A method and system for extracting power grid operation extreme scenarios

Abstract

The application provides a power grid operation extreme scene extraction method and system, comprising the following steps: obtaining time sequence data of wind power and load output of each node in the power grid as scene data; clustering all the scene data to obtain a plurality of typical wind power-load operation scenes considering time sequence correlation and geographical distribution characteristics; and taking the scene data farthest from the cluster center of each typical wind power-load operation scene as an extreme scene corresponding to the typical wind power-load operation scene. The application extracts the scene point farthest from the cluster center as the extreme scene by using the clustering method on the basis of considering the time sequence corresponding relationship of wind power and load and the geographical distribution characteristics of wind power and load. The method can help the planning personnel to quickly evaluate the operation safety level and improve the rationality and scientificity of off-line safety and stability analysis of large-scale power grids.

Description

Technical Field

[0001] This invention relates to the field of power grid planning and operation analysis, and specifically to a method for extracting extreme power grid operation scenarios based on weighted clustering. Background Technology

[0002] The power system is undergoing rapid transformation and disruptive change driven by the growth of renewable energy. With the large-scale integration of renewable energy, the traditional operation of the power system has been altered. The randomness and volatility of renewable energy bring significant uncertainty to grid planning and operation. Analyzing the grid's operational sequence scenarios in different years during the technical evaluation of planning schemes is time-consuming. Generally, several key scenarios are selected for analysis and verification. Traditional system analysis often involves analyzing some extreme scenarios; if the system can maintain safe and stable operation under extreme scenarios, then the system will be stable under all operating modes. Currently, the selection of extreme scenarios often relies on historical information, the experience and judgment of planners, and lacks corresponding theoretical support. Especially given the uncertainties brought about by the large-scale integration of renewable energy, it remains to be verified whether extreme grid operation scenarios will occur during periods of high line pressure or typical periods of high pressure in winter, low pressure in winter, high pressure in summer, and low pressure in summer. This presents a challenge to the selection of extreme scenarios. Therefore, it is necessary to re-analyze, mine, and screen power grid operation scenarios to identify high-risk extreme scenarios in system operation, improve the efficiency of power grid operation mode flow analysis, risk analysis and assessment, and safety and stability verification, and help planners and decision-makers to conduct rapid technical feasibility assessments of power grid planning schemes. Summary of the Invention

[0003] To overcome the shortcomings of the existing technology, this invention proposes a method for extracting extreme scenarios of power grid operation, comprising:

[0004] Acquire time-series data of wind power and load output per hour at each node in the power grid during a horizontal year as scenario data;

[0005] Clustering of all scenario data yields several typical wind power-load operation scenarios that take into account temporal correlation and geographical distribution characteristics;

[0006] Based on various typical wind power-load operation scenarios, the scenario data with the furthest cluster center corresponding to the scenario is taken as the extreme scenario corresponding to the typical wind power-load operation scenario.

[0007] Preferably, the clustering of all scenario data yields multiple typical wind power-load operation scenarios that consider temporal correlation and geographical distribution characteristics, including:

[0008] The wind power-load variables are weighted according to the degree of influence of the wind power-load variables in the scenario data on the preset indicators. The wind power-load variables include wind power and load at each node in the power grid.

[0009] The wind power-load variables are dimensionality reduced based on their weights to obtain dimensionality-reduced scenario data.

[0010] Clustering of the dimensionality-reduced scenario data based on weighted Euclidean distance identifies several typical wind power-load operation scenarios.

[0011] Preferably, the step of assigning weights to the wind power-load variables based on the degree of influence of the wind power-load variables in the scenario data on the preset indicators includes:

[0012] Based on the wind power-load variables in the scenario data, a variable sample is constructed, the dimension of which is the total number of wind power and load variables;

[0013] Obtain the target value of the preset indicator corresponding to each variable sample, combine each variable sample and the corresponding target value to form a training sample, and form a training sample set by all training samples.

[0014] Using the variable samples in the training samples as input values ​​and the target values ​​corresponding to the variable samples as output values, a machine learning algorithm is used to train the training sample set to obtain the weights of each wind power-load variable in the variable samples.

[0015] Preferably, the step of performing dimensionality reduction processing on the wind power-load variables based on the weights of the wind power-load variables to obtain dimensionality-reduced scenario data includes:

[0016] Compare the weights of each wind power-load variable with the preset weight thresholds;

[0017] Wind power-load variables with weights lower than the weight threshold are removed to obtain dimensionality-reduced scenario data.

[0018] Preferably, the clustering of the dimensionality-reduced scenario data based on weighted Euclidean distance to determine multiple typical wind power-load operation scenarios includes:

[0019] Based on the preset number of typical scenarios, multiple dimensionality reduction scenario data are randomly selected from all dimensionality reduction scenario data as initial cluster centers;

[0020] The cluster centers of each category are adjusted based on the weighted Euclidean distance from the data of each dimensionality reduction scenario to each cluster center until clustering is completed;

[0021] Various cluster centers are used as typical wind power-load operation scenarios.

[0022] Preferably, the weighted Euclidean distance is calculated as follows:

[0023]

[0024] In the formula p x For the x-th dimensionality reduction scenario data, p x For the y-th dimensionality reduction scenario data, d(p) x ,p y ) represents p x and p x The weighted Euclidean distance between them, w' h Let N' be the weight of the h-th dimension variable in the dimensionality reduction scenario data, and p be the dimension of the dimensionality reduction scenario data. xh Let p be the value of the h-th dimension variable in the x-th dimensionality reduction scenario data. yh Let h be the value of the h-th dimension variable in the y-th dimensionality reduction scenario data.

[0025] Preferably, the step of using the scenario data with the furthest cluster center corresponding to the scenario as the extreme scenario corresponding to the typical wind power-load operation scenario includes:

[0026] For each typical wind power-load operation scenario, based on the preset required quantity, from the classes corresponding to the typical wind power-load operation scenario, the scenario data that is farthest from the cluster center is selected as the extreme scenario corresponding to the typical wind power-load operation scenario, based on the principle of the farthest weighted Euclidean distance.

[0027] Based on the same inventive concept, this application also provides a power grid operation extreme scenario extraction system, including: a data acquisition module, a typical scenario module and an extreme scenario module;

[0028] The data acquisition module is used to acquire time-series data of wind power and load output per hour at each node in the power grid in a horizontal year as scenario data.

[0029] The typical scenario module clusters all scenario data to obtain multiple typical wind power-load operation scenarios that take into account temporal correlation and geographical distribution characteristics.

[0030] The extreme scenario module is used to select the scenario data with the furthest cluster center corresponding to each typical wind power-load operation scenario as the extreme scenario corresponding to the typical wind power-load operation scenario.

[0031] Preferably, the typical scenario module includes: a weighting unit, a dimensionality reduction unit, and a typical scenario unit;

[0032] The weighting unit is used to assign weights to the wind power-load variables based on the degree of influence of the wind power-load variables in the scenario data on the preset indicators. The wind power-load variables include wind power and load at each node in the power grid.

[0033] The dimensionality reduction unit is used to perform dimensionality reduction processing on the wind power-load variable based on the weights of the wind power-load variable to obtain dimensionality-reduced scenario data;

[0034] The typical scenario unit is used to cluster the dimensionality-reduced scenario data based on weighted Euclidean distance to determine multiple typical wind power-load operation scenarios.

[0035] Preferably, the weighting unit includes: a variable sample subunit, a training sample set subunit, and a weight subunit;

[0036] The variable sample subunit is used to construct a variable sample based on the wind power-load variable in the scenario data. The dimension of the variable sample is the total number of wind power and load variables.

[0037] The training sample set subunit is used to obtain the target value of the preset index corresponding to each variable sample, and to form a training sample by combining each variable sample and the corresponding target value, and to form a training sample set by combining all training samples.

[0038] The weighting subunit is used to take the variable samples in the training samples as input values, the target values ​​corresponding to the variable samples as output values, and train the training sample set using a machine learning algorithm to obtain the weights of each wind power-load variable in the variable samples.

[0039] Compared with the closest existing technology, the present invention has the following beneficial effects:

[0040] This invention provides a method and system for extracting extreme scenarios of power grid operation, comprising: acquiring time-series data of wind power and load output per hour at each node in the power grid in a horizontal year as scenario data; clustering all scenario data to obtain multiple typical wind power-load operation scenarios considering temporal correlation and geographical distribution characteristics; and, based on each typical wind power-load operation scenario, selecting the scenario data with the furthest cluster center as the extreme scenario corresponding to the typical wind power-load operation scenario. This invention, considering the temporal correspondence between wind power and load and the geographical distribution characteristics of wind power and load, uses a clustering method to extract the scenario point furthest from the cluster center as the extreme scenario. This method can help planners quickly assess the level of operational safety, while improving the rationality and scientific rigor of large-scale offline safety and stability analysis of the power grid.

[0041] The present invention further determines the variable weights based on the importance of the variables to the research problem, such as the preset indicators, and uses a weighted clustering method to extract the scene points that are farthest from the cluster center by weighted Euclidean distance as extreme scenes, thereby improving the rationality of scene extraction. Attached Figure Description

[0042] Figure 1 This invention provides a schematic flowchart of a method for extracting extreme scenarios of power grid operation.

[0043] Figure 2 This invention relates to a scenario set consisting of two wind power nodes and one node load, which constitutes a wind power-load configuration.

[0044] Figure 3 This is an illustration of an embodiment of an extreme scene extraction method provided by the present invention;

[0045] Figure 4 This invention relates to a schematic diagram of a wind power grid connection system topology;

[0046] Figure 5 The power angle difference of the critical unit under five extreme scenarios corresponding to typical scenario 1;

[0047] Figure 6 The power angle difference of the critical unit under five extreme scenarios corresponding to typical scenario 2;

[0048] Figure 7 The power angle difference of the critical unit under five extreme scenarios corresponding to typical scenario 3;

[0049] Figure 8 The power angle difference between critical units in extreme scenarios and scenarios with maximum and minimum load;

[0050] Figure 9 A schematic diagram of the basic structure of a power grid operation extreme scenario extraction system provided by the present invention;

[0051] Figure 10 This invention provides a detailed structural diagram of a power grid operation extreme scenario extraction system. Detailed Implementation

[0052] The purpose of this invention is to improve the rationality and efficiency of scenario analysis and help planners and decision-makers conduct rapid technical feasibility assessments of power grid planning schemes based on wind power and load information. This invention proposes a method for extracting extreme scenarios of power grid operation.

[0053] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0054] Example 1:

[0055] A flowchart illustrating a method for extracting extreme scenarios in power grid operation provided by this invention is shown below. Figure 1 As shown, it includes:

[0056] Step 1: Obtain time-series data of wind power and load output per hour at each node in the power grid during a horizontal year as scenario data;

[0057] Step 2: Cluster all scenario data to obtain multiple typical wind power-load operation scenarios that take into account temporal correlation and geographical distribution characteristics;

[0058] Step 3: Based on each typical wind power-load operation scenario, the scenario data with the furthest cluster center corresponding to the scenario is taken as the extreme scenario corresponding to the typical wind power-load operation scenario.

[0059] Clustering is a common method for obtaining power grid operation scenarios. It involves directly clustering raw data along the time dimension to obtain the desired operation scenarios, which helps maintain the temporal correlation of the data. The extraction of extreme operation scenarios can be based on clustering to obtain typical power grid scenarios. The selection of clustering variables is crucial; different clustering variables are used for different analytical problems, and different variable weights also have a significant impact on clustering. Scholars such as He Hailei et al. proposed using improved K-means clustering to generate key scenarios of system time-series operation for large-scale power grid planning transient security risk assessment. Scholars such as Ruidong Liu et al. proposed using the PSO-k-means clustering method to scan power grid time-series operation scenarios to identify key scenarios and conduct small-signal stability analysis. These studies have made valuable explorations on how to scientifically identify key scenarios at the level of power grid operation stability. However, research on how to directly extract extreme scenarios from time-series scenarios is still rare, especially regarding the selection of clustering variables and the analysis of the variables' impact on the research problem. Further in-depth research is needed on related technologies. In this embodiment, a weighted clustering method is used to extract the scene point that is farthest from the cluster center by weighted Euclidean distance as an extreme scene, which improves the rationality of scene extraction.

[0060] This invention provides a method for extracting extreme scenarios of power grid operation based on weighted clustering, specifically including the following:

[0061] Step 1: Obtain time-series data of wind power and load output per hour at each node in the power grid during a horizontal year as scenario data; that is:

[0062] Step 11: Construct a wind power-load joint operation scenario that considers temporal correlation and geographical distribution characteristics;

[0063] Step 11 specifically involves considering temporal correlation, which means constructing the model and scenario set to time the temporal correlation between wind power and load; and considering geographical distribution characteristics, which means using the output of each node wind farm and load for wind power output and load. Constructing a wind power-load joint operation scenario that considers temporal correlation and geographical distribution characteristics can utilize hourly and annual time-series data of the system node wind power and node load levels.

[0064] Preferably, the scenario construction method also includes directly processing the raw wind power-load data to maintain the correlation between the data, or constructing a wind power-load model based on the raw data to generate more sample data.

[0065] The time-series operational scenario variable matrix for the research period is constructed as follows:

[0066] P = [p1, p2, ..., p T ] T ∈R T×N (1)

[0067] p i =[p i1 ,p i2 ,···,p iN ]∈R 1×N (2)

[0068] In the formula, each row of matrix P represents a scene vector, with a total of T scenes. Typically, the time period is in hours, and a horizontal year can contain 8760 scenes. Scene p i It contains N variables, where N represents the node wind power and load dimensions, p ij This represents the wind power output or load of the j-th node in the i-th time period, which is also the wind power output or load of the j-th node in the i-th scenario.

[0069] Step 2 clusters all scenario data to obtain several typical wind power-load operation scenarios that consider temporal correlation and geographical distribution characteristics, including:

[0070] Step 21: Assign weights to the wind power-load variables in the scenario and construct a dimension-reduced matrix;

[0071] Step 21 specifically involves: weighting variables, which means assigning weights to variables based on their influence on the research question; and constructing a dimensionality reduction matrix, which means eliminating variables with weights below a threshold based on their weights. The research question can be further refined to pre-defined indicators, such as power generation or abandoned power.

[0072] Preferably, a machine learning algorithm is used to train a model based on samples to identify the degree of influence of different variables on the research question, and then assign weights to them. The method includes the following steps:

[0073] Based on N-dimensional wind power and load variables, sample data are selected to construct a variable sample set X = {x}. a |x a ∈R N}, where a is the sample number, x a For each sample x, the vector containing N-dimensional feature variables is... a Analysis yields x a The target value of the corresponding research question is used as the target value τ of the sample, and the sample data and the target value of the sample are combined to form the training set [X,τ].

[0074] For input variable X and output variable τ, the RLeifF algorithm is used to train the training set to obtain the weight and ranking of each variable.

[0075] A variable threshold is set, and variables with an impact level lower than the threshold, i.e., node wind power and load variables with a weight lower than the threshold, are directly removed, thereby constructing a dimension-reduced variable matrix with a dimension of N'.

[0076] Step 22: Use weighted clustering to obtain several typical wind power-load operation scenarios;

[0077] Step 22 specifically refers to using the idea of ​​weighted clustering to obtain several typical scenarios C by performing weighted clustering on the time-series scenario P in the time dimension. These typical scenarios can represent most scenarios related to the analysis problem.

[0078] Preferably, the commonly used weighted clustering method—K-means method—is used to cluster and reduce time-series scenarios to obtain typical wind power-load operation scenarios.

[0079] (1) For the aforementioned dimensionality reduction matrix dataset P = {p i |p i ∈R N' Let N' be the dimension of the data objects and T be the number of data objects in the dataset. K initial cluster centers c1, c2, ..., c3 are randomly selected. k k is typically the number of typical scenarios to be selected.

[0080] (2) Calculate the weighted Euclidean distance between each data object and the k cluster centers respectively, and assign the data object to the class to which the nearest cluster center belongs. The weighted Euclidean distance is as follows.

[0081] Any two data objects p x ,p y The weighted Euclidean distance is expressed as follows:

[0082]

[0083] Among them w' h Let N' be the weight of each dimension of the data object, and N' be the number of variables, corresponding to the number of variables after dimensionality reduction. xh Let h be the value of the h-th dimension variable in the x-th time period.

[0084] (3) Recalculate the cluster centers, as defined below:

[0085]

[0086] In the formula N j C1, C2, ..., C is the number of samples in class j. k Cluster centers c1, c2, ..., c k The data collection of the class to which it belongs.

[0087] (4) Calculate the distance between each cluster center and the previously calculated cluster center. If the distance is less than the set threshold,

[0088] Then the process ends; if the conditions are not met, repeat steps (2)-(4).

[0089] The resulting cluster centers represent typical wind power-load scenarios relevant to the research question.

[0090] Step 3, based on typical wind power-load operation scenarios, uses the scenario data with the furthest cluster center corresponding to each scenario as the extreme scenario corresponding to the typical wind power-load operation scenario, i.e.

[0091] Step 31: Select the edge points of each class as the extreme scenarios corresponding to the typical scenario.

[0092] Step 31 specifically refers to the extreme scenario, which is a high-risk and severe scenario in the time series scenario that is prone to system instability, power flow exceeding limits, wind curtailment, and load loss.

[0093] Selecting the edge points of each class as the extreme scenarios corresponding to the typical scenario means selecting the edge points that rank highest in weighted Euclidean distance from the typical scenario (i.e., the cluster center) as the extreme scenarios corresponding to that typical scenario. The method is as follows: based on the determined typical wind power-load scenarios, in each class, select the sample point farthest from the cluster center by Euclidean distance as the edge point. This edge point can be considered the corresponding extreme operating point of the system under the typical scenario, i.e., the extreme operating scenario, expressed as:

[0094]

[0095] In the formula, k is the number of clusters, and c i e is the cluster center i This represents the extreme scenario corresponding to the i-th typical scenario.

[0096] Furthermore, if n edge points are taken from each category, that is, each typical scenario corresponds to n extreme scenarios, then a total of n*k extreme operating scenarios will be formed.

[0097]

[0098] In the formula, m represents the m-th extreme scenario corresponding to each typical scenario, where m = 1, 2, ..., n. Let m be the extreme scenario at the m-th level corresponding to the i-th typical scenario.

[0099] Extreme scenarios are extracted using clustering methods. The number and coverage of extreme scenarios can include scenarios where wind power and load are relatively extreme, or they may include times when wind power or load is relatively average. That is, extreme scenarios are not always on the outer contour of the data points, but may also be inside the data points. It is related to the number of typical scenarios selected.

[0100] Example 2:

[0101] Specifically, taking transient power angle stability as an example, the implementation process and effects of the present invention are introduced.

[0102] A transient power angle stability analysis was conducted using a New England 10-unit 39-bus system to verify the effectiveness and rationality of the extracted extreme scenarios. The analysis used 8760 hours of uncurtailed wind power time-series data from two wind farms (W1 and W2) and one load (L1) throughout the year. The scenario set constructed from the 8760 hours of wind power and load data is shown below. Figure 2 As shown, the extreme scene extraction process is as follows: Figure 3 As shown, the wind power grid connection topology is as follows: Figure 4 As shown.

[0103] From 8760 wind power-load scenarios, 100 scenarios were randomly selected to perform transient power angle stability simulations according to the set fault modes. The transient stability target value was calculated and formed into a sample training set. The RReliefF algorithm was used to train the samples, identify the dominant variables affecting the transient power angle stability of the system, and assign weights.

[0104] The transient stability target value TSI is:

[0105]

[0106] In the formula δ max The maximum power angle difference between any two synchronous machines after a system failure is taken. The larger the TSI, the more stable the system.

[0107] Clustering was performed on 8760 scenarios based on variable weighting to obtain 3 typical scenarios. Five edge points were selected from each typical scenario as extreme scenarios, resulting in a total of 15 extreme scenarios.

[0108] Compare the power angle difference curves of critical units in typical scenarios and their corresponding extreme scenarios, by Figure 5 , Figure 6 and Figure 7 It can be seen that the initial swing amplitude of the power angle difference curve of the critical unit in the three typical scenarios is smaller than that of the corresponding extreme scenarios, that is, the transient power angle stability of the extracted extreme scenarios is weaker than that of the corresponding typical scenarios.

[0109] Figure 8 The power angle difference between the critical units in extreme scenarios and scenarios with maximum and minimum load was compared. Figure 8 It can be seen that the extracted extreme scenarios have a larger initial swing amplitude of the power angle difference than the critical unit in the scenarios with the maximum and minimum loads, that is, the transient stability is more severe. The comparison demonstrates the effectiveness of the extreme scenarios extracted by the method proposed in this invention.

[0110] Example 3:

[0111] Based on the same inventive concept, this application also provides a power grid operation extreme scenario extraction system, the basic structure of which is as follows: Figure 9 As shown, it includes: a data acquisition module, a typical scenario module, and an extreme scenario module;

[0112] The data acquisition module is used to acquire time-series data of wind power and load output per hour at each node in the power grid in a horizontal year as scenario data;

[0113] The typical scenario module clusters all scenario data to obtain multiple typical wind power-load operation scenarios that take into account temporal correlation and geographical distribution characteristics;

[0114] The extreme scenario module is used to identify the extreme scenarios corresponding to typical wind power-load operation scenarios based on the scenario data with the furthest cluster center corresponding to each scenario.

[0115] The detailed structure of the power grid operation extreme scenario extraction system is as follows: Figure 10 As shown.

[0116] The typical scenario module includes: weighting unit, dimensionality reduction unit, and typical scenario unit;

[0117] The weighting unit is used to assign weights to the wind power-load variables based on the degree of influence of the wind power-load variables on preset indicators in the scenario data. The wind power-load variables include wind power and load at each node in the power grid.

[0118] The dimensionality reduction unit is used to perform dimensionality reduction processing on the wind power-load variables based on their weights to obtain dimensionality-reduced scenario data.

[0119] Typical scenario units are used to cluster dimensionality-reduced scenario data based on weighted Euclidean distance to identify multiple typical wind power-load operation scenarios.

[0120] The weighting unit includes: variable sample subunit, training sample set subunit, and weight subunit;

[0121] The variable sample subunit is used to construct variable samples based on the wind power-load variables in the scenario data. The dimension of the variable sample is the total number of wind power and load variables.

[0122] The training sample set sub-unit is used to obtain the target value of the preset index corresponding to each variable sample, and to form a training sample by combining each variable sample and its corresponding target value, and to form a training sample set by combining all training samples.

[0123] The weighting subunit is used to take the variable samples in the training samples as input values ​​and the target values ​​of the corresponding variable samples as output values. The training sample set is trained using machine learning algorithms to obtain the weights of each wind power-load variable in the variable samples.

[0124] The dimension reduction unit includes a comparison subunit and a dimension reduction subunit;

[0125] The comparison sub-unit is used to compare the weights and preset weight thresholds of each wind power-load variable.

[0126] The dimensionality reduction sub-unit is used to remove wind power-load variables with weights below the weight threshold to obtain dimensionality-reduced scenario data.

[0127] The typical scenario unit includes: initialization subunit, clustering subunit, and typical scenario subunit;

[0128] The initialization sub-unit is used to randomly select multiple dimensionality reduction scenario data as initial cluster centers from all dimensionality reduction scenario data according to the preset number of typical scenarios.

[0129] The clustering subunit is used to adjust the cluster centers of each category based on the weighted Euclidean distance from the data of each dimensionality reduction scenario to each cluster center until clustering is completed;

[0130] Typical scenario sub-units are used to represent typical wind power-load operation scenarios with various cluster centers.

[0131] Specifically, the extreme scenario module is used to select the extreme scenarios corresponding to each typical wind power-load operation scenario, based on the preset required number of scenarios and the principle of selecting the scenario data that is farthest from the cluster center in terms of the number of data points corresponding to the typical wind power-load operation scenario.

Claims

1. A method for extracting extreme scenarios of power grid operation, characterized in that, include: Acquire time-series data of wind power and load output per hour at each node in the power grid during a horizontal year as scenario data; Clustering of all scenario data yields several typical wind power-load operation scenarios that take into account temporal correlation and geographical distribution characteristics; Based on each typical wind power-load operation scenario, the scenario data with the farthest cluster center corresponding to the scenario is taken as the extreme scenario corresponding to the typical wind power-load operation scenario. The clustering of all scenario data yields several typical wind power-load operation scenarios that consider temporal correlation and geographical distribution characteristics, including: The wind power-load variables are weighted according to the degree of influence of the wind power-load variables in the scenario data on the preset indicators. The wind power-load variables include wind power and load at each node in the power grid. The wind power-load variables are dimensionality reduced based on their weights to obtain dimensionality-reduced scenario data. Clustering of the dimensionality-reduced scenario data based on weighted Euclidean distance identifies several typical wind power-load operation scenarios; The step of assigning weights to the wind power-load variables based on the degree of influence of the wind power-load variables in the scenario data on the preset indicators includes: Based on the wind power-load variables in the scenario data, a variable sample is constructed, the dimension of which is the total number of wind power and load variables; Obtain the target value of the preset indicator corresponding to each variable sample, combine each variable sample and the corresponding target value to form a training sample, and form a training sample set by all training samples. Using the variable samples in the training samples as input values ​​and the target values ​​corresponding to the variable samples as output values, a machine learning algorithm is used to train the training sample set to obtain the weights of each wind power-load variable in the variable samples.

2. The method as described in claim 1, characterized in that, The step of performing dimensionality reduction processing on the wind power-load variables based on their weights to obtain dimensionality-reduced scenario data includes: Compare the weights of each wind power-load variable with the preset weight thresholds; Wind power-load variables with weights lower than the weight threshold are removed to obtain dimensionality-reduced scenario data.

3. The method as described in claim 1, characterized in that, The clustering of dimensionality-reduced scenario data based on weighted Euclidean distance identifies several typical wind power-load operation scenarios, including: Based on the preset number of typical scenarios, multiple dimensionality reduction scenario data are randomly selected from all dimensionality reduction scenario data as initial cluster centers; The cluster centers of each category are adjusted based on the weighted Euclidean distance from the data of each dimensionality reduction scenario to each cluster center until clustering is completed; Various cluster centers are used as typical wind power-load operation scenarios.

4. The method as described in claim 3, characterized in that, The formula for calculating the weighted Euclidean distance is as follows: In the formula For the first x Data from a dimensionality reduction scenario, For the first y Data from a dimensionality reduction scenario, express and The weighted Euclidean distance between them For dimensionality reduction scenario data h Weights of dimensional variables, To reduce the dimensionality of scenario data, For the first x The data for the first dimensionality reduction scenario h The value of the dimension variable, For the first y The data for the first dimensionality reduction scenario h The value of the dimension variable.

5. The method as described in claim 1, characterized in that, The step of using the scenario data with the furthest cluster center corresponding to the scenario as the extreme scenario corresponding to the typical wind power-load operation scenario includes: For each typical wind power-load operation scenario, based on the preset required quantity, from the classes corresponding to the typical wind power-load operation scenario, the scenario data that is farthest from the cluster center is selected as the extreme scenario corresponding to the typical wind power-load operation scenario, based on the principle of the farthest weighted Euclidean distance.

6. A system for extracting extreme scenarios of power grid operation, characterized in that, include: Data acquisition module, typical scenario module, and extreme scenario module; The data acquisition module is used to acquire time-series data of wind power and load output per hour at each node in the power grid in a horizontal year as scenario data. The typical scenario module clusters all scenario data to obtain multiple typical wind power-load operation scenarios that take into account temporal correlation and geographical distribution characteristics. The extreme scenario module is used to take the scenario data with the farthest cluster center corresponding to each typical wind power-load operation scenario as the extreme scenario corresponding to the typical wind power-load operation scenario. The typical scenario module includes: a weighting unit, a dimensionality reduction unit, and a typical scenario unit; The weighting unit is used to assign weights to the wind power-load variables based on the degree of influence of the wind power-load variables in the scenario data on the preset indicators. The wind power-load variables include wind power and load at each node in the power grid. The dimensionality reduction unit is used to perform dimensionality reduction processing on the wind power-load variable based on the weights of the wind power-load variable to obtain dimensionality-reduced scenario data; The typical scenario unit is used to cluster the dimensionality-reduced scenario data based on weighted Euclidean distance to determine multiple typical wind power-load operation scenarios. The weighting unit includes: a variable sample subunit, a training sample set subunit, and a weight subunit; The variable sample subunit is used to construct a variable sample based on the wind power-load variable in the scenario data. The dimension of the variable sample is the total number of wind power and load variables. The training sample set subunit is used to obtain the target value of the preset index corresponding to each variable sample, and to form a training sample by combining each variable sample and the corresponding target value, and to form a training sample set by combining all training samples. The weighting subunit is used to take the variable samples in the training samples as input values, the target values ​​corresponding to the variable samples as output values, and train the training sample set using a machine learning algorithm to obtain the weights of each wind power-load variable in the variable samples.

Images