CN-122022253-A - Large-scale power system key scene generation method considering source load distribution difference

CN122022253ACN 122022253 ACN122022253 ACN 122022253ACN-122022253-A

Abstract

The invention discloses a large-scale power system key scene generation method considering source load distribution differences, which belongs to the technical field of power system operation analysis and safety evaluation and comprises the steps of obtaining source load data, respectively calculating Kendel correlation coefficient groups of power time sequence curves among source load nodes to form aggregation units, aggregating power sequences of all source load nodes in the aggregation units to obtain aggregation power sequences corresponding to the aggregation units, clustering to obtain corresponding scene class clusters, selecting matching features based on weather factors, combining and matching the scene class clusters to construct an initial clustering scene set, decomposing and restoring aggregation power corresponding to each scene to the source load nodes to form a reduction scene set to calculate risk feature indexes of each scene, and re-clustering the reduction scene set to generate a key scene set. The method solves the problem that the prior art can not screen and obtain the key scene set with both representativeness and risk distinction from the large-scale candidate scenes.

Inventors

SUN RAN
LIU HUICHAO
GUO HONGZUO
YAO FENG
CUI ZHAOHUI
WANG JIANBO
YANG SHA
WU MENGJIAO
ZHANG LIN
YAO YULI

Assignees

国网河南省电力公司
国网河南省电力公司安阳供电公司

Dates

Publication Date: 20260512
Application Date: 20251225

Claims (10)

1. The key scene generation method of the large-scale power system considering the source load distribution difference is characterized by comprising the following steps of: acquiring source load data; Based on the source load data, respectively calculating Kendell correlation coefficients of power time sequence curves among all source load nodes, and grouping and aggregating all the source load nodes according to the Kendell correlation coefficients to form an aggregation unit; the power sequences of all source load nodes in each aggregation unit are aggregated to obtain an aggregation power sequence corresponding to each aggregation unit; Selecting matching features based on weather factors, and carrying out combination matching on the scene clusters by calculating Euclidean distance between feature vectors of clustering centers to construct an initial clustering scene set; Decomposing and restoring the aggregate power corresponding to each scene in the initial clustering scene set to a source load node according to a preset distribution coefficient to form a restored scene set; and calculating risk characteristic indexes of each scene by calculating the restored scene set, and re-clustering the restored scene set by adopting a Gaussian mixture model to generate a key scene set.
2. The method for generating a key scene of a large-scale power system according to claim 1, wherein the source load data comprises wind power data, photovoltaic power data and load power data, the source load nodes comprise wind power stations, photovoltaic power stations and load nodes, the aggregation unit comprises a wind power aggregation unit, a photovoltaic aggregation unit and a load aggregation unit, and preprocessing is further included before calculating the kendel correlation coefficient of a power time sequence curve between the source load nodes.
3. The method for generating a key scene of a large-scale power system according to claim 2, wherein grouping and aggregating each source load node according to a kendel correlation coefficient to form an aggregation unit comprises: constructing a Kendell correlation coefficient matrix of the source load node according to the Kendell correlation coefficient; constructing a Laplacian matrix of the source load node based on the Kendell correlation coefficient matrix of the source load node; Solving the eigenvalue and eigenvector of the Laplace matrix by a characteristic extraction formula based on the Laplace matrix of the source load node; the feature extraction formula is expressed as: ; In the formula, Laplacian matrix as source load node Corresponding characteristic value Corresponding feature vectors; Arranging all the characteristic values of the Laplace matrix of the same source load node from small to large, taking characteristic vectors corresponding to the preset number of characteristic values to construct a characteristic vector matrix, and carrying out normalization processing to construct a normalized characteristic vector matrix; Taking each row of the normalized feature vector matrix as a sample vector, clustering all sample vectors by using a k-means clustering algorithm, determining the centers of various clusters, and dividing all samples into different clusters to obtain an aggregation unit; wherein, the same source load node corresponding to each class cluster.
4. The method for generating a key scene of a large-scale power system according to claim 3, wherein the kendel correlation coefficient matrix of the source load node is expressed as: ; In the formula, Is a matrix of Kendell correlation coefficients of source load nodes, As the number of the source load nodes, Is the first Source load node and the first The kendel correlation between the power timing curves of the individual source load nodes, Indicating that the maximum value is taken, Representing the order as Is a matrix of (a) in the matrix.
5. The method for generating a key scene of a large-scale power system according to claim 4, wherein the laplace matrix of the source load nodes is expressed as: ; In the formula, Is the laplace matrix of the source load nodes, Is that The order-unit matrix is used for the data processing, Is that Column vectors with dimensions and elements of all 1, In column vectors Is a diagonal matrix constructed of principal diagonal elements.
6. The method for generating a key scene of a large-scale power system according to claim 5, wherein aggregating power sequences of all source-load nodes in each aggregation unit to obtain an aggregate power sequence corresponding to each aggregation unit comprises: And summing the power values of all source load nodes in the same aggregation unit at corresponding sampling moments based on the power sequences of all source load nodes in each aggregation unit, so as to obtain an aggregation power sequence corresponding to each aggregation unit.
7. The method for generating the key scene of the large-scale power system considering the source load distribution difference as claimed in claim 6, wherein the k-means algorithm is adopted to cluster the aggregation power sequences corresponding to the aggregation units respectively to obtain corresponding scene clusters, the matching characteristics are selected based on weather factors, the combination matching is carried out on the scene clusters by calculating Euclidean distance between characteristic vectors of the clustering center, and an initial clustering scene set is constructed, and the method comprises the following steps: Respectively taking the aggregation power sequences corresponding to the aggregation units as clustering samples, and clustering the clustering samples of the same aggregation unit by adopting a k-means clustering algorithm to obtain a preset number of source load scene clusters and corresponding clustering centers thereof; based on the aggregate power sequences corresponding to the aggregate units and weather factor sequences in the corresponding sampling moments, calculating Kendell correlation coefficients between each weather factor and each aggregate power sequence according to each weather factor, averaging to obtain average Kendell correlation coefficients, and selecting the weather factor with the largest average Kendell correlation coefficient as a matching feature; Taking the time sequence of the source load scene clusters and the corresponding clustering centers under the weather factors corresponding to the matching features as feature vectors, calculating Euclidean distance between the feature vectors, and carrying out combined matching on the source load scene clusters according to the distance minimum principle to obtain an initial clustering scene set; The weather factor sequence includes a rainfall factor, a wind speed factor, an irradiance factor, a temperature factor, and a humidity factor.
8. The method for generating a key scene of a large-scale power system according to claim 2, wherein the preset distribution coefficient is calculated by a distribution coefficient formula, and wherein: When the source load node is a wind farm, the distribution coefficient formula is expressed as: ; In the formula, Is the first The distribution coefficients of the individual wind farms, Is the first In a wind farm The power sequence of the moment in time, For the sampling time intervals of the wind farm, Is the inner first The wind power aggregation units are arranged in An aggregate power sequence of time of day, The method comprises the steps of collecting wind power aggregation units; When the source load node is a photovoltaic power station, the distribution coefficient formula is expressed as: ; In the formula, Is the first The distribution coefficient of the individual photovoltaic power plants, Is the first The photovoltaic power station is arranged at The power sequence of the moment in time, For the sampling time intervals of the photovoltaic power plant, Is the first The photovoltaic polymerization units are arranged in An aggregate power sequence of time of day, Is a photovoltaic aggregation unit set; When the source load node is a load node, the distribution coefficient formula is: ; In the formula, Is the first The distribution coefficients of the individual load nodes are, Is the first At each load node The power sequence of the moment in time, For the sampling time interval of the load node, Is the first At each load node An aggregate power sequence of time of day, Is a set of load aggregation units.
9. The method for generating key scenes of a large-scale power system according to claim 1, wherein calculating the set of recovery scenes to calculate risk characteristic indexes of each scene comprises: Aiming at each fault scene corresponding to each scene in the restoration scene set, calculating the limit of the tide of each line to the limit of the voltage of each line node; The calculation formula of the load flow limit is expressed as follows: ; In the formula, Is the first The first scene is Circuit under fault scene Is more limited by the flow of water, Is the first Circuit under fault scene Is used for the control of the apparent power of (a), Is a circuit Limit values for the apparent power that can be tolerated; The calculation formula of the voltage threshold is expressed as follows: ; In the formula, Is the first The first scene is Line node under individual fault scene The more limited the voltage of (c) is, Is the first Line node under individual fault scene Is used for the voltage amplitude of (a), And (3) with Respectively, line nodes A voltage upper limit value and a voltage lower limit value of (a); Based on the trend limit of each line of each scene under each fault scene and the voltage limit of each node, obtaining corresponding result indexes of each scene under each fault scene by adopting a scene result evaluation method, and calculating risk characteristics of each scene according to the result indexes; The calculation formula of the result index is expressed as follows: ; In the formula, Is the first The first scene is The outcome indicators in the individual fault scenarios, For the collection of lines, Is a line node set; Is a circuit Is used for the control of the dynamic balance of the vehicle, Is a line node Is set according to the preset weight coefficient of the model; The risk characteristics of each scene are expressed as follows: ; In the formula, Is the first The risk characteristics of the individual scenes are determined, To be still the first The probability of occurrence of the individual fault scenarios, A fault scene set; And constructing a risk feature matrix based on the risk features of each scene in the restored scene set, and adopting Z-score standardization processing to obtain a standardized risk feature matrix serving as a risk feature index of each scene.
10. The method for generating a key scene of a large-scale power system according to claim 1, wherein the generating a key scene set by reclustering the restored scene set by using a gaussian mixture model comprises: Based on risk characteristic indexes of each scene, constructing a Gaussian mixture model for clustering each scene in a restored scene set, wherein the Gaussian mixture model consists of a plurality of Gaussian distributions, and determining the target clustering number of the Gaussian mixture model by adopting a Bayesian information criterion; after the target cluster number is determined, initializing parameters of the Gaussian mixture model by adopting a K-means clustering algorithm; based on the Gaussian mixture model after initializing the parameters, repeating the following steps until the parameters are converged: based on parameters of the Gaussian mixture model, calculating posterior probability of risk feature indexes corresponding to each scene belonging to various Gaussian distributions; Updating parameters of the Gaussian mixture model based on posterior probability of the various Gaussian distributions; after the model parameters are converged, based on posterior probabilities that risk characteristic indexes corresponding to all scenes belong to all kinds of Gaussian distributions, all scenes in the restored scene set are divided into categories corresponding to the Gaussian distributions with the maximum posterior probabilities, and a key scene set is generated.

Description

Large-scale power system key scene generation method considering source load distribution difference Technical Field The invention relates to a large-scale power system key scene generation method considering source load distribution differences, and belongs to the technical field of power system operation analysis and safety evaluation. Background The existing scene generation and reduction technology for reflecting the future possible running states of the power system is generally based on historical data of uncertainty factors such as wind power, photovoltaic output and load, and a scene set is constructed through probability modeling or data-driven analysis. In the aspect of scene construction, a common method comprises sampling and fitting based on statistical characteristics, time sequence modeling, correlation modeling, mode recognition on a source load history curve through means of typical day extraction, principal component analysis, cluster analysis and the like to obtain a representative typical operation mode and occurrence probability thereof, and in the aspect of scene reduction, a cluster reduction method based on distance measurement, a representative sample selection method or a hierarchical screening strategy is often adopted to select a small number of representative scenes from large-scale candidate scenes so as to reduce calculation burden of subsequent power flow calculation, scheduling optimization and risk assessment, thereby supporting planning, operation and scheduling research. However, the prior art still has the following defects that firstly, a plurality of methods often adopt a processing mode of regional summarization, full-network equivalence or simple superposition/average when a source load modeling and a scene are constructed, source load time sequence differences and mutual correlation characteristics of different spatial positions are difficult to describe, so that the influence of the source load space distribution differences on running states such as power flow distribution, line load rate, voltage level and the like is difficult to fully embody, secondly, the prior scene reduction mainly uses probability distribution fitting or curve similarity as a main basis, an explicit description and layering recognition mechanism for the running risk of a system is lacking, the generated scene set has limited capability in distinguishing the safety margin differences of the system, key running scenes with obvious influence on the safety margin are difficult to effectively recognize, and finally, if the scene construction and clustering are directly carried out on the original wind-solar load node dimension in a large-scale power system, the problems of high dimension, large sample scale, high calculation complexity and the like are often faced, and engineering application efficiency is limited. Disclosure of Invention The invention aims to provide a key scene generation method of a large-scale power system, which takes account of source-load distribution differences, and solves the problem that a key scene set with both representativeness and risk differentiation can not be obtained from large-scale candidate scenes through aggregation dimension reduction, cluster matching and risk driving reclustering in the prior art. In order to solve the technical problems, the invention is realized by adopting the following technical scheme. The invention provides a method for generating a key scene of a large-scale power system considering source load distribution difference, which comprises the following steps: acquiring source load data; Based on the source load data, respectively calculating Kendell correlation coefficients of power time sequence curves among all source load nodes, and grouping and aggregating all the source load nodes according to the Kendell correlation coefficients to form an aggregation unit; the power sequences of all source load nodes in each aggregation unit are aggregated to obtain an aggregation power sequence corresponding to each aggregation unit; Selecting matching features based on weather factors, and carrying out combination matching on the scene clusters by calculating Euclidean distance between feature vectors of clustering centers to construct an initial clustering scene set; Decomposing and restoring the aggregate power corresponding to each scene in the initial clustering scene set to a source load node according to a preset distribution coefficient to form a restored scene set; and calculating risk characteristic indexes of each scene by calculating the restored scene set, and re-clustering the restored scene set by adopting a Gaussian mixture model to generate a key scene set. Further, the source load data comprises wind power data, photovoltaic power data and load power data, the source load nodes comprise wind power stations, photovoltaic power stations and load nodes, the aggregation unit comprises a wind power aggregation unit, a photovolta