A method, apparatus, and equipment for predicting the amplitude of transient overvoltages in power grids at the sending end.

Abstract

This invention discloses a method, apparatus, and equipment for predicting transient overvoltage amplitude in power grids at the sending end, relating to the field of dynamic security analysis technology for power systems. The method includes: generating a transient overvoltage sample set through time-domain simulation; training a transient overvoltage amplitude prediction model and a risk classification model based on the transient overvoltage sample set; inputting real-time key response features into the transient overvoltage amplitude prediction model to obtain the corresponding predicted transient overvoltage amplitude; and inputting the real-time key response features and the corresponding predicted transient overvoltage amplitude into the risk classification model to obtain the corresponding risk level; if the risk is low, the predicted transient overvoltage amplitude is directly adopted; if the risk is high, time-domain simulation is triggered to verify the predicted transient overvoltage amplitude, and the simulation result is used as the final predicted transient overvoltage amplitude. This method resolves the inherent contradiction between the low computational efficiency of time-domain simulation and the poor reliability of data-driven models in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of dynamic security analysis technology for power systems, specifically to a method, apparatus, and equipment for predicting the amplitude of transient overvoltages in the sending-end power grid. Background Technology

[0002] Currently, the installed capacity of new energy sources is growing rapidly, with wind power, photovoltaic, and other power electronic equipment being connected to the sending-end grid on a large scale and transmitted to the receiving-end grid through ultra-high-voltage direct current (UHVDC) transmission systems. In UHVDC transmission systems, when commutation failures or DC blocking occur, short-term reactive power excess may trigger transient overvoltages, causing new energy sources to experience transient overvoltages exceeding 1.3 pu. Large-scale grid disconnection severely impacts the safe and stable operation of the sending-end grid. Therefore, accurately predicting the amplitude of sending-end transient overvoltages is crucial for assessing the operational risks of the grid during disturbances, developing reasonable control measures, and ensuring the safe and stable operation of the system.

[0003] Currently, transient overvoltage analysis methods for sending-end power grids are mainly divided into two categories: model-driven and data-driven. Model-driven methods are based on system mechanisms and characterize voltage changes during the transient process after a fault by analyzing the dynamic equations of the power system. These methods typically use time-domain simulation or equivalent calculations to solve for transient overvoltages, accurately obtaining the magnitude of the transient overvoltages. However, they involve large computational loads and complex solution processes, making them difficult to adapt to the needs of online analysis under multiple operating conditions.

[0004] Data-driven methods rely on system operation data and artificial intelligence algorithms to establish transient overvoltage amplitude prediction models based on offline simulation data, which can meet the needs of real-time online analysis. However, data-driven models are highly dependent on training samples. When the distribution of operation data and training data differs significantly, it will affect the reliability of the prediction results, limiting the application of data-driven methods in the prediction of transient overvoltage amplitudes in actual power grids.

[0005] In summary, existing time-domain simulation methods based on rigorous physical models suffer from low computational efficiency and cannot meet the requirements of online applications; while data-driven artificial intelligence prediction methods suffer from insufficient reliability and limited engineering applicability. The inherent contradiction between these two approaches regarding speed and reliability hinders the effective application of transient overvoltage analysis technology in the actual operation of new power systems. Summary of the Invention

[0006] The technical problem to be solved by this invention is how to overcome the contradiction between "low efficiency of time-domain simulation calculation and inability to be applied online" and "poor reliability and insufficient engineering practicality of data-driven prediction models" in existing transient overvoltage analysis methods. The purpose is to provide a method, device and equipment for predicting the amplitude of transient overvoltage in the sending-end power grid, thereby solving the above problems.

[0007] This invention is achieved through the following technical solution:

[0008] In a first aspect, the present invention provides a method for predicting the amplitude of transient overvoltages in a power grid at the sending end, comprising:

[0009] A transient overvoltage sample set is generated through time-domain simulation; each sample in the transient overvoltage sample set includes sample input features and the corresponding true amplitude of transient overvoltage;

[0010] Based on the transient overvoltage sample set, a transient overvoltage amplitude prediction model and a risk classification model are trained and obtained; the transient overvoltage amplitude prediction model is used to predict the transient overvoltage amplitude according to the sample input features; the risk classification model is used to evaluate the risk level of the amplitude prediction result according to the sample input features and their corresponding transient overvoltage amplitudes.

[0011] Extract real-time key response features that are of the same origin as the sample input features from the real-time measurement data of the power grid. Input the real-time key response features into the transient overvoltage amplitude prediction model to obtain the corresponding transient overvoltage prediction amplitude. Input the real-time key response features and the corresponding transient overvoltage prediction amplitude into the risk classification model to obtain the corresponding risk level.

[0012] If the risk level is low, the predicted amplitude of the transient overvoltage will be used as the final predicted amplitude of the transient overvoltage; if the risk level is high, a time-domain simulation will be triggered to verify the predicted amplitude of the transient overvoltage, and the simulation result will be used as the final predicted amplitude of the transient overvoltage.

[0013] Optionally, generating the transient overvoltage sample set through time-domain simulation includes:

[0014] Set up simulation scenarios that include various operating conditions and various fault disturbances, perform simulations based on the time-domain simulation model of the sending-end power grid, and output the original electrical quantity data and the corresponding true amplitude of transient overvoltage before the power grid fault.

[0015] Key electrical quantity features are selected from the raw electrical quantity data;

[0016] Extract the anticipated fault information from the configuration of the simulation scenario;

[0017] The key electrical quantity features and the anticipated fault information are combined into the sample input features, and the corresponding true amplitude of transient overvoltage is used as the label to construct the transient overvoltage sample set.

[0018] Optionally, the step of filtering key electrical quantity features from the original electrical quantity data includes:

[0019] Calculate the correlation coefficient between each feature in the original electrical quantity data and the true amplitude of the transient overvoltage, remove features with correlation coefficients lower than a first preset threshold, and obtain the remaining feature set;

[0020] Calculate the correlation coefficient between each feature in the remaining feature set, remove features with a correlation coefficient higher than a second preset threshold, and obtain the key electrical quantity features.

[0021] Optionally, the step of training a transient overvoltage amplitude prediction model and a risk classification model based on the transient overvoltage sample set includes:

[0022] The transient overvoltage sample set is normalized, and the normalized sample set is divided into a training set, a validation set, and a test set.

[0023] Using the sample input features in the training set as training input and the actual amplitude of the transient overvoltage corresponding to the training set as the training target, the transient overvoltage amplitude prediction model is trained to obtain the model.

[0024] The sample input features in the validation set are input into the transient overvoltage amplitude prediction model to obtain the corresponding transient overvoltage predicted amplitude as the amplitude to be evaluated.

[0025] The risk grading model is trained by using the sample input features and the corresponding amplitude to be evaluated in the validation set.

[0026] Optionally, the step of training the risk grading model using the sample input features and corresponding amplitudes to be evaluated from the validation set as training input includes:

[0027] The prediction error is calculated based on the amplitude to be evaluated and the corresponding true amplitude of transient overvoltage in the verification set;

[0028] Based on the prediction error, calculate the risk value for each sample;

[0029] Based on a preset risk threshold, the risk value of each sample is mapped to a risk level;

[0030] The risk grading model is trained by using the combination of sample input features and corresponding amplitude to be evaluated for each sample in the validation set as training input and the mapped risk level as training objective.

[0031] Optionally, calculating the risk value for each sample based on the prediction error includes:

[0032] The actual transient overvoltage amplitude of each sample is compared with the preset transient overvoltage safety threshold.

[0033] Based on the comparison results, the corresponding risk sensitivity coefficient is selected to weight the prediction error and obtain the risk value of each sample.

[0034] Optionally, the formula for calculating the risk value is as follows:

[0035] when hour, ;

[0036] when hour, ;

[0037] in, The transient overvoltage safety threshold; R represents the true amplitude of the transient overvoltage of the i-th sample; R is the risk value. Let be the prediction error for the i-th sample; , , and The risk sensitivity coefficient is mentioned above; .

[0038] Optionally, after obtaining the final predicted amplitude of the transient overvoltage, the method further includes:

[0039] The final predicted amplitude of transient overvoltage is compared with the preset transient overvoltage safety threshold.

[0040] If the final predicted amplitude of transient overvoltage exceeds the transient overvoltage safety threshold, an early warning message is generated.

[0041] In a second aspect, the present invention provides a device for predicting the amplitude of transient overvoltages in a power grid at the sending end, comprising:

[0042] The sample construction module is used to generate a transient overvoltage sample set through time-domain simulation; each sample in the transient overvoltage sample set includes sample input features and the corresponding true amplitude of transient overvoltage;

[0043] An offline training module is used to train a transient overvoltage amplitude prediction model and a risk grading model based on the transient overvoltage sample set. The transient overvoltage amplitude prediction model is used to predict the transient overvoltage amplitude based on the sample input features. The risk grading model is used to evaluate the risk level of the amplitude prediction result based on the sample input features and their corresponding transient overvoltage amplitudes.

[0044] The online application module is used to extract real-time key response features from real-time power grid measurement data that are of the same origin as the sample input features, input the real-time key response features into the transient overvoltage amplitude prediction model to obtain the corresponding transient overvoltage prediction amplitude, and input the real-time key response features and the corresponding transient overvoltage prediction amplitude into the risk classification model to obtain the corresponding risk level.

[0045] The acquisition module is used to take the predicted transient overvoltage amplitude as the final predicted transient overvoltage amplitude if the risk level is low; and to trigger time-domain simulation to verify the predicted transient overvoltage amplitude if the risk level is high, and to take the simulation result as the final predicted transient overvoltage amplitude.

[0046] Thirdly, the present invention provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement a method for predicting the amplitude of transient overvoltage in a power grid as described in any one of the first aspects.

[0047] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0048] This application provides a method for predicting the amplitude of transient overvoltages in power grids at the sending end. It utilizes an offline-trained transient overvoltage amplitude prediction model to perform millisecond-level forward inference calculations in practical online applications. This replaces the traditional method of performing complete time-domain simulations for each scenario, overcoming the shortcomings of time-domain simulations, which are often too complex to meet real-time online requirements due to complex modeling and excessive computation time. Simultaneously, a prediction risk grading model is introduced, which synchronously outputs a clearly quantified risk level for each prediction result. The model prediction result is directly adopted only when the risk level is low, effectively solving the problems of poor reliability and insufficient engineering applicability of data-driven prediction models. When the risk level is high, time-domain simulation is triggered for verification, further ensuring the accuracy of the predicted transient overvoltage amplitude. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0050] Figure 1 A flowchart illustrating the method for predicting the amplitude of transient overvoltage in the sending-end power grid provided in this application embodiment;

[0051] Figure 2 This is a schematic diagram of the structure of a transient overvoltage prediction model provided in an embodiment of this application;

[0052] Figure 3 This is a schematic diagram of the structure of a risk classification model provided in an embodiment of this application;

[0053] Figure 4 A schematic diagram of the risk function provided in this application embodiment under the condition that the actual amplitude does not exceed the safety threshold;

[0054] Figure 5 A schematic diagram of the risk function provided in this application embodiment when the actual amplitude exceeds the safety threshold;

[0055] Figure 6 A flowchart illustrating the offline training phase provided in an embodiment of this application;

[0056] Figure 7 A flowchart illustrating the online application stage provided in this application embodiment;

[0057] Figure 8 This is a schematic diagram of the structure of the transient overvoltage amplitude prediction device for the sending-end power grid provided in an embodiment of this application. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0059] Please refer to Figure 1 This is a flowchart illustrating the method for predicting the amplitude of transient overvoltage in the sending-end power grid provided in this application embodiment. The following is a further explanation. Figure 1 The method for predicting the amplitude of transient overvoltages in the sending-end power grid is introduced.

[0060] S1. Generate a transient overvoltage sample set through time-domain simulation.

[0061] In the transient overvoltage sample set, each sample contains the following core elements:

[0062] (1) Sample input features: A multi-dimensional feature vector used to accurately characterize the operating state of the sending-end power grid before a specific anticipated fault occurs. This feature integrates key electrical quantity information extracted from the power grid with preset fault scenario description information.

[0063] (2) Actual amplitude of transient overvoltage: A scalar value, serving as the label for the input feature of this sample. It represents the actual peak value of the transient overvoltage calculated through high-fidelity time-domain simulation when the anticipated fault actually occurs under this operating condition. The specific implementation process of S1 will be described below.

[0064] S2. Based on the transient overvoltage sample set, a transient overvoltage amplitude prediction model and a risk classification model are trained and obtained.

[0065] The transient overvoltage amplitude prediction model is used to predict the transient overvoltage amplitude based on the sample input characteristics; the risk grading model is used to assess the risk level of the amplitude prediction result based on the sample input characteristics and their corresponding transient overvoltage amplitudes. The specific implementation process of S2 will be described below.

[0066] S3. Extract real-time key response features from real-time power grid measurement data that are of the same origin as the sample input features. Input the real-time key response features into the transient overvoltage amplitude prediction model to obtain the corresponding transient overvoltage prediction amplitude. Input the real-time key response features and the corresponding transient overvoltage prediction amplitude into the risk classification model to obtain the corresponding risk level.

[0067] In the specific implementation process, real-time key response features are extracted from the real-time measurement data of the power grid. These real-time key response features must be completely consistent with the sample input features constructed in the training phase in terms of type, physical meaning and calculation logic, so as to ensure the consistency between online input data and offline training data in the feature space.

[0068] In actual deployment, the transient overvoltage amplitude prediction model and the risk classification model can be integrated into a fusion network. A set of real-time key response features can be input into the trained transient overvoltage amplitude prediction model to obtain the transient overvoltage prediction amplitude at the current moment.

[0069] Subsequently, the same set of real-time key response features and the predicted transient overvoltage amplitude are input into the risk grading model, which outputs the risk level corresponding to the prediction result. This risk level is used to determine whether the prediction result can be directly adopted and whether further simulation verification is required.

[0070] S4. If the risk level is low, the predicted amplitude of transient overvoltage will be used as the final predicted amplitude of transient overvoltage; if the risk level is high, time-domain simulation will be triggered to verify the predicted amplitude of transient overvoltage, and the simulation result will be used as the final predicted amplitude of transient overvoltage.

[0071] In practical implementation, when the risk level output by the risk grading model is low risk, it indicates that the prediction result is reliable and has little impact on operational decisions, and the predicted amplitude of the transient overvoltage can be directly adopted. When the risk level output by the risk grading model is high risk, it indicates that the uncertainty of the prediction result is high, and direct adoption may result in underreporting (i.e., the actual transient overvoltage amplitude has exceeded the safety threshold but is underestimated by the model prediction). At this time, the system immediately triggers the verification mechanism: it calls the sending-end power grid time-domain simulation model, uses the current state of the power grid represented by the real-time key response features extracted from real-time measurement data as the initial operating conditions of the simulation model, defines the disturbance event parameters of the simulation model with the currently assessed expected fault information, performs a complete time-domain simulation, and uses the peak value of the transient overvoltage calculated by the simulation as the final predicted amplitude of the transient overvoltage.

[0072] In one possible embodiment, after using the simulation results as the final predicted amplitude of the transient overvoltage, the method further includes:

[0073] The scenario data corresponding to the high-risk level is stored as an enhanced sample in the historical database. The enhanced sample includes at least: the real-time key response features used to trigger the simulation review, the transient overvoltage predicted amplitude output by the transient overvoltage amplitude prediction model, the final transient overvoltage predicted amplitude and the corresponding risk level obtained by time-domain simulation calculation. The enhanced samples accumulated in the historical database are periodically used to incrementally learn the transient overvoltage amplitude prediction model and / or risk classification model.

[0074] In the embodiments of this application, each simulation review of high-risk prediction not only resolves the current uncertainty, but also generates an enhanced sample. Regularly using these enhanced samples generated from actual operation to retrain the model can effectively correct cognitive biases of the model, expand its effective decision boundary, and thus continuously improve the model's prediction accuracy and risk assessment reliability when facing complex and novel operating modes.

[0075] In one possible embodiment, after obtaining the final predicted voltage amplitude, the method further includes:

[0076] The final predicted amplitude of transient overvoltage is compared with the preset transient overvoltage safety threshold.

[0077] If the final predicted amplitude of transient overvoltage exceeds the transient overvoltage safety threshold, an early warning message will be generated.

[0078] In this embodiment, the final predicted amplitude of transient overvoltage is compared with the engineering standard threshold in real time. Once the predicted amplitude exceeds the standard, the system automatically generates an early warning, providing operators with a clear and explicit risk warning. This provides valuable decision-making and response time for subsequent emergency control measures such as switching off the machine, switching off the load, and adjusting the operating mode, thus achieving closed-loop active defense against transient overvoltage risks.

[0079] In one possible embodiment, step S1 includes:

[0080] A simulation scenario with multiple operating conditions and fault disturbances is set up. The simulation is performed based on the time-domain simulation model of the sending-end power grid, and the original electrical quantity data before the power grid fault and the corresponding true amplitude of transient overvoltage are output. Key electrical quantity features are screened from the original electrical quantity data. The expected fault information is extracted from the configuration of the simulation scenario. The key electrical quantity features and the expected fault information are combined into sample input features, and the corresponding true amplitude of transient overvoltage is used as the label to construct a transient overvoltage sample set.

[0081] In the specific implementation process, firstly, in order to comprehensively simulate the complex causal relationship between faults and transient overvoltages in actual power grid operation, a simulation scenario was systematically designed that includes multiple operating conditions and multiple fault disturbances. The multiple operating conditions are generated by combining different load levels, renewable energy penetration rates, and DC power. The multiple fault disturbances mainly simulate various typical commutation failure modes. Corresponding anticipated fault information is set for different commutation failure modes, including the number of faults and their duration, to characterize the different severity levels of the disturbances.

[0082] Secondly, a time-domain simulation model of the sending-end power grid is established. This model is a high-fidelity electromagnetic-electromechanical transient hybrid simulation model built based on the actual power grid topology and equipment parameters. Through simulation, this model can realistically reproduce the complex transient dynamic processes of the power grid under fault disturbances.

[0083] Then, for each "operating condition-fault disturbance" scenario in the above design, a complete time-domain simulation is performed using the sending-end power grid time-domain simulation model: the simulation starts from the steady-state operating point before the fault, injects the fault at a preset time, simulates the dynamic response of the system, and records the detailed transient process within hundreds of milliseconds after the fault is cleared. For each scenario, at the moment before the fault occurs (steady state), electrical quantities (such as node voltage, line power flow, etc.) at a wide range of monitoring points in the sending-end power grid are collected as raw electrical quantity data. From the simulation results, the transient peak voltage of the key bus in the sending-end power grid after the fault is extracted as the corresponding true amplitude of the transient overvoltage.

[0084] Finally, key electrical quantity features are selected from the original electrical quantity data, and anticipated fault information (such as fault type, number of faults, and duration) is extracted from the simulation scenario configuration. These key electrical quantity features and anticipated fault information are combined to form a sample input feature. This sample input feature is then paired with the corresponding actual amplitude of the transient overvoltage to form a complete training sample. This process is repeated for all simulation scenarios to ultimately construct a large-scale transient overvoltage sample set.

[0085] In this embodiment, by covering different load levels, renewable energy penetration rates, and DC power combinations, and simulating various commutation failure modes and their severity (different numbers and durations), the generated transient overvoltage sample set comprehensively covers the mainstream and extreme risk scenarios faced by the sending-end power grid under the new power system. This enables the subsequently trained intelligent model to learn a wider range of mapping relationships between operating states and faults, greatly enhancing the model's generalization ability and robustness under the complex and variable operating modes of the actual power grid. This effectively overcomes the shortcomings of traditional data-driven methods, such as over-reliance on specific training samples and inaccurate predictions in out-of-sample scenarios.

[0086] In one possible embodiment, key electrical quantity features are filtered from the raw electrical quantity data, including:

[0087] Calculate the correlation coefficient between each feature in the original electrical quantity data and the true amplitude of the transient overvoltage, remove features whose absolute value of the correlation coefficient is lower than the first preset threshold, and obtain the remaining feature set; calculate the correlation coefficient between each feature in the remaining feature set, remove features whose correlation coefficient is higher than the second preset threshold, and obtain the key electrical quantity features.

[0088] In the specific implementation process, since the original features have high dimensionality and redundancy, Spearman correlation analysis is used to screen the features in order to extract the key features that can effectively reflect the transient overvoltage level.

[0089] The Spearman correlation coefficient is calculated as follows: Given two variables M and N, and a total of L sets of data, represented as (m1, n1), (m2, n2)...(m... L n L Spearman correlation coefficient R s The calculation formula is:

[0090]

[0091] In the formula: , These represent the levels corresponding to the i-th data group; , These represent the average rank of the two groups of variables.

[0092] Based on the above formula, the Spearman correlation coefficient between each feature and the true amplitude of the transient overvoltage is calculated. If the correlation coefficient between a feature and the true amplitude of the transient overvoltage is lower than the first preset threshold, the feature is determined to be a weakly correlated feature. All weakly correlated features are removed from the original electrical quantity data to obtain a set of remaining features after preliminary screening.

[0093] For the remaining feature set, based on the above formula, the Spearman correlation coefficient between every two features is calculated. For any two features, if their correlation coefficient is higher than the second preset threshold, these two features are considered highly overlapping redundant features, and features with smaller correlation coefficients to the transient overvoltage amplitude are deleted. After two rounds of screening, a set of key electrical quantity features that are streamlined in number, strongly correlated with the transient overvoltage mechanism, and complementary in information are finally determined.

[0094] In this embodiment, by eliminating weakly correlated and redundant features, the dimensionality of the input data is significantly reduced. This directly reduces the computational complexity and training time of the subsequent model, while also reducing the risk of overfitting due to data noise and multicollinearity, which helps to improve the model's generalization ability and prediction accuracy.

[0095] In one possible embodiment, the specific steps of S2 include:

[0096] S2.1 Normalize the transient overvoltage sample set and divide the normalized sample set into a training set, a validation set, and a test set.

[0097] In the specific implementation process, considering that the various electrical quantities constituting the sample input features (such as voltage, power, angle, etc.) have different physical dimensions and orders of magnitude, and that the true amplitude of transient overvoltage is itself a continuous value, the sample input features and the true amplitude of transient overvoltage in the transient overvoltage sample set are first normalized. The normalization formula is as follows:

[0098]

[0099] in, X represents the normalized data value; X represents the original data, which can be the value of any sample input feature or the true amplitude of transient overvoltage (label). This is the minimum value of the feature or label in the sample set; This represents the maximum value of the feature or label in the sample set.

[0100] Through normalization, all input and output data are mapped to a uniform numerical range. Then, the normalized transient overvoltage sample set is randomly divided into training set, validation set and test set according to a predetermined ratio (e.g., 70% : 15% : 15%).

[0101] S2.2. Using the sample input features in the training set as training input and the corresponding true amplitude of transient overvoltage in the training set as the training target, the transient overvoltage amplitude prediction model is trained to obtain the model.

[0102] Please refer to Figure 2 This is a schematic diagram of the structure of a transient overvoltage amplitude prediction model provided in an embodiment of this application. The transient overvoltage amplitude prediction model is constructed using a deep neural network. The input layer receives a sample input feature vector consisting of d feature elements, represented as... ; Figure 2 This illustrates a hidden layer containing n neurons. The neurons in this layer... Receive the weighted signal from the input layer and pass it through a nonlinear activation function. Transformations are performed to learn the complex nonlinear mapping relationships between features. This represents the connection weight matrix between layers, and its output layer uses the Sigmoid activation function, denoted as:

[0103]

[0104] This is the normalized predicted transient overvoltage amplitude of the transient overvoltage amplitude prediction model for the input features of the i-th sample.

[0105] During the training phase, mean squared error is used as the loss function, as shown in the following formula:

[0106]

[0107] Where L1 is the loss function value of the transient overvoltage amplitude prediction model, and N is the number of samples in the current training batch. Let the input feature be the true transient overvoltage amplitude corresponding to the i-th sample. This represents the normalized predicted transient overvoltage amplitude of the transient overvoltage amplitude prediction model for the input features of the i-th sample.

[0108] Based on this loss function, the backpropagation algorithm (BP) is used to... Figure 2 The deep neural network shown is trained using supervised learning. The training process is as follows:

[0109] The sample input features from the training set are batch-input into the network for forward propagation. The deviation between the predicted amplitude of transient overvoltage output by the network and the corresponding true amplitude of transient overvoltage is calculated. Based on this deviation, the gradient information of the loss function with respect to the weights and biases of each layer in the network is calculated. The gradient descent optimization algorithm is used to dynamically adjust the network parameters (weights and biases) along the direction of gradient descent to achieve accurate fitting of the feature mapping relationship.

[0110] Through multiple rounds of iterative updates, the network parameters are continuously corrected, causing the loss function value to gradually decrease and converge. The iteration process stops when the network's error on the training set meets the preset accuracy requirements and its performance on the validation set stabilizes. At this point, the optimized network parameters are saved, resulting in the finally trained transient overvoltage amplitude prediction model.

[0111] S2.3 Input the sample input features from the validation set into the transient overvoltage amplitude prediction model to obtain the corresponding transient overvoltage prediction amplitude as the amplitude to be evaluated.

[0112]

[0113] in, Input features for the i-th sample; This represents a transient overvoltage amplitude prediction model with parameter θ. This represents the normalized predicted transient overvoltage amplitude of the transient overvoltage amplitude prediction model for the input features of the i-th sample.

[0114] S2.4. Using the sample input features from the validation set and the corresponding amplitude to be evaluated as training inputs, a risk grading model is trained to obtain the model.

[0115] In one possible embodiment, step S2.4 includes:

[0116] The prediction error is calculated based on the amplitude to be evaluated and the corresponding true amplitude of transient overvoltage in the validation set. Based on the prediction error, the risk value of each sample is calculated. According to the preset risk threshold, the risk value of each sample is mapped to a risk level. The combination of the sample input features of each sample in the validation set and the corresponding amplitude to be evaluated is used as the training input, and the mapped risk level is used as the training target to train and obtain the risk grading model.

[0117] In the specific implementation process, the formula for calculating the prediction error is as follows:

[0118]

[0119] in, Let be the prediction error for the i-th sample; Let be the true amplitude of the transient overvoltage of the i-th sample; The amplitude to be evaluated for the i-th sample is obtained from the transient overvoltage amplitude prediction model and after inverse normalization. The inverse normalization formula is as follows:

[0120]

[0121] This represents the normalized predicted transient overvoltage amplitude of the transient overvoltage amplitude prediction model output for the i-th sample input feature. and These are the maximum and minimum values ​​used when normalizing the true amplitude of transient overvoltage for all samples in the sample set.

[0122] For the i-th sample in the validation set, input the sample into the feature vector. Its corresponding amplitude to be evaluated By combining these features, the combined input features of the risk grading model are formed. .

[0123] Please refer to Figure 3 This is a schematic diagram of the structure of a risk grading model provided in an embodiment of this application. Risk grading is also constructed using a deep neural network. The input layer receives a combined input feature vector consisting of d feature elements, represented as... ; Figure 3 This illustrates a hidden layer containing n neurons. The neurons in this layer... Receive the weighted signal from the input layer and pass it through a nonlinear activation function. Transformations are performed to learn the complex nonlinear mapping relationships between features. This represents the connection weight matrix between layers. The output layer contains two neurons, corresponding to two risk levels.

[0124] The risk grading model outputs the predicted probability vectors of the two types of risks for the i-th sample. for:

[0125]

[0126] in, This represents the probability that the prediction result of the i-th sample belongs to low risk. The prediction result represents the probability that the i-th sample belongs to the high-risk category.

[0127] The risk level predicted by the final risk grading model for:

[0128]

[0129] in, This indicates taking the maximum value.

[0130] The loss function during the training phase is:

[0131]

[0132] Where L2 is the loss function value of the risk grading model, and N is the number of samples in the current training batch; Let i be the true risk category of the i-th sample. This represents the predicted probability of the actual risk category by the risk grading model.

[0133] In one possible implementation, the risk value for each sample is calculated based on the prediction error, including:

[0134] The actual transient overvoltage amplitude of each sample is compared with the preset transient overvoltage safety threshold. Based on the comparison results, the corresponding risk sensitivity coefficient is selected to weight the prediction error and obtain the risk value of each sample.

[0135] In the specific implementation process, the formula for calculating the risk value is as follows:

[0136] when hour, ;

[0137] when hour, ;

[0138] in, This is the safety threshold for transient overvoltage. Let be the true amplitude of the transient overvoltage of the i-th sample; R is the risk value; Let be the prediction error for the i-th sample; , , and This represents the risk sensitivity coefficient.

[0139] Please refer to Figure 4 This is a schematic diagram of the risk function provided in this application embodiment under the condition that the actual amplitude of the transient overvoltage does not exceed the safety threshold. The actual amplitude of the transient overvoltage does not exceed the transient overvoltage safety threshold (…). ) condition, when When the predicted amplitude is less than the actual amplitude, and neither exceeds the transient overvoltage threshold, the impact on the safe operation of the system is relatively small. When the predicted amplitude is greater than the actual amplitude, it may cause the predicted amplitude to exceed the transient overvoltage safety threshold. However, due to conservative decision-making, this situation will not directly reduce system safety, but it may still lead to unnecessary control and losses. .

[0140] Please refer to Figure 5 This is a schematic diagram of the risk function provided in the embodiments of this application when the actual amplitude exceeds the safety threshold. The actual amplitude of the transient overvoltage exceeds the transient overvoltage safety threshold (…). ) condition, when At this time, the predicted amplitude is less than the actual amplitude, but since the actual amplitude exceeds the transient overvoltage threshold, the prediction deviation may lead to an insufficient assessment of the system's operational safety, neglecting necessary protective measures and threatening the safe operation of the system. Therefore, a very high risk coefficient is selected. .when When the predicted amplitude is greater than the actual amplitude, the system will activate protection based on a more severe overvoltage level. This could firstly cause unnecessary control and losses, and secondly, increase the risk of equipment damage under severe overvoltage levels. Therefore, a larger risk coefficient should be selected. ,in .

[0141] Combining the two scenarios, the risk sensitivity coefficient can be obtained as follows: .

[0142] In one possible embodiment, the risk value of each sample is mapped to a risk level according to a preset risk threshold, including: if the risk value of any sample is greater than the preset risk threshold, the risk level of the sample is high risk; if the risk value of any sample is less than the preset risk threshold, the risk level of the sample is low risk.

[0143] In the specific implementation process, the mapping formula is as follows:

[0144]

[0145] Where C represents the risk level; 0 represents low risk; 1 represents high risk; and R represents the risk value. This indicates a preset risk threshold.

[0146] In summary, this application provides a method for predicting the amplitude of transient overvoltages in power grids at the sending end. Based on traditional transient overvoltage amplitude prediction models, a risk classification model is innovatively introduced and constructed. This method combines the transient overvoltage amplitude prediction model with the risk classification model, avoiding the problems of complex modeling and high computational load associated with model-driven methods, while overcoming the shortcomings of insufficient prediction accuracy and reliability of data-driven methods under different operating conditions.

[0147] The flowchart for the offline training phase is as follows: Figure 6 As shown, a sample set is constructed and divided using selected key power grid response characteristics and anticipated fault information as input, and the simulated transient overvoltage amplitude as output. A transient overvoltage amplitude prediction model is trained using the training set; then, using the validation set and the output of the prediction model, a risk classification model is trained based on a risk function reflecting the power grid safety mechanism. The transient overvoltage amplitude prediction model and the risk classification model are stored in an offline model library.

[0148] The flowchart for the online application phase is shown in Figure 7. By combining the transient overvoltage amplitude prediction model with the risk classification model, a fusion network is formed that can simultaneously output the predicted amplitude of the transient overvoltage and its risk level. For low-risk samples, the prediction results can be directly used for subsequent evaluation. For high-risk samples, the prediction conclusions should not be directly used; simulation verification should be triggered first to ensure both the accuracy and safety of the prediction results.

[0149] Based on the same inventive concept, please refer to Figure 8 This application also provides a device for predicting the amplitude of transient overvoltages in a power grid at the sending end, the device comprising:

[0150] The sample construction module is used to generate a transient overvoltage sample set through time-domain simulation; each sample in the transient overvoltage sample set includes sample input features and the corresponding true amplitude of transient overvoltage;

[0151] The offline training module is used to train a transient overvoltage amplitude prediction model and a risk grading model based on a transient overvoltage sample set. The transient overvoltage amplitude prediction model is used to predict the transient overvoltage amplitude based on the sample input features. The risk grading model is used to evaluate the risk level of the amplitude prediction result based on the sample input features and their corresponding transient overvoltage amplitudes.

[0152] The online application module is used to extract real-time key response features from real-time power grid measurement data that are of the same origin as the sample input features. The real-time key response features are input into the transient overvoltage amplitude prediction model to obtain the corresponding transient overvoltage prediction amplitude. The real-time key response features and the corresponding transient overvoltage prediction amplitude are input into the risk classification model to obtain the corresponding risk level.

[0153] The module is used to obtain the transient overvoltage prediction amplitude as the final transient overvoltage prediction amplitude if the risk level is low; if the risk level is high, it triggers time-domain simulation to verify the transient overvoltage prediction amplitude and uses the simulation result as the final transient overvoltage prediction amplitude.

[0154] Optionally, the sample building module is specifically used for:

[0155] Set up simulation scenarios that include various operating conditions and various fault disturbances, perform simulations based on the time-domain simulation model of the sending-end power grid, and output the original electrical quantity data and the corresponding true amplitude of transient overvoltage before the power grid fault.

[0156] Key electrical quantity features are selected from the raw electrical quantity data;

[0157] Extract anticipated fault information from the configuration of the simulation scenario;

[0158] The key electrical quantity characteristics and anticipated fault information are combined into sample input features, and the corresponding true amplitude of transient overvoltage is used as the label to construct a transient overvoltage sample set.

[0159] Optionally, the sample building module is specifically used for:

[0160] Calculate the correlation coefficient between each feature in the original electrical quantity data and the true amplitude of the transient overvoltage, remove features with correlation coefficients lower than the first preset threshold, and obtain the remaining feature set;

[0161] Calculate the correlation coefficients between the features in the remaining feature set, remove features with correlation coefficients higher than the second preset threshold, and obtain the key electrical quantity features.

[0162] Optionally, the offline training module is specifically used for:

[0163] The transient overvoltage sample set is normalized, and the normalized sample set is divided into a training set, a validation set, and a test set.

[0164] The transient overvoltage amplitude prediction model is trained by using the sample input features in the training set as the training input and the corresponding true amplitude of transient overvoltage in the training set as the training target.

[0165] The sample input features in the validation set are input into the transient overvoltage amplitude prediction model to obtain the corresponding transient overvoltage prediction amplitude as the amplitude to be evaluated;

[0166] The risk grading model is trained by using the sample input features from the validation set and the corresponding amplitude to be evaluated as training inputs.

[0167] Optionally, the offline training module is specifically used for:

[0168] The prediction error is calculated based on the amplitude to be evaluated and the corresponding true amplitude of transient overvoltage in the validation set.

[0169] Calculate the risk value for each sample based on the prediction error;

[0170] Based on a preset risk threshold, the risk value of each sample is mapped to a risk level;

[0171] The risk grading model is trained by using the combination of sample input features and corresponding amplitudes to be evaluated for each sample in the validation set as training input and the mapped risk level as training objective.

[0172] Optionally, the offline training module is specifically used for:

[0173] The actual transient overvoltage amplitude of each sample is compared with the preset transient overvoltage safety threshold.

[0174] Based on the comparison results, the corresponding risk sensitivity coefficient is selected to weight the prediction error and obtain the risk value of each sample.

[0175] Optionally, the formula for calculating the risk value is as follows:

[0176] when hour, ;

[0177] when hour, ;

[0178] in, This is the safety threshold for transient overvoltage. Let be the true amplitude of the transient overvoltage of the i-th sample; R is the risk value; Let be the prediction error for the i-th sample; , , and Risk sensitivity coefficient; .

[0179] Optionally, the device also includes an early warning module, which is specifically used to: after obtaining the final predicted amplitude of transient overvoltage, compare the final predicted amplitude of transient overvoltage with a preset transient overvoltage safety threshold.

[0180] If the final predicted amplitude of transient overvoltage exceeds the transient overvoltage safety threshold, an early warning message will be generated.

[0181] It should be noted that each module in the transient overvoltage amplitude prediction device for the sending-end power grid in this embodiment corresponds one-to-one with each step in the transient overvoltage amplitude prediction method for the sending-end power grid in the aforementioned embodiment. Therefore, the specific implementation of this embodiment can refer to the implementation of the transient overvoltage amplitude prediction method for the sending-end power grid described above, and will not be repeated here.

[0182] Based on the same inventive concept, this application also provides a computer device, which includes a processor, a memory, and a computer program stored in the memory. The computer program is executed by the processor to implement the aforementioned method for predicting the amplitude of transient overvoltage in the sending-end power grid.

[0183] Based on the same inventive concept, this application also provides a computer storage medium storing a computer program, which is executed by a processor to implement the aforementioned method for predicting the amplitude of transient overvoltage in the sending-end power grid.

[0184] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a device including one or any combination of the above-mentioned memories. The computer may be a variety of computing devices, including smart terminals and servers.

[0185] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0186] As an example, executable instructions may, but do not necessarily, correspond to files in the file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborative files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0187] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0188] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0189] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0190] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. 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 predicting the amplitude of transient overvoltages in a power grid at the sending end, characterized in that, include: A transient overvoltage sample set is generated through time-domain simulation; Each sample in the transient overvoltage sample set includes sample input features and the corresponding true amplitude of transient overvoltage; Based on the transient overvoltage sample set, a transient overvoltage amplitude prediction model and a risk classification model are trained and obtained; the transient overvoltage amplitude prediction model is used to predict the transient overvoltage amplitude according to the sample input features; The risk grading model is used to assess the risk level of the amplitude prediction result based on the sample input features and their corresponding transient overvoltage amplitudes. Extract real-time key response features that are of the same origin as the sample input features from the real-time measurement data of the power grid. Input the real-time key response features into the transient overvoltage amplitude prediction model to obtain the corresponding transient overvoltage prediction amplitude. Input the real-time key response features and the corresponding transient overvoltage prediction amplitude into the risk classification model to obtain the corresponding risk level. If the risk level is low risk, then the predicted amplitude of the transient overvoltage will be used as the final predicted amplitude of the transient overvoltage. If the risk level is high, a time-domain simulation is triggered to verify the predicted amplitude of the transient overvoltage, and the simulation result is used as the final predicted amplitude of the transient overvoltage.

2. The method for predicting the amplitude of transient overvoltage in a power grid oriented towards the sending end, as described in claim 1, is characterized in that... The generation of transient overvoltage sample sets through time-domain simulation includes: Set up simulation scenarios that include various operating conditions and various fault disturbances, perform simulations based on the time-domain simulation model of the sending-end power grid, and output the original electrical quantity data and the corresponding true amplitude of transient overvoltage before the power grid fault. Key electrical quantity features are selected from the raw electrical quantity data; Extract the anticipated fault information from the configuration of the simulation scenario; The key electrical quantity features and the anticipated fault information are combined into the sample input features, and the corresponding true amplitude of transient overvoltage is used as the label to construct the transient overvoltage sample set.

3. The method for predicting the amplitude of transient overvoltage in a power grid at the sending end according to claim 2, characterized in that, The step of filtering key electrical quantity features from the original electrical quantity data includes: Calculate the correlation coefficient between each feature in the original electrical quantity data and the true amplitude of the transient overvoltage, remove features with correlation coefficients lower than a first preset threshold, and obtain the remaining feature set; Calculate the correlation coefficient between each feature in the remaining feature set, remove features with a correlation coefficient higher than a second preset threshold, and obtain the key electrical quantity features.

4. The method for predicting the amplitude of transient overvoltage in a power grid at the sending end according to claim 1, characterized in that, The process of training a transient overvoltage amplitude prediction model and a risk classification model based on the transient overvoltage sample set includes: The transient overvoltage sample set is normalized, and the normalized sample set is divided into a training set, a validation set, and a test set. Using the sample input features in the training set as training input and the actual amplitude of the transient overvoltage corresponding to the training set as the training target, the transient overvoltage amplitude prediction model is trained to obtain the model. The sample input features in the validation set are input into the transient overvoltage amplitude prediction model to obtain the corresponding transient overvoltage predicted amplitude as the amplitude to be evaluated. The risk grading model is trained by using the sample input features and the corresponding amplitude to be evaluated in the validation set.

5. The method for predicting the amplitude of transient overvoltage in a power grid at the sending end according to claim 4, characterized in that, The step of training a risk grading model using sample input features and corresponding amplitudes to be evaluated from the validation set includes: The prediction error is calculated based on the amplitude to be evaluated and the corresponding true amplitude of transient overvoltage in the verification set; Based on the prediction error, calculate the risk value for each sample; Based on a preset risk threshold, the risk value of each sample is mapped to a risk level; The risk grading model is trained by using the combination of sample input features and corresponding amplitude to be evaluated for each sample in the validation set as training input and the mapped risk level as training objective.

6. The method for predicting the amplitude of transient overvoltage in a power grid at the sending end according to claim 5, characterized in that, The calculation of the risk value for each sample based on the prediction error includes: The actual transient overvoltage amplitude of each sample is compared with the preset transient overvoltage safety threshold. Based on the comparison results, the corresponding risk sensitivity coefficient is selected to weight the prediction error and obtain the risk value of each sample.

7. The method for predicting the amplitude of transient overvoltage in a power grid oriented towards the sending end, as described in claim 6, is characterized in that... The formula for calculating the risk value is as follows: when hour, ; when hour, ; in, The transient overvoltage safety threshold; R represents the true amplitude of the transient overvoltage of the i-th sample; R is the risk value. Let be the prediction error for the i-th sample; , , and The risk sensitivity coefficient is mentioned above; .

8. The method for predicting the amplitude of transient overvoltage in a power grid oriented towards the sending end, as described in claim 1, is characterized in that... After obtaining the final predicted amplitude of the transient overvoltage, the method further includes: The final predicted amplitude of transient overvoltage is compared with the preset transient overvoltage safety threshold. If the final predicted amplitude of transient overvoltage exceeds the transient overvoltage safety threshold, an early warning message is generated.

9. A device for predicting the amplitude of transient overvoltages in a power grid at the sending end, characterized in that, include: The sample construction module is used to generate transient overvoltage sample sets through time-domain simulation. Each sample in the transient overvoltage sample set includes sample input features and the corresponding true amplitude of transient overvoltage; An offline training module is used to train a transient overvoltage amplitude prediction model and a risk classification model based on the transient overvoltage sample set; the transient overvoltage amplitude prediction model is used to predict the transient overvoltage amplitude based on the sample input features; The risk grading model is used to assess the risk level of the amplitude prediction result based on the sample input features and their corresponding transient overvoltage amplitudes. The online application module is used to extract real-time key response features from real-time power grid measurement data that are of the same origin as the sample input features, input the real-time key response features into the transient overvoltage amplitude prediction model to obtain the corresponding transient overvoltage prediction amplitude, and input the real-time key response features and the corresponding transient overvoltage prediction amplitude into the risk classification model to obtain the corresponding risk level. The module is configured to use the predicted transient overvoltage amplitude as the final predicted transient overvoltage amplitude if the risk level is low. If the risk level is high, a time-domain simulation is triggered to verify the predicted amplitude of the transient overvoltage, and the simulation result is used as the final predicted amplitude of the transient overvoltage.

10. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement a method for predicting the amplitude of transient overvoltage in a power grid as described in any one of claims 1-8.

Images