Large-scale power system key scene generation method considering source load distribution difference

Abstract

The invention discloses a large-scale electric power system key scene generation method considering source load distribution difference, and belongs to the technical field of electric power system operation analysis and safety assessment, and the method comprises the steps: obtaining source load data, respectively calculating Kendall correlation coefficient grouping aggregation of a power time sequence curve between source load nodes, and forming an aggregation unit; aggregating the power sequences of all the source load nodes in each aggregation unit to obtain an aggregation power sequence corresponding to each aggregation unit, and clustering to obtain a corresponding scene class cluster; selecting matching features based on weather factors to carry out combination matching on each scene class cluster to construct an initial clustering scene set; and decomposing and restoring the aggregation power corresponding to each scene to a source load node to form a restored scene set, calculating a risk characteristic index of each scene, and performing re-clustering on the restored scene set to generate a key scene set. The problem that in the prior art, a key scene set with representativeness and risk discrimination cannot be screened from large-scale candidate scenes is solved.

Description

Technical Field

[0001] This invention relates to a method for generating critical scenarios of large-scale power systems that takes into account differences in source-load distribution, and belongs to the field of power system operation analysis and security assessment technology. Background Technology

[0002] Existing technologies for scenario generation and reduction to reflect the possible future operating states of power systems typically rely on historical data of uncertainties such as wind power, photovoltaic output, and load. Scenario sets are constructed through probabilistic modeling or data-driven analysis. Common methods for scenario construction include sampling and fitting based on statistical features, time series modeling, correlation modeling, and pattern recognition of source-load historical curves using typical day extraction, principal component analysis, and cluster analysis to obtain representative typical operating modes and their probabilities. For scenario reduction, methods such as distance-based clustering reduction, representative sample selection, or hierarchical screening strategies are often employed to select a small number of representative scenarios from a large pool of candidate scenarios. This reduces the computational burden of subsequent power flow calculations, scheduling optimization, and risk assessment, thereby supporting planning, operation, and scheduling research.

[0003] However, the aforementioned existing technologies still have the following shortcomings: First, many methods often adopt regional aggregation, network-wide equivalence, or simple superposition / averaging methods when modeling source loads and constructing scenarios, which makes it difficult to characterize the temporal differences and interrelationships of source loads in different spatial locations. This makes it difficult to fully reflect the impact of spatial distribution differences of source loads on power flow distribution, line load rate, voltage level, and other operating states. Second, existing scenario reduction methods mainly rely on probability distribution fitting or curve similarity, lacking explicit characterization and hierarchical identification mechanisms for system operation risks. The generated scenario sets have limited ability to distinguish differences in system safety margins and are difficult to effectively identify key operating scenarios that significantly affect safety margins. Finally, in large-scale power systems, if scenario construction and clustering are directly performed at the original wind and solar load node dimension, problems such as high dimensionality, large sample size, and high computational complexity are often encountered, which limits the efficiency of engineering applications. Summary of the Invention

[0004] The purpose of this invention is to provide a method for generating key scenarios of large-scale power systems that takes into account the differences in source and load distribution. By using aggregation dimensionality reduction, cluster matching, and risk-driven re-clustering, this method solves the problem that existing technologies cannot select a set of key scenarios that are both representative and risk-discriminative from a large number of candidate scenarios.

[0005] To solve the above-mentioned technical problems, the present invention is implemented using the following technical solution.

[0006] This invention provides a method for generating critical scenarios of large-scale power systems that takes into account differences in source-load distribution, including:

[0007] Obtain source payload data;

[0008] Based on the source load data, the Kendall correlation coefficient of the power time series curves between each source load node is calculated, and the source load nodes are grouped and aggregated according to the Kendall correlation coefficient to form an aggregation unit.

[0009] Aggregate the power sequences of all source-load nodes within each aggregation unit to obtain the aggregated power sequence corresponding to each aggregation unit.

[0010] The k-medoids algorithm is used to cluster the aggregation power sequences corresponding to each aggregation unit to obtain the corresponding scene clusters. Matching features are selected based on weather factors, and the Euclidean distance between the feature vectors of the cluster centers is calculated to combine and match each scene cluster to construct an initial set of clustered scenes.

[0011] The aggregate power corresponding to each scenario in the initial clustering scenario set is decomposed and restored to the source load node according to the preset allocation coefficient to form a restored scenario set;

[0012] The risk characteristic indicators of each scenario in the reconstructed scenario set are calculated, and the reconstructed scenario set is re-clustered using a Gaussian mixture model to generate a key scenario set.

[0013] Furthermore, the source-load data includes wind power data, photovoltaic power data, and load power data. The source-load nodes include wind farms, photovoltaic power plants, and load nodes. The aggregation units include wind power aggregation units, photovoltaic aggregation units, and load aggregation units. Before calculating the Kendall correlation coefficient of the power time-series curves between each source-load node, the source-load data is preprocessed.

[0014] Furthermore, based on the Kendall correlation coefficient, each source load node is grouped and aggregated to form an aggregate unit, including:

[0015] Construct the Kendall correlation coefficient matrix of the source load nodes based on the Kendall correlation coefficient;

[0016] Construct the Laplace matrix of the source-load node based on the Kendall correlation coefficient matrix of the source-load node;

[0017] Based on the Laplacian matrix of the source and load nodes, the eigenvalues ​​and eigenvectors of the Laplacian matrix are solved by feature extraction formula;

[0018] The feature extraction formula is expressed as follows:

[0019] ;

[0020] In the formula, The Laplace matrix of the source load nodes Corresponding eigenvalues The corresponding feature vector;

[0021] Arrange the eigenvalues ​​of the Laplacian matrix of the same source load node in ascending order, take the eigenvectors corresponding to the first preset number of eigenvalues ​​to construct an eigenvector matrix, and then perform normalization to construct a normalized eigenvector matrix.

[0022] Each row of the normalized feature vector matrix is ​​regarded as a sample vector. The k-means clustering algorithm is used to cluster all sample vectors, determine the center of each cluster, and divide all samples into different clusters to obtain the aggregation unit.

[0023] Each cluster corresponds to the same source load node.

[0024] Furthermore, the Kendall correlation coefficient matrix of the source load node is expressed as:

[0025] ;

[0026] In the formula, The Kendall correlation coefficient matrix of the source load nodes. The number of source load nodes. For the first The source load node and the first Kendall correlation coefficient between the power time-series curves of each source-load node. This indicates taking the maximum value. Indicates the order is The matrix.

[0027] Furthermore, the Laplace matrix of the source load node is expressed as:

[0028] ;

[0029] In the formula, Let Laplace's matrix be the source load node. for An identity matrix of order 1. for A column vector of dimension 1 with all elements equal to 1. As column vectors The elements of are a diagonal matrix constructed from the elements on the main diagonal.

[0030] Furthermore, the power sequences of all source-load nodes within each aggregation unit are aggregated to obtain the aggregated power sequence corresponding to each aggregation unit, including:

[0031] Based on the power sequence of all source-load nodes within each aggregation unit, the power values ​​of each source-load node within the same aggregation unit are summed at the corresponding sampling time to obtain the aggregation power sequence corresponding to each aggregation unit.

[0032] Furthermore, the k-medoids algorithm is used to cluster the aggregation power sequences corresponding to each aggregation unit to obtain the corresponding scene clusters; matching features are selected based on weather factors, and the scene clusters are combined and matched by calculating the Euclidean distance between the feature vectors of the cluster centers to construct an initial set of clustered scenes, including:

[0033] The aggregation power sequence corresponding to each aggregation unit is used as a clustering sample. The k-medoids clustering algorithm is used to cluster the clustering samples of the same aggregation unit to obtain a preset number of source load scene clusters and their corresponding cluster centers.

[0034] Based on the aggregated power sequence corresponding to each aggregated unit and the weather factor sequence within the corresponding sampling time, for each weather factor, the Kendall correlation coefficient between each weather factor and each aggregated power sequence is calculated and the average Kendall correlation coefficient is obtained by averaging. The weather factor with the largest average Kendall correlation coefficient is selected as the matching feature.

[0035] Using the time series of source load scene clusters and their corresponding cluster centers under the weather factors corresponding to the matching features as feature vectors, the Euclidean distance between feature vectors is calculated, and the source load scene clusters are combined and matched according to the principle of minimum distance to obtain the initial cluster scene set;

[0036] The weather factor sequence includes rainfall, wind speed, irradiance, temperature, and humidity.

[0037] Furthermore, the preset allocation coefficient is calculated using the allocation coefficient formula, wherein:

[0038] When the source-load node is a wind farm, the distribution coefficient formula is expressed as:

[0039] ;

[0040] In the formula, For the first Allocation coefficients for individual wind farms For the first A wind farm in Power sequence at time t, The sampling time interval for the wind farm. For the innermost Each wind power aggregation unit in The aggregate power sequence at time t, A collection of wind power aggregation units;

[0041] When the source-load node is a photovoltaic power station, the allocation coefficient formula is expressed as:

[0042] ;

[0043] In the formula, For the first Allocation coefficient of each photovoltaic power station For the first A photovoltaic power station in Power sequence at time t, The sampling time interval for the photovoltaic power station. For the first Each photovoltaic aggregation unit in The aggregate power sequence at time t, A collection of photovoltaic aggregation units;

[0044] When the source load node is also a load node, the distribution coefficient formula is expressed as:

[0045] ;

[0046] In the formula, For the first The allocation coefficient for each load node. For the first Each load node Power sequence at time t, The sampling time interval for the load node is [missing information]. For the first Each load node The aggregate power sequence at time t, This is a set of load aggregation units.

[0047] Furthermore, the risk characteristic indicators for each scenario in the reconstructed scenario set are calculated, including:

[0048] For each fault scenario corresponding to each scenario in the restored scenario set, calculate the power flow limit of each line and the voltage limit of each line node;

[0049] The formula for calculating the power flow exceedance limit is expressed as follows:

[0050] ;

[0051] In the formula, For the first The scene in Line under various fault scenarios The more extreme the trend, For the first Line under various fault scenarios Apparent power For the line The maximum apparent power that can be tolerated;

[0052] The formula for calculating the voltage over-limit is as follows:

[0053] ;

[0054] In the formula, For the first The scene in Under each fault scenario Line nodes The higher the voltage limit, For the first Line nodes under various fault scenarios voltage amplitude, and Line nodes The upper and lower voltage limits;

[0055] Based on the power flow limit exceedance and voltage limit exceedance of each node under each fault scenario, the scenario consequence assessment method is used to obtain the corresponding consequence index of each scenario under each fault scenario, and the risk characteristics of each scenario are calculated based on the consequence index.

[0056] The formula for calculating the consequence index is as follows:

[0057] ;

[0058] In the formula, For the first The scene in Consequence indicators under various failure scenarios For the collection of routes, For the set of line nodes; For the line The preset weighting coefficients, For line nodes The preset weighting coefficients;

[0059] The risk characteristics of each scenario are represented as follows:

[0060] ;

[0061] In the formula, For the first Risk characteristics of each scenario To repay the first The probability of occurrence of each failure scenario. A set of fault scenarios;

[0062] A risk feature matrix is ​​constructed based on the risk characteristics of each scenario in the restored scenario set, and a standardized risk feature matrix is ​​obtained by Z-score standardization, which serves as the risk feature index for each scenario.

[0063] Furthermore, a Gaussian mixture model is used to re-cluster the reconstructed scene set to generate a key scene set, including:

[0064] Based on the risk characteristic indicators of each scenario, a Gaussian mixture model is constructed to cluster each scenario in the set of restored scenarios. The Gaussian mixture model is composed of multiple Gaussian distributions. The target number of clusters in the Gaussian mixture model is determined by the Bayesian information criterion.

[0065] After determining the target number of clusters, the parameters of the Gaussian mixture model are initialized using the K-means clustering algorithm;

[0066] Based on the Gaussian mixture model with initialized parameters, repeat the following steps until the parameters converge:

[0067] Based on the parameters of the Gaussian mixture model, the posterior probability of the risk characteristic indicators corresponding to each scenario belonging to various Gaussian distributions is calculated.

[0068] The parameters of the Gaussian mixture model are updated based on the posterior probabilities of the various Gaussian distributions.

[0069] After the model parameters converge, based on the posterior probabilities of the risk feature indicators corresponding to each scenario belonging to various Gaussian distributions, each scenario in the restored scenario set is divided into the category corresponding to the Gaussian distribution with the largest posterior probability, generating a key scenario set.

[0070] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0071] 1. This invention forms aggregated units by grouping source and load nodes based on Kendall correlation coefficient, and constructs initial scenarios by combining k-medoids algorithm for clustering aggregated power sequences and weather factor feature matching. Then, the restored scenarios are further clustered using Gaussian mixture model to generate a set of key scenarios. This effectively achieves dimensionality reduction and compression of scenarios with different source and load distributions in large-scale power systems and accurate extraction of key scenarios, significantly improving scenario generation efficiency and the accuracy of system operation risk assessment. It solves the problem that existing technologies cannot select a set of key scenarios that are both representative and risk-discriminating from a large number of candidate scenarios.

[0072] 2. This invention calculates the Kendall correlation coefficient of the power time-series curves of each source-load node and constructs a correlation coefficient matrix. Combined with Laplace matrix feature extraction and k-means clustering, it realizes the grouping and aggregation of each source-load node into an aggregation unit based on the Kendall correlation coefficient. This effectively reduces the dimensionality complexity of directly processing large-scale source-load nodes and improves the computational efficiency of scene generation.

[0073] 3. This invention uses the k-medoids algorithm to cluster the aggregated power sequence, and uses the weather factor with the largest average Kendall correlation coefficient as the matching feature. It constructs an initial cluster scene set by calculating the Euclidean distance between the feature vectors of the cluster centers and performing combination matching. This enhances the correlation between the scene and the actual weather conditions and improves the accuracy of the physical meaning of the scene generation.

[0074] 4. This invention uses a Gaussian mixture model to re-cluster the restored scene set, and combines the Bayesian information criterion to determine the target number of clusters and initialize the K-means parameters, thereby achieving accurate extraction of key scenes. At the same time, through Z-score standardization of risk feature indicators, it significantly improves the quantitative accuracy and decision reliability of system operation risk assessment. Attached Figure Description

[0075] Figure 1 This is a schematic diagram of the overall process of a method for generating key scenarios of a large-scale power system that takes into account the differences in source and load distribution, provided by an embodiment of the present invention.

[0076] Figure 2 This is a detailed flowchart illustrating a method for generating key scenarios of a large-scale power system that takes into account differences in source and load distribution, provided by an embodiment of the present invention. Detailed Implementation

[0077] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0078] Example 1

[0079] like Figure 1 As shown in the figure, this embodiment introduces a method for generating key scenarios of a large-scale power system that takes into account the differences in source and load distribution, including:

[0080] Step 1: Obtain source payload data.

[0081] This embodiment acquires power data from three types of sources and loads—wind power, photovoltaic power, and load—as well as corresponding node data, providing complete raw input for subsequent scenario generation. This ensures that the data foundation covers the source and load characteristics of the entire system and supports accurate scenario modeling.

[0082] Step 2: Based on the source-load data, calculate the Kendall correlation coefficient of the power time-series curves between each source-load node, and group and aggregate each source-load node according to the Kendall correlation coefficient to form an aggregation unit.

[0083] This embodiment constructs a source-load node correlation matrix based on Kendall correlation coefficient and constructs a Laplacian matrix. It achieves "highly correlated node aggregation" through eigenvalue extraction and k-means clustering, effectively reducing the risk of dimensionality explosion when directly processing large-scale nodes and improving the physical rationality of group aggregation.

[0084] Step 3: Aggregate the power sequences of all source-load nodes within each aggregation unit to obtain the aggregated power sequence corresponding to each aggregation unit.

[0085] This embodiment sums the power sequences of each node within the same aggregation unit at different times to form an aggregated power sequence. While preserving the overall power characteristics of the unit, it compresses the data size, providing high-quality input after dimensionality reduction for subsequent clustering algorithms, thus balancing computational efficiency and information integrity.

[0086] Step 4: Use the k-medoids algorithm to cluster the aggregation power sequence corresponding to each aggregation unit to obtain the corresponding scene clusters; select matching features based on weather factors, and combine and match each scene cluster by calculating the Euclidean distance between the feature vectors of the cluster centers to construct the initial clustered scene set.

[0087] In this embodiment, k-medoids is used to cluster the aggregated sequences to generate scene clusters. Combined with the weather factor feature with the highest average Kendall correlation coefficient, Euclidean distance matching is used to achieve a strong weather-scene association combination, making the initial scene set more in line with the actual operating environment and enhancing the physical interpretability of the scene.

[0088] Step 5: Decompose and restore the aggregated power corresponding to each scene in the initial clustering scene set to the source load node according to the preset allocation coefficient to form a restored scene set.

[0089] In this embodiment, the aggregated power is precisely decomposed to each source-load node according to the source-load node type using a preset allocation coefficient, ensuring that the restored scenario is highly consistent with the original source-load node power characteristics, and providing real node-level data support for risk characteristic calculation.

[0090] Step Six: Calculate the risk characteristic indicators of each scenario in the reconstructed scenario set, and use a Gaussian mixture model to re-cluster the reconstructed scenario set to generate a key scenario set.

[0091] This embodiment calculates risk characteristics such as power flow exceedance, voltage exceedance, and consequence indicators, and constructs a risk feature matrix through Z-score standardization. It then combines this with Gaussian mixture model for re-clustering to achieve accurate extraction of high-risk and low-redundancy key scenarios, significantly improving the decision reliability of system operation risk assessment.

[0092] Example 2

[0093] Based on the same inventive concept as Embodiment 1, such as Figure 2 As shown in the figure, this embodiment introduces the implementation steps of a method for generating key scenarios of a large-scale power system that takes into account the differences in source and load distribution, including:

[0094] Step 1: Obtain source payload data.

[0095] In this embodiment, the source-load nodes include wind farms, photovoltaic power plants, and load nodes, and the aggregation unit includes a wind power aggregation unit, a photovoltaic aggregation unit, and a load aggregation unit. Before calculating the Kendall correlation coefficient of the power time-series curves between each source-load node, the source-load data is preprocessed.

[0096] Step 2: Based on the source load data, calculate the Kendall correlation coefficient of the power time series curves between each source load node, and group and aggregate each source load node according to the Kendall correlation coefficient to form an aggregation unit.

[0097] Step 2.1: Construct the Kendall correlation coefficient matrix of the source load nodes based on the Kendall correlation coefficient.

[0098] In this embodiment, the Kendall correlation coefficient matrix of the source load node is represented as:

[0099] ;

[0100] In the formula, The Kendall correlation coefficient matrix of the source load nodes. The number of source load nodes. For the first The source load node and the first Kendall correlation coefficient between the power time-series curves of each source-load node. This indicates taking the maximum value. Indicates the order is The matrix.

[0101] Step 2.2: Construct the Laplace matrix of the source-load nodes based on the Kendall correlation coefficient matrix of the source-load nodes.

[0102] In this embodiment, the Laplace matrix of the source load node is represented as:

[0103] ;

[0104] In the formula, Let Laplace's matrix be the source load node. for An identity matrix of order 1. for A column vector of dimension 1 with all elements equal to 1. As column vectors The elements of are a diagonal matrix constructed from the elements on the main diagonal.

[0105] Step 2.3: Based on the Laplacian matrix of the source load nodes, solve for the eigenvalues ​​and eigenvectors of the Laplacian matrix using the feature extraction formula.

[0106] In this embodiment, the feature extraction formula is expressed as:

[0107] ;

[0108] In the formula, The Laplace matrix of the source load nodes Corresponding eigenvalues The corresponding feature vector.

[0109] Step 2.4: Arrange the eigenvalues ​​of the Laplacian matrix of the same source load node in ascending order, take the eigenvectors corresponding to the first preset number of eigenvalues ​​to construct an eigenvector matrix, and perform normalization processing to construct a normalized eigenvector matrix.

[0110] Step 2.5: Treat each row of the normalized feature vector matrix as a sample vector, use the k-means clustering algorithm to cluster all sample vectors, determine the center of each cluster, and divide all samples into different clusters to obtain aggregation units; wherein, each cluster corresponds to the same source load node.

[0111] Step 3: Aggregate the power sequences of all source-load nodes within each aggregation unit to obtain the aggregated power sequence corresponding to each aggregation unit.

[0112] In this embodiment, based on the power sequence of all source-load nodes within each aggregation unit, the power values ​​of each source-load node within the same aggregation unit are summed at the corresponding sampling time to obtain the aggregation power sequence corresponding to each aggregation unit.

[0113] Step 4: Use the k-medoids algorithm to cluster the aggregation power sequence corresponding to each aggregation unit to obtain the corresponding scene clusters; select matching features based on weather factors, and combine and match each scene cluster by calculating the Euclidean distance between the feature vectors of the cluster centers to construct the initial clustered scene set.

[0114] Step 4.1: Take the aggregate power sequence corresponding to each aggregation unit as the clustering sample, and use the k-medoids clustering algorithm to cluster the clustering samples of the same aggregation unit to obtain a preset number of source-load scene clusters and their corresponding cluster centers.

[0115] Step 4.2: Based on the aggregated power sequence corresponding to each aggregated unit and the weather factor sequence within the corresponding sampling time, for each weather factor, calculate the Kendall correlation coefficient between each weather factor and each aggregated power sequence, and calculate the average Kendall correlation coefficient. Select the weather factor with the largest average Kendall correlation coefficient as the matching feature.

[0116] Step 4.3: Using the time series of source load scene clusters and their corresponding cluster centers under the weather factors corresponding to the matching features as feature vectors, calculate the Euclidean distance between the feature vectors, and combine and match the source load scene clusters according to the principle of minimum distance to obtain the initial cluster scene set.

[0117] In this embodiment, the weather factor sequence includes rainfall, wind speed, irradiance, temperature, and humidity.

[0118] Step 5: Decompose and restore the aggregate power corresponding to each scene in the initial clustering scene set to the source load node according to the preset allocation coefficient to form a restored scene set.

[0119] In this embodiment, the preset allocation coefficient is calculated using the allocation coefficient formula, wherein:

[0120] When the source-load node is a wind farm, the distribution coefficient formula is expressed as:

[0121] ;

[0122] In the formula, For the first Allocation coefficients for individual wind farms For the first A wind farm in Power sequence at time t, The sampling time interval for the wind farm. For the innermost Each wind power aggregation unit in The aggregate power sequence at time t, It is a collection of wind power aggregation units.

[0123] When the source-load node is a photovoltaic power station, the allocation coefficient formula is expressed as:

[0124] ;

[0125] In the formula, For the first Allocation coefficient of each photovoltaic power station For the first A photovoltaic power station in Power sequence at time t, The sampling time interval for the photovoltaic power station. For the first Each photovoltaic aggregation unit in The aggregate power sequence at time t, It is a collection of photovoltaic aggregation units.

[0126] When the source load node is also a load node, the distribution coefficient formula is expressed as:

[0127] ;

[0128] In the formula, For the first The allocation coefficient for each load node. For the first Each load node Power sequence at time t, The sampling time interval for the load node is [missing information]. For the first Each load node The aggregate power sequence at time t, This is a set of load aggregation units.

[0129] Step 6: Calculate the risk characteristic indicators of each scenario in the restored scenario set, and use a Gaussian mixture model to re-cluster the restored scenario set to generate a key scenario set.

[0130] Step 6.1: Calculate the risk characteristic indicators for each scenario in the set of restored scenarios.

[0131] Step 6.1.1 For each fault scenario corresponding to each scenario in the restored scenario set, calculate the power flow limit of each line and the voltage limit of each line node.

[0132] In this embodiment, the formula for calculating the power flow exceedance is expressed as follows:

[0133] ;

[0134] In the formula, For the first The scene in Line under various fault scenarios The more extreme the trend, For the first Line under various fault scenarios Apparent power For the line The limit of apparent power that can be tolerated.

[0135] In this embodiment, the formula for calculating the voltage over-limit is expressed as follows:

[0136] ;

[0137] In the formula, For the first The scene in Under each fault scenario Line nodes The higher the voltage limit, For the first Line nodes under various fault scenarios voltage amplitude, and Line nodes The upper and lower voltage limits.

[0138] Step 6.1.2: Based on the power flow limit exceedance of each line and the voltage limit exceedance of each node under each fault scenario, the scenario consequence assessment method is used to obtain the corresponding consequence index of each scenario under each fault scenario, and the risk characteristics of each scenario are calculated based on the consequence index.

[0139] In this embodiment, the formula for calculating the consequence index is expressed as follows:

[0140] ;

[0141] In the formula, For the first The scene in Consequence indicators under various failure scenarios For the collection of routes, For the set of line nodes; For the line The preset weighting coefficients, For line nodes The preset weighting coefficients.

[0142] In this embodiment, the risk characteristics of each scenario are represented as follows:

[0143] ;

[0144] In the formula, For the first Risk characteristics of each scenario To repay the first The probability of occurrence of each failure scenario. This is a set of fault scenarios.

[0145] Step 6.1.3: Construct a risk feature matrix based on the risk features of each scenario in the restored scenario set, and use Z-score standardization to obtain a standardized risk feature matrix, which serves as the risk feature index for each scenario.

[0146] Step 6.2: Use a Gaussian mixture model to re-cluster the restored scene set to generate a key scene set.

[0147] Step 6.2.1: Based on the risk characteristic indicators of each scenario, construct a Gaussian mixture model for clustering each scenario in the restored scenario set. The Gaussian mixture model consists of multiple Gaussian distributions, and the target number of clusters in the Gaussian mixture model is determined by the Bayesian information criterion.

[0148] Step 6.2.2: After determining the target number of clusters, the parameters of the Gaussian mixture model are initialized using the K-means clustering algorithm.

[0149] Step 6.2.3: Based on the Gaussian mixture model with initialized parameters, repeat the following steps until the parameters converge:

[0150] Based on the parameters of the Gaussian mixture model, the posterior probability of the risk characteristic indicators corresponding to each scenario belonging to various Gaussian distributions is calculated.

[0151] The parameters of the Gaussian mixture model are updated based on the posterior probabilities of the various Gaussian distributions.

[0152] Step 6.2.4: After the model parameters converge, based on the posterior probabilities of the risk feature indicators corresponding to each scenario belonging to various Gaussian distributions, the scenarios in the restored scenario set are divided into the categories corresponding to the Gaussian distribution with the largest posterior probability, generating a key scenario set.

[0153] Example 3

[0154] Based on the same inventive concept as other embodiments, this embodiment describes a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the methods of Embodiment 1 or 2 described above.

[0155] Example 4

[0156] Based on the same inventive concept as other embodiments, this embodiment introduces a computer program product, including computer instructions that, when executed by a processor, implement the steps of the methods described in Embodiment 1 or 2 above.

[0157] In summary, this invention forms aggregated units by grouping source and load nodes based on Kendall correlation coefficient, and constructs initial scenarios by combining k-medoids algorithm for clustering aggregated power sequences and weather factor feature matching. Then, it generates a key scenario set by re-clustering the restored scenarios using Gaussian mixture model. This effectively achieves dimensionality reduction and compression of scenarios with different source and load distributions in large-scale power systems and accurate extraction of key scenarios, significantly improving scenario generation efficiency and the accuracy of system operation risk assessment. It solves the problem that existing technologies cannot select a set of key scenarios that are both representative and risk-discriminating from a large number of candidate scenarios.

[0158] This invention calculates the Kendall correlation coefficient of the power time-series curves of each source-load node and constructs a correlation coefficient matrix. Combined with Laplace matrix feature extraction and k-means clustering, it achieves the grouping and aggregation of each source-load node into an aggregation unit based on the Kendall correlation coefficient. This effectively reduces the dimensionality complexity of directly processing large-scale source-load nodes and improves the computational efficiency of scene generation.

[0159] This invention uses the k-medoids algorithm to cluster aggregated power sequences, and uses the weather factor with the largest average Kendall correlation coefficient as the matching feature. It constructs an initial clustered scene set by calculating the Euclidean distance between the feature vectors of the cluster centers and performing combined matching. This enhances the correlation between the scene and the actual weather conditions and improves the accuracy of the physical meaning of the generated scene.

[0160] This invention employs a Gaussian mixture model to re-cluster the reconstructed scenario set, and combines the Bayesian information criterion to determine the target number of clusters and initialize the K-means parameters, thereby achieving accurate extraction of key scenarios. At the same time, through Z-score standardization of risk feature indicators, it significantly improves the quantitative accuracy and decision reliability of system operation risk assessment.

[0161] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0162] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0163] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0164] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0165] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for generating key scenarios of a large-scale power system considering differences in source-load distribution, characterized in that, include: Obtain source payload data; Based on the source load data, the Kendall correlation coefficient of the power time series curves between each source load node is calculated, and the source load nodes are grouped and aggregated according to the Kendall correlation coefficient to form an aggregation unit. Aggregate the power sequences of all source-load nodes within each aggregation unit to obtain the aggregated power sequence corresponding to each aggregation unit. The k-medoids algorithm is used to cluster the aggregation power sequences corresponding to each aggregation unit to obtain the corresponding scene clusters. Matching features are selected based on weather factors, and the Euclidean distance between the feature vectors of the cluster centers is calculated to combine and match each scene cluster to construct an initial set of clustered scenes. The aggregate power corresponding to each scenario in the initial clustering scenario set is decomposed and restored to the source load node according to the preset allocation coefficient to form a restored scenario set; The risk characteristic indicators of each scenario in the reconstructed scenario set are calculated, and the reconstructed scenario set is re-clustered using a Gaussian mixture model to generate a key scenario set.

2. The method for generating key scenarios of large-scale power systems considering source-load distribution differences according to claim 1, characterized in that, The source-load data includes wind power data, photovoltaic power data, and load power data. The source-load nodes include wind farms, photovoltaic power plants, and load nodes. The aggregation units include wind power aggregation units, photovoltaic aggregation units, and load aggregation units. Before calculating the Kendall correlation coefficient of the power time-series curves between each source-load node, the source-load data is preprocessed.

3. The method for generating key scenarios of large-scale power systems considering source-load distribution differences according to claim 2, characterized in that, Based on the Kendall correlation coefficient, each source load node is grouped and aggregated to form an aggregate unit, including: Construct the Kendall correlation coefficient matrix of the source load nodes based on the Kendall correlation coefficient; Construct the Laplace matrix of the source-load node based on the Kendall correlation coefficient matrix of the source-load node; Based on the Laplacian matrix of the source and load nodes, the eigenvalues ​​and eigenvectors of the Laplacian matrix are solved by feature extraction formula; The feature extraction formula is expressed as follows: ； In the formula, The Laplace matrix of the source load nodes Corresponding eigenvalues The corresponding feature vector; Arrange the eigenvalues ​​of the Laplacian matrix of the same source load node in ascending order, take the eigenvectors corresponding to the first preset number of eigenvalues ​​to construct an eigenvector matrix, and then perform normalization to construct a normalized eigenvector matrix. Each row of the normalized feature vector matrix is ​​regarded as a sample vector. The k-means clustering algorithm is used to cluster all sample vectors, determine the center of each cluster, and divide all samples into different clusters to obtain the aggregation unit. Each cluster corresponds to the same source load node.

4. The method for generating key scenarios of large-scale power systems considering source-load distribution differences according to claim 3, characterized in that, The Kendall correlation coefficient matrix of the source load nodes is represented as: ； In the formula, The Kendall correlation coefficient matrix of the source load nodes. The number of source load nodes. For the first The source load node and the first Kendall correlation coefficient between the power time-series curves of each source-load node. This indicates taking the maximum value. Indicates the order is The matrix.

5. The method for generating key scenarios of large-scale power systems considering source-load distribution differences according to claim 4, characterized in that, The Laplace matrix of the source load node is expressed as: ； In the formula, Let Laplace's matrix be the source load node. for An identity matrix of order 1. for A column vector of dimension 1 with all elements equal to 1. As column vectors The elements of are a diagonal matrix constructed from the elements on the main diagonal.

6. The method for generating key scenarios of large-scale power systems considering source-load distribution differences according to claim 5, characterized in that, Aggregate the power sequences of all source-load nodes within each aggregation unit to obtain the aggregated power sequence corresponding to each aggregation unit, including: Based on the power sequence of all source-load nodes within each aggregation unit, the power values ​​of each source-load node within the same aggregation unit are summed at the corresponding sampling time to obtain the aggregation power sequence corresponding to each aggregation unit.

7. The method for generating key scenarios of large-scale power systems considering source-load distribution differences according to claim 6, characterized in that, The k-medoids algorithm is used to cluster the aggregation power sequences corresponding to each aggregation unit to obtain the corresponding scene clusters. Matching features are selected based on weather factors, and the scene clusters are combined and matched by calculating the Euclidean distance between the feature vectors of the cluster centers to construct an initial set of clustered scenes, including: The aggregation power sequence corresponding to each aggregation unit is used as a clustering sample. The k-medoids clustering algorithm is used to cluster the clustering samples of the same aggregation unit to obtain a preset number of source load scene clusters and their corresponding cluster centers. Based on the aggregated power sequence corresponding to each aggregated unit and the weather factor sequence within the corresponding sampling time, for each weather factor, the Kendall correlation coefficient between each weather factor and each aggregated power sequence is calculated and the average Kendall correlation coefficient is obtained by averaging. The weather factor with the largest average Kendall correlation coefficient is selected as the matching feature. Using the time series of source load scene clusters and their corresponding cluster centers under the weather factors corresponding to the matching features as feature vectors, the Euclidean distance between feature vectors is calculated, and the source load scene clusters are combined and matched according to the principle of minimum distance to obtain the initial cluster scene set; The weather factor sequence includes rainfall, wind speed, irradiance, temperature, and humidity.

8. The method for generating key scenarios of large-scale power systems considering source-load distribution differences according to claim 2, characterized in that, The preset allocation coefficient is calculated using the allocation coefficient formula, where: When the source-load node is a wind farm, the distribution coefficient formula is expressed as: ； In the formula, For the first Allocation coefficients for individual wind farms For the first A wind farm in Power sequence at time t, The sampling time interval for the wind farm. For the innermost Each wind power aggregation unit in The aggregate power sequence at time t, A collection of wind power aggregation units; When the source-load node is a photovoltaic power station, the allocation coefficient formula is expressed as: ； In the formula, For the first Allocation coefficient of each photovoltaic power station For the first A photovoltaic power station in Power sequence at time t, The sampling time interval for the photovoltaic power station. For the first Each photovoltaic aggregation unit in The aggregate power sequence at time t, It is a collection of photovoltaic aggregation units; When the source load node is also a load node, the distribution coefficient formula is expressed as: ； In the formula, For the first The allocation coefficient for each load node. For the first Each load node Power sequence at time t, The sampling time interval for the load node is [missing information]. For the first Each load node The aggregate power sequence at time t, This is a set of load aggregation units.

9. The method for generating key scenarios of large-scale power systems considering source-load distribution differences according to claim 1, characterized in that, The risk characteristic indicators for each scenario in the set of reconstructed scenarios are calculated, including: For each fault scenario corresponding to each scenario in the restored scenario set, calculate the power flow limit of each line and the voltage limit of each line node; The formula for calculating the power flow exceedance limit is expressed as follows: ； In the formula, For the first The scene in Line under various fault scenarios The more extreme the trend, For the first Line under various fault scenarios Apparent power For the line The maximum apparent power that can be tolerated; The formula for calculating the voltage over-limit is as follows: ； In the formula, For the first The scene in Line nodes under various fault scenarios The higher the voltage limit, For the first Line nodes under various fault scenarios voltage amplitude, and Line nodes The upper and lower voltage limits; Based on the power flow limit exceedance and voltage limit exceedance of each node under each fault scenario, the scenario consequence assessment method is used to obtain the corresponding consequence index of each scenario under each fault scenario, and the risk characteristics of each scenario are calculated based on the consequence index. The formula for calculating the consequence index is as follows: ； In the formula, For the first The scene in Consequence indicators under various failure scenarios For the set of routes, For the set of line nodes; For the line The preset weighting coefficients, For line nodes The preset weighting coefficients; The risk characteristics of each scenario are represented as follows: ； In the formula, For the first Risk characteristics of each scenario To repay the first The probability of occurrence of each failure scenario. A set of fault scenarios; A risk feature matrix is ​​constructed based on the risk characteristics of each scenario in the restored scenario set, and a standardized risk feature matrix is ​​obtained by Z-score standardization, which serves as the risk feature index for each scenario.

10. The method for generating key scenarios of large-scale power systems considering source-load distribution differences according to claim 1, characterized in that, The reconstructed scene set is re-clustered using a Gaussian mixture model to generate a key scene set, including: Based on the risk characteristic indicators of each scenario, a Gaussian mixture model is constructed to cluster each scenario in the set of restored scenarios. The Gaussian mixture model is composed of multiple Gaussian distributions. The target number of clusters in the Gaussian mixture model is determined by the Bayesian information criterion. After determining the target number of clusters, the parameters of the Gaussian mixture model are initialized using the K-means clustering algorithm; Based on the Gaussian mixture model with initialized parameters, repeat the following steps until the parameters converge: Based on the parameters of the Gaussian mixture model, the posterior probability of the risk characteristic indicators corresponding to each scenario belonging to various Gaussian distributions is calculated. The parameters of the Gaussian mixture model are updated based on the posterior probabilities of the various Gaussian distributions. After the model parameters converge, based on the posterior probabilities of the risk feature indicators corresponding to each scenario belonging to various Gaussian distributions, each scenario in the restored scenario set is divided into the category corresponding to the Gaussian distribution with the largest posterior probability, generating a key scenario set.

Images