Power grid stability analysis and evaluation method and system based on artificial intelligence

Abstract

The invention discloses a power grid stability analysis and evaluation method based on artificial intelligence, and the method comprises the steps: obtaining sample data, and dividing the sample data into training data and test data according to a preset proportion; establishing a power grid stability analysis model according to the convolutional neural network and setting model parameters; inserting an extrusion excitation module into the convolutional neural network, and adjusting the weight of the feature channel to optimize the power grid stability analysis model; training a power grid stability analysis model by adopting the sample data and the corresponding annotation data; and analyzing the power grid operation data based on the trained power grid stability analysis model, judging whether the power grid is stable or not, and predicting the instability mode of the power grid. According to the method, the artificial intelligence technology and the power grid stability analysis are combined, the defects of a traditional method in the aspects of alternating current power flow reduction accuracy, transient stability analysis efficiency, steady-state operation risk identification, optimization decision automation and the like are effectively overcome, and the accuracy and efficiency of power grid stability analysis are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of power grid stability analysis, and in particular to a power grid stability analysis and evaluation method and system based on artificial intelligence. Background Technology

[0002] The stable operation of the power system is the cornerstone of modern socio-economic development, directly affecting the safe and reliable operation of industrial production, residential lives, and national infrastructure. With the acceleration of globalization and rapid technological advancements, electricity demand continues to rise, and the power system has undergone large-scale expansion and technological upgrades. The following is the background information on current power grid development:

[0003] The continuous expansion of power systems: To meet the ever-increasing demand for electricity, countries are constantly building and upgrading their power systems, resulting in a continuous expansion of power grid coverage and a sustained increase in installed power generation capacity. Large power plants, ultra-high-voltage transmission lines, and complex distribution networks constitute the massive structure of modern power systems. Taking China as an example, in the past few decades, its installed power capacity has increased dozens of times, and its power grid scale has jumped to the forefront of the world.

[0004] Large-scale integration of renewable energy generation: To address climate change and reduce greenhouse gas emissions, the world is vigorously developing renewable energy generation technologies such as wind power and photovoltaics. Renewable energy generation is characterized by significant intermittency and volatility, and its large-scale integration into the power grid poses a significant challenge to the stable operation of traditional power systems. For example, wind power output is significantly affected by wind speed, exhibiting random fluctuations; photovoltaic power generation depends on solar radiation conditions, showing obvious diurnal variations and weather dependence. The integration of these renewable energy generation methods greatly increases the difficulty of power grid power balance regulation, and makes voltage and frequency stability more difficult to control.

[0005] The widespread application of DC transmission technology: Due to its advantages such as large transmission capacity, low loss, and suitability for long-distance transmission, DC transmission technology has been widely used in modern power systems. However, the fault characteristics of DC transmission systems differ significantly from those of AC transmission systems. When a DC fault occurs, it may lead to rapid changes in grid power, causing system oscillations and other problems, increasing the complexity and difficulty of grid stability analysis.

[0006] The advancement of smart grid construction: The concept of smart grids aims to achieve the informatization, automation, and interactivity of power systems. Through the deep integration of advanced communication, information, control, and power technologies, it enhances the operational efficiency, reliability, and flexibility of the power grid. Smart grid construction involves multiple aspects, including the integration of distributed energy resources, demand-side management, and intelligent dispatching, placing higher demands on the stability analysis and control of the power grid.

[0007] The deepening of electricity market reform: With the continuous advancement of electricity market reform, the operation mode and management model of the power system have undergone profound changes. The competitive mechanism of the electricity market makes generation planning and load dispatching more flexible and diverse, and increases the real-time nature and complexity of electricity trading. In this market environment, the power grid needs more accurate and faster stability analysis tools to support dispatching decisions in order to ensure the reliability and economy of power supply.

[0008] The diversification of user-side demands: Modern electricity users have increasingly higher requirements for power quality, demanding not only reliable power supply but also stringent requirements for indicators such as voltage stability, frequency stability, and harmonic content. Meanwhile, the widespread application of various new types of electrical equipment (such as electric vehicles, energy storage systems, and controllable loads) has made user-side load characteristics more complex and variable, posing new challenges to the stable operation of the power grid.

[0009] Impacts of Extreme Weather and Natural Disasters: Global climate change has led to more frequent extreme weather events, such as torrential rains, floods, typhoons, and snowstorms. These natural disasters pose a serious threat to the safe and stable operation of power systems. Extreme weather can cause damage to power equipment, transmission line failures, and substation outages, resulting in widespread power outages and causing enormous losses to the socio-economic landscape.

[0010] Aging and Upgrading of Power Systems: Some power system equipment has been in operation for a long time, and aging problems are becoming increasingly prominent, leading to a decline in equipment reliability and performance. At the same time, the pace of technological upgrading and replacement in power systems is accelerating, with the continuous emergence of new equipment and technologies, requiring power grid stability analysis methods to adapt to changes in equipment characteristics and technological development needs.

[0011] Against this backdrop, the importance of power system stability analysis has become increasingly prominent. Stability analysis is a crucial link in ensuring the safe and reliable operation of the power grid, involving multiple aspects such as power angle stability, voltage stability, and frequency stability. Traditional stability analysis methods mainly rely on physical models and analytical methods, such as time-domain simulation and eigenvalue analysis. These methods have provided strong support for the stable operation of power systems for a certain period, but they have gradually revealed some limitations when facing the complexity and uncertainty of new power systems. For example, traditional methods are difficult to handle the randomness and volatility brought about by the large-scale integration of new energy sources, are not accurate enough in simulating the dynamic characteristics of complex power grids, have low computational efficiency, and cannot meet the needs of real-time dispatch decision-making. Therefore, there is an urgent need for an efficient, accurate, and intelligent power grid stability analysis method that can adapt to the characteristics of new power systems to address the many challenges facing current and future power systems. Summary of the Invention

[0012] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0013] In view of the problems existing in the prior art, the present invention is proposed.

[0014] Therefore, the technical problem to be solved by this invention is that traditional power grid stability analysis methods are difficult to adapt to the complexity and uncertainty of new power systems. They have problems such as difficulty in AC power flow reconstruction, insufficient identification of steady-state operation risks, low efficiency of transient stability analysis, and low degree of automation of optimization decision-making, which make it difficult to meet the high requirements of modern power systems for stability and reliability.

[0015] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a power grid stability analysis and evaluation method based on artificial intelligence, which includes acquiring power grid steady-state information and dynamic information to form sample data, and dividing the sample data into training data and test data according to a preset ratio;

[0016] A power grid stability analysis model is established based on a convolutional neural network, and the model parameters are set.

[0017] A squeezing excitation module is inserted into the convolutional neural network to adjust the weights of the feature channels in order to optimize the power grid stability analysis model;

[0018] The power grid stability analysis model is trained using the sample data and the corresponding labeled data.

[0019] The trained power grid stability analysis model is used to analyze real-time power grid operation data to determine whether the power grid is stable and to predict the instability mode of the power grid.

[0020] As a preferred embodiment of the artificial intelligence-based power grid stability analysis and evaluation method of the present invention, the specific steps of obtaining power grid steady-state information and dynamic information to form sample data include:

[0021] Simulation conditions are set, and based on these conditions, the power grid simulation operation generates steady-state and dynamic information. The simulation conditions include power output, load level, fault type, and topology information.

[0022] Extract the current, voltage, and power data from the steady-state and dynamic information;

[0023] The current, voltage, and power data are cleaned and preprocessed to form sample data.

[0024] As a preferred embodiment of the artificial intelligence-based power grid stability analysis and evaluation method of the present invention, the specific steps of obtaining power grid steady-state information and dynamic information to form sample data include:

[0025] Extract historical data of the power grid, including current, voltage, and power data of the power grid under various operating conditions;

[0026] The current, voltage, and power data are cleaned and preprocessed to form sample data.

[0027] As a preferred embodiment of the artificial intelligence-based power grid stability analysis and evaluation method of the present invention, training the power grid stability analysis model using the sample data and the labeled data corresponding to the sample data specifically includes:

[0028] By combining the experience of power grid simulation experts, the dominant instability modes of the training data are labeled to generate labeled data;

[0029] Input the training data, perform forward propagation calculations, and obtain the output results;

[0030] The difference between the model output and the labeled data is calculated using a loss function;

[0031] The power grid stability analysis model adopts the Adam optimization algorithm and updates the model parameters with cross-entropy loss as the objective function to minimize the loss function;

[0032] The performance of the power grid stability analysis model is evaluated using the test data, and the generalization ability of the power grid stability analysis model is verified.

[0033] As a preferred embodiment of the power grid stability analysis and evaluation method based on artificial intelligence described in this invention, the training data further includes validation data;

[0034] After training the power grid stability analysis model using the training data, the power grid stability analysis model is evaluated using the validation data, and the loss and accuracy on the validation data are calculated.

[0035] The hyperparameters of the power grid stability analysis model are adjusted based on the losses and accuracy.

[0036] As a preferred embodiment of the artificial intelligence-based power grid stability analysis and evaluation method of the present invention, the method further includes, before training the power grid stability analysis model using the sample data and the labeled data corresponding to the sample data:

[0037] The z-score standardization technique is used to standardize the data of different dimensions in the sample data.

[0038] As a preferred embodiment of the artificial intelligence-based power grid stability analysis and evaluation method of the present invention, the specific steps of the squeezing excitation module for optimizing the power grid stability analysis model include:

[0039] Transformation operation: Convolutional transformation of the initial features, outputting the initial feature map for each channel;

[0040] Squeezing operation: The initial feature map of each feature channel is averaged using global average pooling;

[0041] Activation operation: The features obtained from the previous extrusion are processed through two fully connected layers to generate weights of different importance for each channel;

[0042] Scaling operation: The weights generated in the excitation phase are applied to the original feature map. The feature channels are weighted by multiplying the weight of each channel element-wise with the corresponding feature map.

[0043] As a preferred embodiment of the power grid stability analysis and evaluation method based on artificial intelligence described in this invention, the squeezing operation specifically includes: averaging the initial feature map of each feature channel and compressing each two-dimensional initial feature map into a single value.

[0044] The incentive operation specifically includes:

[0045] The input features are mapped to a low-dimensional space, and nonlinearity is introduced using the ReLU activation function.

[0046] The feature in the low-dimensional space is mapped back to the initial feature space to generate a weight vector with the same number of input feature channels. This weight vector is then compressed to between 0 and 1 using the sigmoid activation function to form the activation coefficient of each channel. The activation coefficient represents the importance of the channel in the current task.

[0047] As a preferred embodiment of the artificial intelligence-based power grid stability analysis and evaluation method of the present invention, the scaling operation specifically includes:

[0048] Obtain the weights after the activation operation. If the weights are greater than a preset value, enlarge the feature map of that channel. If the weights are less than a preset value, suppress the feature map of that channel.

[0049] This invention provides the following technical solution: an artificial intelligence-based power grid stability analysis and evaluation system, which includes a data acquisition module for acquiring power grid operation data;

[0050] The power grid stability analysis model, trained based on the sample data and the corresponding labeled data as described above, is used to analyze and determine whether the power grid is stable and to predict the instability mode of the power grid.

[0051] The beneficial effects of this invention are as follows: By combining artificial intelligence technology with power grid stability analysis, this invention effectively solves the shortcomings of traditional methods in terms of accuracy of AC power flow reconstruction, efficiency of transient stability analysis, identification of steady-state operation risks, and automation of optimization decisions. It significantly improves the accuracy and efficiency of power grid stability analysis, realizes rapid and accurate assessment of power grid stability and optimization decision support, powerfully promotes the intelligent development of new power systems, and provides a solid technical guarantee for the reliable operation of the power grid. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0053] Figure 1 This is a flowchart of an artificial intelligence-based power grid stability analysis and evaluation method in one embodiment.

[0054] Figure 2 This is a diagram of the CEPRI-36 node system in one embodiment.

[0055] Figure 3 This is a graph showing the loss function and accuracy changes of SE-CNN during the training process of the CEPRI-36 node system in one embodiment.

[0056] Figure 4 A visualization of the SENet module output (channel weights) in one embodiment: a comparison of the start and end of training.

[0057] Figure 5 This is a confusion matrix diagram of the prediction results of SE-CNN in a real power grid in one embodiment. Detailed Implementation

[0058] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0059] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0060] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0061] In one embodiment of the present invention, an artificial intelligence-based power grid stability analysis and evaluation method is provided, which can assess the stability of power grid operation. This method primarily utilizes artificial intelligence technology, particularly convolutional neural networks (CNNs) combined with a squeeze excitation module (SENet), to perform real-time analysis and prediction of power grid stability. It can automatically learn complex patterns in power grid operation by processing steady-state and dynamic information, determine whether the power grid is stable, and predict possible instability modes. This system plays a crucial role in monitoring the stable operation of power systems, preventing faults, and improving power grid operating efficiency. It is particularly suitable for modern complex power grid environments, helping power dispatchers take timely measures to prevent power grid accidents and ensure the reliability and security of power supply.

[0062] like Figure 1 As shown, the power grid stability analysis and assessment methods based on artificial intelligence include:

[0063] Step S1: Obtain steady-state and dynamic information of the power grid to form sample data, and divide the sample data into training data and test data according to a preset ratio.

[0064] Steady-state information of the power grid refers to the data performance of the power grid under normal operating conditions, such as the voltage amplitude and phase of each node, and the active power and reactive power of the lines, which are stable operating data when there are no faults or disturbances.

[0065] Dynamic information covers the dynamic response data of the power grid after being subjected to various disturbances (such as short-circuit faults, load changes, generator tripping, etc.), including voltage and current changes at the moment of the fault and in the subsequent transient process, as well as fluctuations in system frequency.

[0066] Step S2: Establish a power grid stability analysis model based on the convolutional neural network and set the model parameters.

[0067] Convolutional Neural Networks (CNNs) are deep learning models specifically designed for processing data with a grid structure (such as images, time series data, etc.), and they excel in feature extraction and pattern recognition.

[0068] When constructing a power grid stability analysis model, it is necessary to reasonably set parameters such as the number of layers in the convolutional neural network, the number of neurons in each layer, the kernel size, the stride, and the padding method, based on the characteristics of the power grid data and the needs of the analysis task.

[0069] CNNs, with their unique structures including convolutional layers, pooling layers, and fully connected layers, can automatically extract deep features when processing power grid data with spatial or temporal correlations. For example, convolutional layers perform convolution operations by sliding convolution kernels across the input data; the calculation formula is as follows:

[0070] Where yi,j represents the value of the output feature map at position (i,j), wk is the convolution kernel parameter, xi+k,j is the input data, and b is the bias term. This operation can extract local features from power grid data, such as voltage fluctuation patterns and power change trends.

[0071] Step S3: Insert a squeezing excitation module into the convolutional neural network and adjust the weights of the feature channels to optimize the power grid stability analysis model.

[0072] The Squeeze Excitation Module (SENet) can automatically learn and adjust the weights of feature channels, making the model focus more on feature channels that are important for power grid stability analysis and suppressing unimportant feature channels, thereby improving the model's performance and generalization ability.

[0073] Step S4: Train the power grid stability analysis model using the sample data and the labeled data corresponding to the sample data.

[0074] During training, sample data is input into the model, and the model output is calculated through forward propagation. Then, the loss function is used to measure the difference between the model output and the actual labeled data. The Adam optimization algorithm is used with cross-entropy loss as the objective function to update the model parameters and minimize the loss function, so that the model can gradually learn the characteristic patterns of power grid stability and instability modes.

[0075] Step S5: Analyze the real-time power grid operation data based on the trained power grid stability analysis model to determine whether the power grid is stable and predict the instability mode of the power grid.

[0076] The real-time power grid operation data is preprocessed using the same process as the training sample data and then input into the trained model. The model evaluates the current power grid operation status based on the learned characteristics and rules, determines whether it is stable, and further predicts possible instability modes, such as power angle instability and voltage instability, so that power dispatchers can take corresponding control measures in a timely manner.

[0077] This invention effectively addresses the shortcomings of traditional methods in terms of accuracy of AC power flow reconstruction, efficiency of transient stability analysis, identification of steady-state operation risks, and automation of optimization decisions by combining artificial intelligence technology with power grid stability analysis. It significantly improves the accuracy and efficiency of power grid stability analysis, enables rapid and accurate assessment of power grid stability and provides support for optimization decisions, powerfully promotes the intelligent development of new power systems, and provides a solid technical guarantee for the reliable operation of the power grid.

[0078] In one embodiment, obtaining steady-state and dynamic information of the power grid to form sample data specifically includes:

[0079] Simulation conditions are set, and based on these conditions, the power grid simulation operation generates steady-state and dynamic information. The simulation conditions include power output, load level, fault type, and topology information.

[0080] Extract the current, voltage, and power data from the steady-state and dynamic information;

[0081] The current, voltage, and power data are cleaned and preprocessed to form sample data.

[0082] Specifically, by using specialized power system simulation software (such as PSASP, PSS / E, etc.) to set different simulation conditions, the behavior of the power grid under various normal and fault operating conditions is simulated, and a large amount of steady-state and dynamic data is obtained. Power output can cover the active and reactive power output of different generator sets, load levels can include peak load, off-peak load, and different load distribution scenarios, fault types include various short-circuit faults, open-circuit faults, etc., and topology information involves changes in the network connection structure of the power grid.

[0083] From the power grid operation data obtained from the simulation, key physical quantities such as current, voltage, and power are accurately extracted. These data are core indicators reflecting the power grid's operating status and are crucial for subsequent stability analysis.

[0084] Data cleaning is the process of removing outliers, missing values, noise, and other contaminants from data to ensure its quality and reliability.

[0085] Preprocessing includes operations such as data normalization and standardization, which transform data with different dimensions and numerical ranges into a uniform scale suitable for model input, thereby improving the efficiency and stability of model training.

[0086] In one implementation, obtaining steady-state and dynamic information of the power grid to form sample data specifically includes:

[0087] Extract historical data of the power grid, including current, voltage, and power data of the power grid under various operating conditions;

[0088] The current, voltage, and power data are cleaned and preprocessed to form sample data.

[0089] Historical power grid data is a data resource accumulated over a long period of time during the actual operation of the power grid. It contains real operating information of the power grid under various operating conditions such as different seasons, different load demands, and different fault events. It has high value for model training and validation.

[0090] Data cleaning is the process of removing outliers, missing values, noise, and other contaminants from data to ensure its quality and reliability.

[0091] Preprocessing includes operations such as data normalization and standardization, which transform data with different dimensions and numerical ranges into a uniform scale suitable for model input, thereby improving the efficiency and stability of model training.

[0092] In one implementation, training the power grid stability analysis model using the sample data and the corresponding labeled data specifically includes:

[0093] By combining the experience of power grid simulation experts, the dominant instability modes of the training data are labeled to generate labeled data.

[0094] Due to the specialized and complex nature of power grid stability analysis, it is necessary to rely on the experience of power grid simulation experts to annotate the training data, clearly indicating the power grid instability mode corresponding to each set of data, such as power angle instability and voltage instability, so as to provide accurate supervision information for model training.

[0095] Input the training data, perform forward propagation calculations, and obtain the output results.

[0096] The labeled training data is input into the power grid stability analysis model. The model calculates the output results corresponding to the input data through the forward propagation process, based on the learned weights and bias parameters, which are the judgment of power grid stability and the prediction of instability modes.

[0097] The difference between the model output and the labeled data is calculated using a loss function.

[0098] Loss functions are used to quantify the difference between the model output and the actual labeled data. Commonly used loss functions include mean squared error and cross-entropy loss. Cross-entropy loss is more commonly used in classification problems. It can measure the difference between the model's predicted probability distribution and the true label distribution.

[0099] The power grid stability analysis model employs the Adam optimization algorithm and updates the model parameters using cross-entropy loss as the objective function to minimize the loss function.

[0100] The Adam optimization algorithm is an adaptive learning rate optimization method that combines the advantages of gradient descent and momentum methods. It can automatically adjust the learning rate based on gradient information during model training, thereby improving the convergence speed and stability of the model.

[0101] Using cross-entropy loss as the objective function, the gradient is calculated through backpropagation, and the Adam optimization algorithm is used to update the model parameters (such as weights and biases), continuously reducing the value of the loss function so that the model's prediction results gradually approach the true labeled data.

[0102] The performance of the power grid stability analysis model is evaluated using the test data, and the generalization ability of the power grid stability analysis model is verified.

[0103] After the model training is completed, the model is evaluated using an independent test dataset. The test dataset was not involved in the model training process and can objectively reflect the model's ability to adapt to new data.

[0104] By calculating metrics such as precision, recall, and F1 score on the test dataset, the performance and generalization ability of the model are evaluated to ensure that the model can reliably analyze and predict power grid stability in practical applications. The F1 score is an important metric for evaluating the performance of a classification model; it comprehensively considers the model's precision and recall, and is the harmonic mean of the two. The F1 score ranges from 0 to 1, with higher values ​​indicating better model performance.

[0105] In one embodiment, the training data further includes validation data;

[0106] After training the power grid stability analysis model using the training data, the power grid stability analysis model is evaluated using the validation data, and the loss and accuracy on the validation data are calculated.

[0107] The hyperparameters of the power grid stability analysis model are adjusted based on the losses and accuracy.

[0108] Specifically, before training the power grid stability analysis model using the sample data and the corresponding labeled data, the method further includes:

[0109] The z-score standardization technique is used to standardize the data of different dimensions in the sample data.

[0110] The specific steps of the squeezing excitation module for optimizing the power grid stability analysis model include:

[0111] Transformation operation: Convolutional transformation of the initial features, outputting the initial feature map for each channel;

[0112] Squeezing operation: The initial feature map of each feature channel is averaged using global average pooling;

[0113] Activation operation: The features obtained from the previous extrusion are processed through two fully connected layers to generate weights of different importance for each channel;

[0114] Scaling operation: The weights generated in the excitation phase are applied to the original feature map. The feature channels are weighted by multiplying the weight of each channel element-wise with the corresponding feature map.

[0115] The compression operation specifically includes: averaging the initial feature map of each feature channel and compressing each two-dimensional initial feature map into a single value.

[0116]

[0117] Where zc represents the compression feature of the c-th channel, H and W are the height and width of the feature map, respectively, and Xi,j,c represents the value of the c-th channel at position (i,j).

[0118] The incentive operation specifically includes:

[0119] The input features are mapped to a low-dimensional space, and nonlinearity is introduced using the ReLU activation function.

[0120] The feature in the low-dimensional space is mapped back to the initial feature space to generate a weight vector with the same number of input feature channels. This weight vector is then compressed to between 0 and 1 using the sigmoid activation function to form the activation coefficient of each channel. The activation coefficient represents the importance of the channel in the current task.

[0121] s = F ex (z,W)=σ(g(z,W))=σ(W2δ(W1z));

[0122] In the formula, σ(·) and δ(·) are the sigmoid and ReLU activation functions, respectively; W1 and W2 are the parameters of the two fully connected layers, respectively; r is the proportional parameter that balances model performance and computational complexity; and the generated s is the weight that characterizes the importance of each channel in the initial feature map U.

[0123] The scaling operation specifically includes:

[0124] Obtain the weights after the activation operation. If the weights are greater than a preset value, enlarge the feature map of that channel. If the weights are less than a preset value, suppress the feature map of that channel.

[0125] Another embodiment of the present invention provides an artificial intelligence-based power grid stability analysis and evaluation system, comprising:

[0126] The data acquisition module is used to acquire data on power grid operation.

[0127] The power grid stability analysis model, trained based on the sample data and the corresponding labeled data, is used to analyze and determine whether the power grid is stable and to predict the instability mode of the power grid.

[0128] To verify the effectiveness of the power grid stability analysis model based on SE-CNN (a convolutional neural network with inserted squeezing excitation modules) described in this invention, experiments were conducted on the CEPRI-36 node system of the China Electric Power Research Institute. For example... Figure 2 As shown, the system comprises 8 generators, 36 buses, 26 lines, and 10 loads. A sample dataset was generated according to the scheme shown in Table 1.

[0129] Table 1. Sample Generation Scheme for CEPRI-36 Node System

[0130] Detailed settings Number of species Test System CEPRI-36 Node System 1 Trend level 90％，100％，110％ 3 load motor component ratio 50％，60％，70％，80％，90％ 5 Faulty circuit All non-transformer lines 26 Fault location 2％，20％，50％，80％，98％ 5 Fault duration 0.05s, 0.15s, 0.25s, 0.3s 4 Fault type Three-phase metallic short circuit and tangent (N-1) 1

[0131] A total of 7800 samples were generated. Each sample contains generator power angle data (dimension `200x8`) and bus voltage data (dimension `200x36`) within a 2-second observation window after the fault, with a sampling interval of 0.01 seconds. The dominant instability mode of the samples was labeled according to expert experience rules.

[0132] After the sample data is standardized by `z-score`, it is randomly divided into a training set (85%) and a test set (15%), and the training set is further divided into a validation set.

[0133] The SE-CNN model described in this invention is used for training, with Adam as the optimizer and cross-entropy loss as the objective function.

[0134] Experimental results:

[0135] During training, the loss function continuously decreased, and the loss and accuracy on the training and validation sets converged well, with no obvious overfitting. See [link / reference]. Figure 3 Indication.

[0136] On an independent test set, the SE-CNN model described in this invention achieved an average accuracy of 97.7% in identifying the dominant instability modes (stable, power angle instability, voltage instability).

[0137] Compared to a standard CNN model without the squeeze excitation module (96.9% test accuracy), SE-CNN achieves a 0.8 percentage point improvement in accuracy, validating the effectiveness of the squeeze excitation module in improving model performance through adaptive channel weighting.

[0138] Compared with traditional machine learning models such as decision tree (DT), support vector machine (SVM), k-nearest neighbors (KNN), and random forest (RF), SE-CNN has a significant accuracy advantage (see Table 2 for comparison results).

[0139] Table 2. Comparison of test accuracy of different methods in the CEPRI-36 node system.

[0140]

[0141] SE-CNN increases training time by approximately 16.5% (391.53 seconds vs 456.29 seconds) compared to conventional CNN, but only increases online application (testing) time by 7.8% (0.1146 seconds vs 0.1235 seconds), meeting the real-time requirements of simulation analysis scenarios.

[0142] Visual analysis of the output (channel weights) of the extrusion excitation module (see...) Figure 4 (Illustration) shows that after training, the weights of each channel are significantly differentiated, and the model has successfully learned the relative importance of different feature channels, effectively amplifying the information of key channels.

[0143] To verify the adaptability and scalability of the method described in this invention in large-scale power grids, experiments were conducted on a provincial-level actual power grid model. A sample dataset was generated according to the scheme shown in Table 3.

[0144] Table 3. Actual Power Grid Sample Generation Scheme

[0145] Detailed settings Number of species Test System Actual power grid 1 Trend level Four types were selected from the trend samples. 4 load motor component ratio 60％，75％，90％ 3 Faulty circuit All 500kV lines 313 Fault location 2％，50％ 2 Fault duration 0.1s, 0.2s, 0.25s, 0.3s 4 Fault type Three-phase metallic short circuit and tangent (N-1) 1

[0146] A total of 30,048 labeled samples were generated. Each sample contains power angle data (dimension `200xM`, where M is the number of generators) and voltage data (dimension `200xB`, where B is the number of buses) within the observation window after the fault. The data were also normalized using `z-score`.

[0147] The SE-CNN model, which has the same structure as Example 1 but is adapted in scale, was used for training and testing.

[0148] Experimental results:

[0149] On an independent test set, the SE-CNN model described in this invention still achieves an average accuracy of 97.7% in identifying the dominant instability pattern.

[0150] The conventional CNN model achieved an accuracy of 97.2% on the actual power grid, while SE-CNN maintained a 0.5 percentage point advantage.

[0151] Compared with other machine learning models (DT, SVM, KNN, RF), SE-CNN still leads in accuracy (see Table 4 for comparison results).

[0152] Table 4 Comparison of test accuracy of different methods in actual power grids

[0153]

[0154] Confusion matrix analysis (see Figure 5 (Illustrative image) shows that the model has excellent overall discrimination performance and a very low rate of misclassified samples.

[0155] The results fully demonstrate that the power grid stability analysis and evaluation method based on SE-CNN described in this invention does not decrease in performance with a significant increase in power grid size, has good scalability and engineering applicability, and is suitable for rapid time-series stability evaluation of actual large power grids.

[0156] It is important to note that the constructions and arrangements of this application shown in several different exemplary embodiments are merely illustrative. Although only a few embodiments are described in detail in this disclosure, those who consult this disclosure will readily understand that many modifications are possible (e.g., changes in the size, dimensions, structure, shape, and proportions of various elements, as well as parameter values ​​(e.g., temperature, pressure, etc.), mounting arrangements, use of materials, color, orientation, etc.) without substantially departing from the novel teachings and advantages of the subject matter described in this application). For example, an element shown as integrally formed may be composed of multiple parts or elements, the position of elements may be inverted or otherwise altered, and the nature or number or position of discrete elements may be changed or altered. Therefore, all such modifications are intended to be included within the scope of the invention. The order or sequence of any process or method steps may be changed or rearranged according to alternative embodiments. In the claims, any "device plus function" clause is intended to cover the structure described herein that performs the function, and not only structurally equivalent but also equivalent in structure. Other substitutions, modifications, alterations, and omissions may be made in the design, operation, and arrangement of the exemplary embodiments without departing from the scope of the invention. Therefore, the present invention is not limited to the specific embodiments, but extends to various modifications that still fall within the scope of the appended claims.

[0157] Furthermore, in order to provide a concise description of exemplary embodiments, not all features of actual embodiments (i.e., those features that are not relevant to the currently considered best mode for carrying out the invention, or those features that are not relevant to implementing the invention) may be omitted.

[0158] It should be understood that numerous specific implementation decisions can be made during the development of any practical implementation, such as in any engineering or design project. Such development efforts may be complex and time-consuming, but for those skilled in the art who benefit from this disclosure, the development effort will be a routine work of design, manufacturing, and production without requiring much experimentation.

[0159] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An artificial intelligence-based power grid stability analysis and assessment method, characterized in that, The method comprises the following steps: obtaining steady-state information and dynamic information of a power grid to form sample data, and dividing the sample data into training data and test data according to a preset proportion; establishing a power grid stability analysis model based on a convolutional neural network and setting model parameters; inserting a squeeze excitation module into the convolutional neural network, and adjusting the weights of feature channels to optimize the power grid stability analysis model; training the power grid stability analysis model by using the sample data and label data corresponding to the sample data; analyzing real-time power grid operation data based on the trained power grid stability analysis model, judging whether the power grid is stable, and predicting the instability mode of the power grid.

2. The artificial intelligence based power grid stability analysis and assessment method as claimed in claim 1 wherein: The method of obtaining steady-state information and dynamic information of a power grid to form sample data specifically comprises: setting simulation conditions, and generating steady-state information and dynamic information based on the simulation conditions, wherein the simulation conditions include power output, load level, fault type and topology information; extracting current, voltage and power data from the steady-state information and dynamic information; cleaning and preprocessing the current, voltage and power data to form sample data.

3. The artificial intelligence based power grid stability analysis and assessment method as claimed in claim 1 wherein: The method of obtaining steady-state information and dynamic information of a power grid to form sample data specifically comprises: extracting historical data of the power grid, wherein the historical data includes current, voltage and power data of the power grid under various operating conditions; cleaning and preprocessing the current, voltage and power data to form sample data.

4. The artificial intelligence based power grid stability analysis and assessment method as claimed in claim 1 wherein: The method of training the power grid stability analysis model by using the sample data and label data corresponding to the sample data specifically comprises: annotating the dominant instability mode of the training data based on the experience of power grid simulation experts to generate label data; inputting the training data, performing forward propagation calculation, and obtaining output results; calculating the difference between the model output and the label data by using a loss function; the power grid stability analysis model uses an Adam optimization algorithm, and updates the model parameters by using cross-entropy loss as a target function to minimize the loss function; evaluating the performance of the power grid stability analysis model by using the test data, and verifying the generalization ability of the power grid stability analysis model.

5. The artificial intelligence based power grid stability analysis and assessment method as claimed in claim 4, wherein: The training data also includes validation data; after training the power grid stability analysis model by using the training data, evaluating the power grid stability analysis model by using the validation data, calculating the loss and accuracy of the validation data, and adjusting the hyperparameters of the power grid stability analysis model according to the loss and accuracy. Before training the power grid stability analysis model by using the sample data and label data corresponding to the sample data, the method further comprises:

6. The artificial intelligence based power grid stability analysis and assessment method as claimed in claim 1 wherein: standardizing and converting different dimensional data in the sample data by using z-score standardization technology. The specific steps of the squeeze excitation module for optimizing the power grid stability analysis model comprise:

7. The artificial intelligence based power grid stability analysis and assessment method as claimed in claim 1 wherein: transformation operation: initial feature convolution transformation, each channel outputs an initial feature map; squeeze operation: average calculation of the initial feature map of each feature channel by using global average pooling; excitation operation: processing the features obtained by squeezing through two fully connected layers to generate weights of different importance degrees of each channel. ​ The scaling operation specifically includes: performing average calculation on the initial feature map of each feature channel, and compressing each two-dimensional initial feature map into a single numerical value.

8. The artificial intelligence based power grid stability analysis and assessment method as claimed in claim 7, wherein: The excitation operation specifically includes: The input feature is mapped to a low-dimensional space, and a ReLU activation function is used to introduce nonlinearity. The feature mapping in the low-dimensional space is mapped back to the initial feature space to generate a weight vector with the same number of input feature channels, and is compressed to between 0 and 1 through a sigmoid activation function to form an excitation coefficient of each channel, which represents the importance of the channel in the current task. The scaling operation specifically includes:

9. The artificial intelligence based power grid stability analysis and assessment method as claimed in claim 8, wherein: The weight after the excitation operation is obtained, and when the weight is greater than a preset value, the feature map of the channel is amplified, and when the weight is less than the preset value, the feature map of the channel is suppressed. It includes:

10. An artificial intelligence based power grid stability analysis and assessment system characterized in that, A data acquisition module for acquiring data of power grid operation; The power grid stability analysis model trained based on the sample data and the labeled data corresponding to the sample data in claim 1 is used to analyze and judge whether the power grid is stable, and to predict the instability mode of the power grid. ​

Images