mode combinations of the converter station are set; the stability calculation program is called through MATLAB software and the CCT is solved by the dichotomy method. The specific solution process is as follows: Figure 2 In this embodiment, the sample set is set to have a fault at 0.1s, the actual removal time of a typical fault is 0.1s, and the sampling interval is set to [0.05s, 0.25s]. A convolutional neural network CNN model is constructed to perform regression learning on the input data set and the output data set.

[0065] In order to further implement the above technical solution, according to the various operation modes of the sending-end power grid R j , transient fault F k And different control modes of flexible DC converter stations at the sending end C n , generate a stable calculation sample set X for stability analysis and power oscillation suppression effect evaluation m The specific contents include:

[0066] Operation Mode R j Including setting different load levels, start and stop states of different generator sets or different grid topologies, j = 1, 2, ..., M;

[0067] Transient fault F k Including three-phase short circuit fault, single-phase short circuit fault, DC single-pole blocking or DC double-pole blocking, k = 1, 2, ..., P;

[0068] Different control modes of sending-end flexible DC converter station C n Including active power and reactive power control methods, n = 1, 2, ..., Q;

[0069] On the simulation platform, batch build stable calculation sample sets X, X including different operation modes, transient faults, and control modes of the sending-end flexible DC converter station m = {R j ,F k ,C n}, m=1,2,...,M*P*Q.

[0070] It should be noted that:

[0071] In actual application, the active type can adopt fixed active power control, fixed DC voltage control, and fixed frequency control; the reactive type can adopt fixed reactive power control and fixed AC voltage control.

[0072] In order to further implement the above technical solution, the binary method is used to solve X mThe fault limit removal time CCT value of each sample in the fault removal time is compared with the actual fault removal time T ac Compare and get the system stability margin T m The specific contents include:

[0073] Set the solution range of batch calculation CCT value for all samples [T min ,T max ];

[0074] The fault removal time range is gradually reduced by iterative binary division: In each iterative calculation, the current intermediate value T is obtained. i =(T min +T max ) / 2, for each sample at the fault removal time T i After each simulation, the calculation result file is read to obtain the maximum power angle difference Δδ between any two generators within the simulation time interval. max And judge the stability of the system by the power angle stability criterion;

[0075] If the system is in T i If the current T i Update T min Otherwise, the current T i Update T max ;

[0076] Until the updated T min and T max When the difference between is less than the preset threshold ε, the current T i For X m The fault limit clearing time CCT i ;

[0077] CCT i The actual fault removal time T ac Compare, record the stability state of the system, and calculate the stability margin T of the system m =CCT i -T ac , T m As X m A quantitative evaluation index of the system power oscillation suppression effect.

[0078] In order to further implement the above technical solution, the content of judging the stability of the system through the power angle stability criterion specifically includes:

[0079] The power angle stability criterion is:

[0080]

[0081] When δ TSIWhen <0, it means that under the current disturbance, the system transient power angle is unstable, otherwise the system remains stable;

[0082] The stability state of the system is stored in the state information S in the form of a 0-1 variable, where S = 1 means that the system is stable at the current T i When S=0, it means the system is stable at the current T i Unstable down.

[0083] In order to further implement the above technical solution, the steady-state feature X is m With transient characteristics F m Normalize the feature data to ensure that the data are compared on the same scale;

[0084] Both the input and output datasets are randomly divided into training and test sets.

[0085] It should be noted that:

[0086] The sample set is randomly divided into training set and test set, with 80% as training set and 20% as test set, to ensure that the model has good generalization ability and stable evaluation performance.

[0087] In order to further implement the above technical solution, the specific contents of building a convolutional neural network model include:

[0088] The input layer is used to receive the input data set and pass the data to the convolutional layer;

[0089] The convolution layer includes at least one single convolution layer, each of which extracts features of different levels of data layer by layer. A batch normalization layer and a ReLU activation layer are set after each single convolution layer.

[0090] The batch normalization layer normalizes the final output of the convolution layer to reduce the difference between the features of different batches; the ReLU activation layer performs nonlinear transformation on the features output by the batch normalization layer;

[0091] The pooling layer downsamples the feature data after nonlinear transformation;

[0092] The Dropout layer randomly discards some neurons during training to ensure the robustness of the model;

[0093] The feature data after downsampling in the pooling layer is flattened into a one-dimensional vector through the fully connected layer, and the predicted value is output through the regression layer to achieve the regression task.

[0094] It should be noted that:

[0095] The multi-layer 2D convolutional layer in this model is a 3x3 convolution kernel, which is used to extract different levels of features of the data layer by layer to improve the expression ability and prediction performance of the model; the batch normalization layer standardizes the output of the convolutional layer to reduce the difference between different batches and accelerate the training process. The ReLU activation layer introduces a nonlinear activation function to help the network learn complex mapping relationships to improve the model's ability to fit complex data; the maximum pooling layer is added to reduce the size of the feature map, retain the main features, and reduce the computational complexity; the Dropout layer randomly discards some neurons during the training process to prevent overfitting and enhance the generalization ability of the model. Finally, the feature map extracted by the convolutional layer is flattened through the fully connected layer, the high-dimensional features are mapped to the target output, and the predicted values ​​are output through the regression layer to achieve the regression task. The CNN model structure diagram built in this embodiment is shown in the figure. Figure 3 shown.

[0096] In order to further implement the above technical solution, the network architecture is set after the CNN model is built: the initial learning rate, batch size and maximum number of training times of the CNN model are set, the training set is shuffled in each training cycle and the training process curve is drawn.

[0097] It should be noted that:

[0098] In this embodiment, the Adam optimizer and the gradient descent algorithm are used to train the model. The Adam optimizer combines the advantages of momentum optimization and RMSProp optimization methods, has the characteristics of adaptive learning rate and momentum, and does not require excessive manual parameter adjustment.

[0099] In order to further implement the above technical solution, the trained CNN model is predicted using the test set, and the root mean square error RMSE and determination coefficient R of the CNN model on the test set are calculated. 2 ,in:

[0100]

[0101] Among them, y (i) is the true value of the sample, is the sample prediction value, is the sample mean, m is the number of samples;

[0102] The RMSE and R 2 To evaluate the generalization ability of the model, the hyperparameters of the model are further adjusted according to the evaluation results to obtain the final CNN model.

[0103] It should be noted that:

[0104] The RMSE and R 2To evaluate the generalization ability of the model, further adjust the model's hyperparameters, such as learning rate, batch size, number of convolution kernels, etc., based on the evaluation results, and further improve the model performance by adjusting these parameters. Save the trained model as a power oscillation suppression effect evaluation model for the sending-end power grid suitable for massive operation scenarios for subsequent use and deployment.

[0105] In order to further implement the above technical solution, a system for evaluating the power oscillation suppression effect of a sending-end power grid based on a convolutional neural network includes: an input module and a model building and training module;

[0106] The input module is used to input the data to be evaluated into the trained convolutional neural network (CNN) model to obtain the evaluation results;

[0107] The model building and training module is used to build and train the convolutional neural network (CNN) model, including:

[0108] According to the various operation modes of the sending-end power grid R j , transient fault F k And different control modes of flexible DC converter stations at the sending end C n , generate a stable calculation sample set X for stability analysis and power oscillation suppression effect evaluation m ; where j = 1, 2, ..., M; k = 1, 2, ..., P; n = 1, 2, ..., Q; X m = {R j ,F k ,C n}, m = 1, 2, ..., M*P*Q;

[0109] Solve for X using the binary method m The fault limit removal time CCT value of each sample in the fault removal time is compared with the actual fault removal time T ac Compare and get the system stability margin T m ;

[0110] The time series data of voltage U, active power P, reactive power Q or frequency f of each node are sampled to form a transient response time window. The sampling interval [T0,T fault ] covers the three moments before the fault, the fault occurrence and the fault removal; the transient response time window is composed of the transient characteristics F of the sample m , X m As steady-state characteristics and transient characteristics F m Integrate the input data set into the stability margin T m As the output data set; wherein each node includes each load node and each generator node;

[0111] Construct a convolutional neural network (CNN) model, train the CNN model through input data sets and output data sets, and optimize the parameters of the CNN model to obtain a trained CNN model.

[0112] In order to further implement the above technical solution, a computer-readable storage medium stores a computer program, characterized in that when the program is executed by a processor, the steps of any of the above methods are implemented.

[0113] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for evaluating the power oscillation suppression effect of a sending-end power grid based on a convolutional neural network, characterized in that: The following steps are involved: Input the data to be evaluated into the trained convolutional neural network (CNN) model to obtain the evaluation results. The specific contents of building and training the convolutional neural network (CNN) model include: According to the various operating modes of the sending power grid R j , transient fault F k And different control modes of sending-end flexible DC converter stations C n , generate a stable calculation sample set X for stability analysis and power oscillation suppression effect evaluation m ; where j = 1, 2, ..., M; k = 1, 2, ..., P; n = 1, 2, ..., Q; X m ={R j ,F k ,C n }, m=1,2,...,M*P*Q; Solve for X using the bisection method m The fault limit removal time CCT value of each sample is compared with the actual fault removal time T ac Compare and get the system stability margin T m ; The time series data of voltage U, active power P, reactive power Q or frequency f of each node are sampled to form a transient response time window. The sampling interval [T0,T fault ] covers the three moments before fault, fault occurrence and fault removal; the transient response time window is composed of the transient characteristics F of the sample m , X m As steady-state characteristics and transient characteristics F m Integrate into the input data set and transform the stability margin T m As the output data set; wherein each node includes each load node and each generator node; Construct a convolutional neural network (CNN) model, train the CNN model using input and output data sets, and optimize the parameters of the CNN model to obtain a trained CNN model.

2. The method for evaluating the power oscillation suppression effect of a sending-end power grid based on a convolutional neural network according to claim 1, characterized in that: According to the various operating modes of the sending power grid R j , transient fault F k And different control modes of sending-end flexible DC converter stations C n , generate a stable calculation sample set X for stability analysis and power oscillation suppression effect evaluation m The specific contents include: Operation Mode R j Including setting different load levels, different generator start and stop states or different grid topologies, j = 1, 2, ..., M; Transient fault F k Including three-phase short circuit fault, single-phase short circuit fault, DC single-pole blocking or DC double-pole blocking, k = 1, 2, ..., P; Different control modes of sending-end flexible DC converter stations C n Including active power and reactive power control methods, n = 1, 2, ..., Q; On the simulation platform, batch builds stable calculation sample sets X, X including different operation modes, transient faults, and control modes of the sending-end flexible DC converter station. m ={R j ,F k ,C n }, m=1,2,...,M*P*Q.

3. The method for evaluating the power oscillation suppression effect of a sending-end power grid based on a convolutional neural network according to claim 1, characterized in that: Solve for X using the bisection method m The fault limit removal time CCT value of each sample is compared with the actual fault removal time T ac Compare and get the system stability margin T m The specific contents include: Set the solution range of batch calculation CCT value for all samples [T min ,T max ]; The fault removal time range is gradually narrowed by iterative bisection: in each iterative calculation, the current intermediate value T is obtained. i =(T min +T max ) / 2, for each sample at the fault removal time T i Perform time domain simulation under the following conditions. After each simulation, read the calculation result file to obtain the maximum power angle difference Δδ between any two generators within the simulation time interval. max And judge the stability of the system by the power angle stability criterion; If the system is in T i If the current T i Update T min Otherwise, the current T i Update T max ; Until the updated T min and T max When the difference between the two is less than the preset threshold ε, the current T i For X m Fault limit clearing time CCT i ; CCT i The actual fault clearing time T ac Compare, record the system's stable state, and calculate the system's stability margin T m =CCT i -T ac , T m As X m A quantitative evaluation index of the system power oscillation suppression effect.

4. The method for evaluating the effect of power oscillation suppression of a sending-end power grid based on a convolutional neural network according to claim 3, characterized in that: The content of judging the stability of the system through the power angle stability criterion specifically includes: The power angle stability criterion is: When δ TSI When <0, it means that under the current disturbance, the system transient power angle is unstable, otherwise the system remains stable; The stability state of the system is stored in the state information S in the form of a 0-1 variable, where S=1 means that the system is in the current T i When S=0, it means the system is stable at the current T i Unstable down.

5. The method for evaluating the effect of power oscillation suppression of a sending-end power grid based on a convolutional neural network according to claim 1, characterized in that: In the X m As steady-state characteristics and transient characteristics F m Before integrating into the input data set, the steady-state feature X m and transient characteristics F m Normalize the feature data to ensure that the data are compared on the same scale; Both the input and output datasets are randomly divided into training and test sets.

6. The method for evaluating the effect of power oscillation suppression of a sending-end power grid based on a convolutional neural network according to claim 1, characterized in that: The specific contents of building a convolutional neural network model include: The input layer is used to receive the input data set and pass the data to the convolutional layer; The convolution layer includes at least one single convolution layer. Each single convolution layer extracts different levels of data features layer by layer. A batch normalization layer and a ReLU activation layer are set after each single convolution layer. The batch normalization layer normalizes the final output of the convolution layer to reduce the difference between features of different batches; the ReLU activation layer performs nonlinear transformation on the features output by the batch normalization layer; The pooling layer downsamples the feature data after nonlinear transformation; The Dropout layer randomly discards some neurons during training to ensure model robustness; The feature data after downsampling in the pooling layer is flattened into a one-dimensional vector through the fully connected layer, and the predicted value is output through the regression layer to achieve the regression task.

7. The method for evaluating the effect of power oscillation suppression in a sending-end power grid based on a convolutional neural network according to claim 1, characterized in that: After building the CNN model, set the network architecture: set the initial learning rate, batch size, and maximum number of training times of the CNN model, set the training set to be shuffled in each training cycle, and draw the training process curve.

8. The method for evaluating the effect of power oscillation suppression in a sending-end power grid based on a convolutional neural network according to claim 1, characterized in that: Use the test set to predict the trained CNN model and calculate the root mean square error RMSE and determination coefficient R of the CNN model on the test set 2 ,in: Among them, y (i) is the true value of the sample, is the sample prediction value, is the sample mean, m is the number of samples; RMSE and R 2 To evaluate the generalization ability of the model, the hyperparameters of the model are further adjusted according to the evaluation results to obtain the final CNN model.

9. A system for evaluating the effect of power oscillation suppression on a sending-end power grid based on a convolutional neural network, based on a method for evaluating the effect of power oscillation suppression on a sending-end power grid based on a convolutional neural network according to any one of claims 1 to 8, characterized in that: include: Input module and model building and training module; The input module is used to input the data to be evaluated into the trained convolutional neural network (CNN) model to obtain the evaluation results; The model building and training module is used to build and train the convolutional neural network (CNN) model, specifically including: According to the various operating modes of the sending power grid R j , transient fault F k And different control modes of sending-end flexible DC converter stations C n , generate a stable calculation sample set X for stability analysis and power oscillation suppression effect evaluation m ; where j = 1, 2, ..., M; k = 1, 2, ..., P; n = 1, 2, ..., Q; X m ={R j ,F k ,C n }, m=1,2,...,M*P*Q; Solve for X using the bisection method m The fault limit removal time CCT value of each sample is compared with the actual fault removal time T ac Compare and get the system stability margin T m ; The time series data of voltage U, active power P, reactive power Q or frequency f of each node are sampled to form a transient response time window. The sampling interval [T0,T fault ] covers the three moments before fault, fault occurrence and fault removal; the transient response time window is composed of the transient characteristics F of the sample m , X m As steady-state characteristics and transient characteristics F m Integrate into the input data set and transform the stability margin T m As the output data set; wherein each node includes each load node and each generator node; Construct a convolutional neural network (CNN) model, train the CNN model using input and output data sets, and optimize the parameters of the CNN model to obtain a trained CNN model.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Citation Information

Cited By

  • Oscillation identification method and device, equipment, storage medium and product

    CN121388762A