Power grid transient stability margin prediction method based on steady-state power flow and anticipated fault

Abstract

The invention discloses a power grid transient stability margin prediction method based on steady-state power flow and anticipated faults, and the method comprises the steps: carrying out the sample generation based on different operation conditions of a power system, an AC power flow, operation modes under different time sequences, and faults under N-1 and N-2 scenes; labeling the generated sample according to two transient stability margin evaluation indexes of power angle stability and voltage stability; the labeled sample data and the anticipated fault data after the embedded coding are combined to serve as input features to be input into a deep learning model and trained, and a trained deep learning model is obtained; and predicting the transient stability margin of the power grid according to the trained deep learning model, and judging the stable state of the power system. The transient stability evaluation sample generation method is provided, the steady-state power flow and the expected fault before the fault are considered, the transient stability margin prediction of the power grid is realized by using the deep learning model, and the transient stability margin prediction speed, precision and reliability are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power grid transient stability evaluation, and relates to a power grid transient stability margin prediction method based on steady-state power flow and expected faults. BACKGROUND

[0002] With the expansion of the scale of new power systems, the rapid growth of the proportion of wind and solar energy access, the influence of large-capacity DC expansion zone transmission and new loads, there is a large gap between the traditional typical mode in single-period power balance analysis and extensive power balance analysis, which has gradually failed to meet the requirements of power grid operation under the influence of massive uncertain factors, and the traditional "high in winter and high in summer" characteristics have gradually changed, the typical day characteristics are no longer obvious, and the typical operation mode has been difficult to cover all possible operation boundaries, which cannot prepare for simulating the operation boundary of the power grid, and the guiding significance for operation is limited. Therefore, in the current power grid simulation analysis, the model needs to have more intelligent data analysis capability and prediction capability.

[0003] At present, the steady-state data identification restoration and transient stability evaluation analysis of power system time sequence production simulation results still long-term rely on a large amount of manual experience, which is tedious and complex. And at present, the N-1 check in the time sequence production simulation still relies on the traditional power system simulation software such as BPA, PSASP, etc., which needs to import the results of the time sequence production simulation into these software for calculation, which is low in efficiency and is not conducive to the implementation of large-scale simulation, and in the application of power grid transient stability prediction, it is committed to rapid judgment after the fault occurs, and rarely pays attention to the transient stability margin prediction based on pre-fault information. Therefore, the present application is proposed. SUMMARY

[0004] The purpose of the present application is to provide a power grid transient stability margin prediction method based on steady-state power flow and expected faults, to propose a transient stability evaluation sample generation method, to consider the steady-state power flow before the fault and the expected fault, and to use a deep learning model to realize the prediction of the transient stability margin of the power grid, thereby improving the speed, accuracy and reliability of the transient stability margin prediction.

[0005] The technical solution for achieving the purpose of the present application is as follows: A power grid transient stability margin prediction method based on steady-state power flow and expected faults, comprising the following steps: Based on different operation conditions of the power system, the alternating current flow is used, and the samples are generated based on the operation mode under different time sequences and the faults under N-1 and N-2 scenarios; The generated samples are labeled according to the two transient stability margin evaluation indexes of power angle stability and voltage stability; The labeled sample data and embedded expected fault data are input into a deep learning model as input features, and the deep learning model is trained to obtain a trained deep learning model. The transient stability margin of the power grid is predicted according to the trained deep learning model, and the stability state of the power system is determined.

[0006] In the preferred technical solution, the sample generation comprises: Under the typical scenario operation mode of the target power system, a plurality of power flow change scenarios and fault change scenarios are uniformly sampled; for each power flow change scenario, the mean and variance are set to randomly sample the unit output value and load level value within the proportional range based on the typical mode, and the specific objects of line maintenance and unit maintenance are obtained through non-replacement random sampling; for each fault change scenario, the fault removal time is determined by fixed value, the fault line and the corresponding fault type are determined by random sampling, and the sample values are combined to form system operation information; The operation mode under the time sequence simulation scenario generates continuous operation states including 8760 hours in a year based on different time sequence simulations, the operation states include fault conditions under N-1 and N-2 scenarios, according to the importance and probability density of the operation states, a non-uniform sampling strategy is adopted to focus on sampling high-risk areas; the fault change scenario sampling method is consistent with that under the typical scenario operation mode.

[0007] In the preferred technical solution, the labeling according to the two transient stability margin evaluation indexes of power angle stability and voltage stability comprises: After generating the samples, the collected samples are labeled, and the transient stability index TSI and TVSI are used to determine whether the system is stable; wherein,

[0008] In the formula, is the maximum relative power angle difference of any generator in the transient process; When , it indicates that the system is in transient stability; when , it indicates that the system is in transient instability;

[0009] In the formula, is the total number of buses, is the transient stability simulation time, is the fault removal time, is the transient voltage deviation index; When , it indicates that the system has good voltage stability after disturbance; when , it indicates that the system has voltage instability under disturbance; On the label, the data is marked as 1, then the system is stable, and the data is marked as 0, then the system is unstable; In the preferred technical solution, the transient voltage deviation index is:

[0010] wherein, is the initial state voltage value, is the voltage value at the end of the transient simulation, is the safety threshold of the voltage amplitude deviation.

[0011] In the preferred technical solution, the embedded encoded expected fault data includes: By embedding encoding, the steady-state power flow data and multi-dimensional attributes under the expected fault situation are mapped to a unified vector space, and the discrete and unstructured fault information is converted into low-dimensional and dense vector representation.

[0012] In the preferred technical solution, the deep learning model is a convolutional neural network, which includes multiple convolutional layers and pooling layers, and is used to extract feature information of the power grid under different fault conditions layer by layer, the convolutional layer performs convolution operation on the input data to generate output feature map :

[0013] wherein, the input matrix , is a real number space, n represents the data dimension, m is the time step, and c is the channel number, represents the convolution kernel weight of the l layer, represents the bias term, f ( ) is a nonlinear activation function; After several layers of convolution and pooling operations, the final feature vector is transmitted to the fully connected layer for regression prediction, and the output transient stability margin value is output, and the output of the fully connected layer is the predicted transient stability margin :

[0014] wherein, W f and b f are the weights and biases of the fully connected layer, respectively, and σ( ) is an activation function.

[0015] In the preferred technical solution, it further includes: According to the prediction result of the transient stability margin of the power grid by the trained deep learning model, the transient stability margin evaluation models of the power angle stability and the voltage stability are embedded in the operation optimization of the unit commitment in a constrained form, and the operation mode generated by the time sequence simulation is ensured to have transient stability through re-solving of the unit commitment.

[0016] The application further discloses a power grid transient stability margin prediction system based on steady-state power flow and expected fault, comprising: A sample generation module generates samples based on different operation conditions of the power system, uses alternating current power flow, and is based on operation modes under different time sequences and faults under N-1 and N-2 scenarios. An evaluation labeling module labels the generated samples according to two transient stability margin evaluation indexes of power angle stability and voltage stability. A deep learning model training module inputs the labeled sample data and embedded encoded expected fault data as input features into a deep learning model and performs training to obtain a trained deep learning model. A prediction module predicts the transient stability margin of the power grid according to the trained deep learning model and judges the stability state of the power system.

[0017] The application further discloses a computer storage medium having a computer program stored thereon, and the computer executes the computer program to implement the power grid transient stability margin prediction method based on steady-state power flow and expected fault.

[0018] Compared with the prior art, the application has the following advantages: 1. A transient stability evaluation sample generation method is proposed, a construction method of a time sequence stability evaluation sample set under typical scenarios and time sequence scenarios is proposed, and a set of reasonable and practical artificial intelligence time sequence stability evaluation sample set is generated. Considering the steady-state power flow before the fault and the expected fault, the deep learning model is used to predict the transient stability margin of the power grid, and the speed, accuracy and reliability of the transient stability margin prediction are improved.

[0019] 2. Based on the stability judgment result of artificial intelligence, a key technology of transforming an artificial intelligence neural network into a linear programming problem is proposed, the transient stability evaluation model of the neural network is embedded in the operation optimization of the unit commitment in a constrained form, and the transient stability of the operation mode generated by the time sequence simulation is ensured through re-solving of the unit commitment. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A flowchart of the power grid transient stability margin prediction method based on steady-state power flow and expected fault is shown in the figure. Figure 2 A flowchart of the power system transient stability evaluation based on machine learning is shown in the figure. Figure 3 a transient power angle stability profile for the sample; Figure 4 a confusion matrix result plot; Figure 5 a receiver operating characteristic (ROC) plot. DETAILED DESCRIPTION

[0021] The principle of the present application is that the present application aims to solve the problem that the operating mode generated by the current time sequence production simulation cannot quickly and accurately evaluate the transient stability of the system. A data-driven time sequence simulation transient stability analysis and optimization method system is constructed, and two innovative methods are proposed: ①In view of the massive transient stability calculation and analysis requirements brought by the complex and changeable time sequence simulation operating mode and transient stability fault set of large power grids, an artificial intelligence-based fast transient stability evaluation method under the massive operating mode of large power grids is innovatively proposed, which realizes the transient stability evaluation of N-1 and N-2 faults in 8760 hours without transient stability calculation in seconds, and solves the bottleneck that the time sequence simulation results cannot be quickly analyzed. ②Based on the stability judgment result of artificial intelligence, a key technology of transforming artificial intelligence neural network into linear programming problem is proposed, which embeds the transient stability evaluation model of neural network in the form of constraint in the operational optimization of unit commitment, and through the re-solution of unit commitment, it ensures that the operating mode generated by the time sequence simulation has transient stability.

[0022] Embodiments As shown in Figure 1 A power grid transient stability margin prediction method based on steady-state power flow and expected fault, comprising the following steps: Based on different operating conditions of the power system, the alternating current flow is utilized, and based on the operating mode under different time sequences and the fault under N-1 and N-2 scenarios, samples are generated; The generated samples are labeled according to the power angle stability and voltage stability two transient stability margin evaluation indexes; The labeled sample data is combined with the embedded and coded expected fault data as input features to input a deep learning model and trained to obtain a trained deep learning model; According to the trained deep learning model, the transient stability margin of the power grid is predicted, and the stability state of the power system is judged.

[0023] In a preferred embodiment, sample generation includes: In the typical scenario operation mode of the target power system, a plurality of power flow change scenarios and fault change scenarios are uniformly sampled; for each power flow change scenario, the mean and variance are set to randomly sample the unit output value and load level value within the proportional range based on the typical mode, and the specific objects of line maintenance and unit maintenance are obtained through non-replacement random sampling; for each fault change scenario, the fault removal time is fixed, the fault line and the corresponding fault type are determined through random sampling, and the sampling values are combined to form system operation information; In the operation mode under the time sequence simulation scenario, continuous operation states including 8760 hours in a year are generated based on different time sequence simulations, the operation states include fault conditions under N-1 and N-2 scenarios, according to the importance and probability density of the operation states, a non-uniform sampling strategy is adopted, and the high-risk area is focused on sampling; the fault change scenario sampling mode is consistent with that in the typical scenario operation mode.

[0024] In a preferred embodiment, the two transient stability margin evaluation indexes of power angle stability and voltage stability are used for labeling, including: After generating the samples, the collected samples are labeled, and the transient stability indexes TSI and TVSI are used to judge whether the system is stable; wherein,

[0025] In the formula, is the maximum relative power angle difference of any generator in the transient process; When , it indicates that the system is in transient stability; when , it indicates that the system is in transient instability;

[0026] In the formula, is the total number of buses, is the transient stability simulation time, is the fault removal time, is the transient voltage deviation index; When , it indicates that the system has good voltage stability after the disturbance; when , it indicates that the system loses voltage stability under the disturbance; In the label, the data is labeled as 1 if the system is stable, and the data is labeled as 0 if the system is unstable; In a preferred embodiment, the transient voltage deviation index is:

[0027] Wherein, is the initial state voltage value, is the voltage value at the end of the transient stability simulation, is a safety threshold of voltage amplitude deviation.

[0028] In a preferred embodiment, the embedded encoded expected fault data includes: By embedding encoding, the steady-state flow data and the multi-dimensional attributes under the expected fault condition are mapped to a unified vector space, and the discrete and unstructured fault information is converted into a low-dimensional and dense vector representation.

[0029] In a preferred embodiment, the deep learning model is a convolutional neural network, which includes multiple convolutional layers and pooling layers for layer-by-layer extraction of feature information of the power grid under different fault conditions, the convolutional layer performs convolution operation on the input data to generate an output feature map :

[0030] wherein the input matrix , is a real number space, n represents the data dimension, m is the time step, c is the number of channels, represents the convolution kernel weight of the l layer, represents the bias term, f ( ) is a nonlinear activation function; After several layers of convolution and pooling operations, the final feature vector is passed to the fully connected layer for regression prediction, and the output transient stability margin value is output, and the output of the fully connected layer is the predicted transient stability margin :

[0031] wherein, W f and b f are the weights and biases of the fully connected layer, respectively, and σ( ) is an activation function.

[0032] In a preferred embodiment, it further comprises: According to the prediction result of the transient stability margin of the power grid by the trained deep learning model, the power angle stability and voltage stability transient stability margin evaluation model is embedded in the operational optimization of unit commitment in a constrained form, and the transient stability of the operation mode generated by the time sequence simulation is realized by re-solving the unit commitment.

[0033] By directly constructing the correlation between the time sequence simulation results and the N-1 results, the N-1 test of the results generated by the time sequence production simulation is realized, and the efficiency of the time sequence production simulation is greatly improved.

[0034] In another embodiment, a computer storage medium having stored thereon a computer program, the computer program, when executed by a computer, implements the power grid transient stability margin prediction method based on steady-state power flow and contingency according to any one of the preceding embodiments.

[0035] In another embodiment, a power grid transient stability margin prediction system based on steady-state power flow and contingency comprises: a sample generation module, which generates samples based on different operating conditions of the power system, uses alternating current power flow, and is based on different time sequences of operating modes and N-1 and N-2 scenarios of faults; an evaluation labeling module, which labels the generated samples according to two transient stability margin evaluation indexes of power angle stability and voltage stability; a deep learning model training module, which inputs the labeled sample data and embedded and encoded contingency data as input features into a deep learning model and trains the deep learning model to obtain a trained deep learning model; a prediction module, which predicts the transient stability margin of the power grid according to the trained deep learning model and judges the stability state of the power system.

[0036] The following describes the working process of the power grid transient stability margin prediction system based on steady-state power flow and contingency with a specific example: As shown in Figure 2 The machine learning-based power system transient stability evaluation is composed of offline training and online evaluation. After the fault disappears, the trained model is used to evaluate the transient stability of the power system. If the result shows that the system is still in a transient stable state, the model will continue to monitor the power system in real time; otherwise, if the result shows that the system has lost stability, emergency control measures should be taken immediately to ensure the safety of the power system.

[0037] The quality of the training sample set directly determines whether the artificial intelligence model can achieve excellent evaluation results. In the time sequence simulation scene, the most complex is that it involves a large number of different operation modes, and the complexity of the fault set also needs to be considered. Under the superposition of these factors, a large number of transient stability samples are required, and too many samples will increase the cost of sample acquisition and storage. Therefore, the method proposes a special transient stability evaluation sample generation technology. Specifically, two different sample generation strategies are proposed, which correspond to ① commonly used typical operation modes. Because the transient stability characteristics in the typical operation mode are more close to the actual operation of the power grid, this sample set is used to strengthen the model's learning of the transient stability characteristics under the typical mode, and at the same time enables the model to cope with the transient stability analysis function under the dispatching scene. ② Operation mode under time sequence simulation scene, based on the operation mode of 8760 sections generated by different time sequence simulations, while considering different N-1, N-2 fault conditions. This sample generation strategy is guided by the target scene to drive the model to better predict the transient stability characteristics of the time sequence simulation results.

[0038] Specifically, under the typical scene operation mode of the target power system, a plurality of power flow change scenes and fault change scenes are uniformly sampled; for each power flow change scene, the mean and variance are randomly sampled within the proportional range based on the typical mode to set the unit output value and load level value, and the specific objects of line maintenance and unit maintenance are obtained by non-replacement random sampling; for each fault change scene, the fault removal time is fixedly valued, and the fault line and the corresponding fault type are determined by random sampling, so as to form the system operation information by combining the sampling values. The time sequence simulation scene operation mode includes 8760 hours of continuous operation state in a year. According to the importance and probability density of the operation state, a non-uniform sampling strategy is adopted to focus on sampling high-risk areas; the fault change scene sampling method is consistent with that in the typical scene operation mode, so as to greatly reduce the required number of samples while maintaining the representativeness and coverage of the sample set.

[0039] Using the power flow calculation module in the new power system planning simulation platform (TEAP), the operation results under the typical scene or time sequence simulation scene can be used as time sequence input for AC power flow calculation to generate transient stability sample sets that are more close to the actual power flow. Based on the above information, the sample generation scheme shown in Table 1 is proposed: Table 1 Sample generation scheme under typical operation mode

[0040] Propose an integrated evaluation model for power angle stability and voltage stability

[0041] In order to better evaluate the transient stability of the system, the model should consider the problem of angle stability and voltage stability at the same time, and build appropriate input features. For this purpose, a multi-source heterogeneous deep learning model is built. Its main innovations are: ① Based on different types of data, the features are fused after being processed by different machine learning methods. ② It can give the transient stability class and stability margin of the system at the same time. ③ Based on the feature sharing of angle stability and voltage stability integration evaluation technology.

[0042] After generating the samples, the collected samples need to be labeled using supervised machine learning techniques. Here, the transient stability index TSI and TVSI are used to determine whether the system is stable.

[0043]

[0044] In the formula, is the maximum relative public angle difference of any generator in the transient process. When , it indicates that the system is in transient stability; when , it indicates that the system is in transient instability. In order to better distinguish between stable and unstable, the data is labeled as 1 if the system is stable, and 0 if the system is not stable.

[0045]

[0046] In the formula, is the total number of buses, is the transient stability simulation time, is the fault clearance time. When , it indicates that the system has good voltage stability after disturbance; when , it indicates that the system will experience voltage instability under disturbance. The transient voltage deviation index is:

[0047] Among them, is the initial state voltage value, is the voltage value at the end of transient stability simulation, is the safety threshold of voltage amplitude deviation, generally 10%.

[0048] Based on the TEAP new power system planning simulation platform AC power flow module, 18000 samples are generated based on 8760 operation modes under a certain time sequence calculation, considering N-1 and N-2 fault scenarios.

[0049] The transient angle stability distribution of its samples is as follows: Figure 3 : TSI in the vicinity of the sample indicates angle instability (accounting for 11.12%)

[0050] The deep learning model used in the present application is a convolutional neural network (CNN). The CNN is particularly suitable for modeling and predicting large-scale, high-dimensional data due to its efficient extraction of local features and parameter sharing mechanism. In power grid transient stability analysis, the steady-state flow data and feature information under the expected fault scenario are highly complex and have spatial correlation. By inputting these data into the CNN model, the spatial features and patterns in the data can be effectively captured, thereby achieving accurate prediction of the transient stability margin. Through embedding technology, the multi-dimensional attributes of the fault can be mapped to a unified vector space, thereby converting the discrete, unstructured fault information into low-dimensional, dense vector representation, which can be used as input features together with the labeled sample data to input into the CNN model for prediction.

[0051] The CNN model includes multiple convolutional layers and pooling layers for layer-by-layer extraction of feature information of the power grid under different fault conditions. Assuming that the input matrix is where n represents the data dimension, m is the time step, and c is the number of channels. The convolutional layer performs convolution operation on the input data to generate the output feature map :

[0052] where represents the convolution kernel weight of the l-th layer, represents the bias term, and f( ) is a nonlinear activation function such as the ReLU function.

[0053] After several layers of convolution and pooling operations, the final feature vector F is passed to the fully connected layer for regression prediction, and the output transient stability margin value is output. The output of the fully connected layer is the predicted transient stability margin :

[0054] where W f and b f are the weight and bias of the fully connected layer, respectively, and σ( ) is the activation function.

[0055] The following Table 2 shows the prediction results of the CNN model: Table 2

[0056] Figure 4 , 5 The following Table 2 shows the prediction results of the CNN model:

[0057] A method for constructing a time sequence stability evaluation sample set under typical scenarios and time sequence scenarios is proposed, and a set of reasonable and practical artificial intelligence time sequence stability evaluation sample set is generated. Considering the steady-state power flow before the fault and the expected fault, the transient stability margin of the power grid is predicted by using a deep learning model, thereby improving the speed, accuracy and reliability of the transient stability margin prediction.

[0058] The above embodiments are preferred embodiments of the present application, but the embodiments of the present application are not limited by the above embodiments, and any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present application shall be equivalent replacement modes and shall be included in the protection scope of the present application.

Claims

1. A method for predicting the transient stability margin of a power grid based on steady-state power flow and anticipated faults, characterized in that, Includes the following steps: Based on different operating conditions of the power system, AC power flow is used to generate samples based on different time-series operating modes and faults in two scenarios, N-1 and N-2. The generated samples are labeled according to two transient stability margin evaluation indices: power angle stability and voltage stability. The labeled sample data, combined with the embedded and encoded expected fault data, are used as input features to input the deep learning model and train it to obtain a trained deep learning model. The transient stability margin of the power grid is predicted based on the trained deep learning model, and the stable state of the power system is determined.

2. The power grid transient stability margin prediction method based on steady-state power flow and anticipated faults according to claim 1, characterized in that, Sample generation includes: Under the typical operating mode of the target power system, multiple power flow change scenarios and fault change scenarios are uniformly sampled. For each power flow change scenario, within the proportional range based on the typical mode, the unit output value and load level value are randomly sampled by setting the mean and variance. The specific objects of line maintenance and unit maintenance are obtained through random sampling without replacement. For each fault change scenario, the fault clearing time is determined by fixed values. The faulted line and the corresponding fault type are determined by random sampling. The sampled values ​​are combined to form the system operation information. The operation mode under the time-series simulation scenario is based on the continuous operation status of 8760 hours throughout the year generated by different time-series simulations. The operation status includes fault conditions under two scenarios, N-1 and N-2. According to the importance and probability density of the operation status, a non-uniform sampling strategy is adopted, focusing on sampling high-risk areas. The sampling method for fault change scenarios is consistent with that under the typical scenario operation mode.

3. The method for predicting power grid transient stability margin based on steady-state power flow and anticipated faults according to claim 1, characterized in that, The annotation is based on two transient stability margin evaluation indicators: power angle stability and voltage stability. After generating samples, the collected samples are labeled, and the transient stability indices TSI and TVSI are used to determine whether the system is stable; among them, In the formula, This represents the maximum relative common angle difference of any generator during the transient process; when When, it indicates that the system is in a transient stable state; when When this occurs, it indicates that the system is in a transient state of instability; In the formula, The total number of busbars, To stabilize the simulation time, For fault clearing time, This refers to the transient voltage deviation index; when When the voltage is stable, it indicates that the system has good voltage stability after experiencing a disturbance; when When this occurs, it indicates that the system has experienced voltage instability under disturbance; On the label, data labeled as 1 indicates that the system is stable, while data labeled as 0 indicates that the system is unstable.

4. The power grid transient stability margin prediction method based on steady-state power flow and anticipated faults according to claim 3, characterized in that, The transient voltage deviation index for: in, This is the initial state voltage value. This is the voltage value at the end of the quasi-stable simulation. This is the safe threshold for voltage amplitude deviation.

5. The power grid transient stability margin prediction method based on steady-state power flow and anticipated faults according to claim 1, characterized in that, The embedded and encoded anticipated fault data includes: By embedding encoding, steady-state power flow data and multi-dimensional attributes under anticipated fault scenarios are mapped to a unified vector space, transforming discretized and unstructured fault information into a low-dimensional, dense vector representation.

6. The method for predicting the transient stability margin of a power grid based on steady-state power flow and anticipated faults according to claim 1, characterized in that, The deep learning model is a convolutional neural network, including multiple convolutional layers and pooling layers, used to extract feature information of the power grid under different fault conditions layer by layer. The convolutional layers perform convolution operations on the input data to generate output feature maps. : Wherein, the input matrix , Let n be the real number space, m be the time step, and c be the number of channels. Indicates the first l The kernel weights of the layer Indicates the bias term. f ( ) is a non-linear activation function; After several layers of convolution and pooling operations, the final feature vector The data is passed to the fully connected layer for regression prediction, and the output is the transient stability margin value. The output of the fully connected layer is the predicted transient stability margin. : in, W f and b f These are the weights and biases of the fully connected layer, σ( ) is the activation function.

7. The method for predicting power grid transient stability margin based on steady-state power flow and anticipated faults according to claim 1, characterized in that, Also includes: Based on the prediction results of the transient stability margin of the power grid by the trained deep learning model, the evaluation models of power angle stability and voltage stability transient stability margin are embedded in the operation optimization of the unit combination in the form of constraints. By resolving the unit combination, the operating mode generated by the time-series simulation has transient stability.

8. A power grid transient stability margin prediction system based on steady-state power flow and anticipated faults, characterized in that, include: The sample generation module generates samples based on different operating conditions of the power system, utilizing AC power flow, and based on different time-series operating modes and faults in two scenarios, N-1 and N-2. The evaluation and labeling module labels the generated samples based on two transient stability margin evaluation indicators: power angle stability and voltage stability. The deep learning model training module takes the labeled sample data and the embedded and encoded expected fault data as input features to the deep learning model and trains it to obtain a trained deep learning model. The prediction module predicts the transient stability margin of the power grid based on the trained deep learning model, and determines the stable state of the power system.

9. The power grid transient stability margin prediction system based on steady-state power flow and anticipated faults according to claim 8, characterized in that, Sample generation includes: Under the typical operating mode of the target power system, multiple power flow change scenarios and fault change scenarios are uniformly sampled. For each power flow change scenario, within the proportional range based on the typical mode, the unit output value and load level value are randomly sampled by setting the mean and variance. The specific objects of line maintenance and unit maintenance are obtained through random sampling without replacement. For each fault change scenario, the fault clearing time is determined by fixed values. The faulted line and the corresponding fault type are determined by random sampling. The sampled values ​​are combined to form the system operation information. The operation mode under the time-series simulation scenario is based on the continuous operation status of 8760 hours throughout the year generated by different time-series simulations. The operation status includes fault conditions under two scenarios, N-1 and N-2. According to the importance and probability density of the operation status, a non-uniform sampling strategy is adopted, focusing on sampling high-risk areas. The sampling method for fault change scenarios is consistent with that under the typical scenario operation mode.

10. A computer storage medium having a computer program stored thereon, characterized in that, When the computer executes the computer program, it implements the power grid transient stability margin prediction method based on steady-state power flow and anticipated faults as described in any one of claims 1-7.

Images