CN114418789A - Power grid operation extreme scene extraction method and system - Google Patents
Power grid operation extreme scene extraction method and system Download PDFInfo
- Publication number
- CN114418789A CN114418789A CN202011078388.9A CN202011078388A CN114418789A CN 114418789 A CN114418789 A CN 114418789A CN 202011078388 A CN202011078388 A CN 202011078388A CN 114418789 A CN114418789 A CN 114418789A
- Authority
- CN
- China
- Prior art keywords
- scene
- wind power
- load
- variable
- typical
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000605 extraction Methods 0.000 title claims description 15
- 238000000034 method Methods 0.000 claims abstract description 34
- 238000012549 training Methods 0.000 claims description 42
- 238000012545 processing Methods 0.000 claims description 8
- 238000010801 machine learning Methods 0.000 claims description 7
- 238000004458 analytical method Methods 0.000 abstract description 13
- 238000011160 research Methods 0.000 description 9
- 230000001052 transient effect Effects 0.000 description 9
- 239000011159 matrix material Substances 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000003064 k means clustering Methods 0.000 description 2
- 238000012502 risk assessment Methods 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000005206 flow analysis Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Water Supply & Treatment (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Public Health (AREA)
- Medical Informatics (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides a method and a system for extracting extreme scenes of power grid operation, which comprise the following steps: acquiring time sequence data of wind power and load output of each node in a power grid in each hour in the horizontal year as scene data; clustering all scene data to obtain a plurality of typical wind power-load operation scenes considering time sequence relevance and geographic distribution characteristics; based on each typical wind power-load operation scene, taking the scene data with the farthest clustering center corresponding to the scene as the extreme scene corresponding to the typical wind power-load operation scene; according to the method, on the basis of considering the time sequence corresponding relation of wind power-load and the geographical distribution characteristics of the wind power-load, the scene point farthest from the clustering center is extracted by adopting a clustering method to serve as an extreme scene, the method can provide help for planners to quickly evaluate the operation safety level, and meanwhile, the rationality and the scientificity of the offline safety and stability analysis of the large-scale power grid are improved.
Description
Technical Field
The invention relates to the field of power grid planning operation analysis, in particular to a method for extracting an extreme scene of power grid operation based on weighted clustering.
Background
The power system is undergoing rapid transformation and subversion change driven by renewable energy source growth, the operation mode of the traditional power system is changed along with the access of large-scale renewable energy sources, and the characteristics of randomness, volatility and the like of the renewable energy sources bring great uncertainty to network source planning and operation. When the planning scheme is technically evaluated, it is time-consuming to analyze the horizontal year power grid operation time sequence scene, a plurality of key scenes are generally selected for analysis and check, in the traditional system analysis, part of extreme scenes are selected for analysis, and if the system can maintain safe and stable operation in the extreme scenes, the system can be stable in all operation modes. The current extreme scene is often selected according to historical information and experience and judgment of planners, and corresponding theoretical support is lacked. Particularly, in the face of uncertainty caused by access of large-scale renewable energy sources, whether an extreme power grid operation scene occurs at a typical time of large line pressure or large winter, small winter, large summer, small summer and the like in the past is still to be verified, and therefore a challenge is brought to selection of the extreme scene. Therefore, the operation scene of the power grid needs to be analyzed, excavated and screened again to identify the high-risk extreme scene of the system operation, so that the efficiency of the work such as power flow analysis, risk analysis and evaluation, safety and stability check and the like of the operation mode of the power grid is improved, and a planner and a decision maker are helped to perform rapid technical feasibility evaluation on the power grid planning scheme.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a method for extracting an extreme scene of power grid operation, which comprises the following steps:
acquiring time sequence data of wind power and load output of each node in a power grid in each hour in the horizontal year as scene data;
clustering all scene data to obtain a plurality of typical wind power-load operation scenes considering time sequence relevance and geographic distribution characteristics;
based on each typical wind power-load operation scene, taking the scene data with the farthest clustering center corresponding to the scene as the extreme scene corresponding to the typical wind power-load operation scene.
Preferably, the clustering all the scene data to obtain a plurality of typical wind power-load operation scenes considering the time sequence relevance and the geographic distribution characteristics includes:
weighting the wind power-load variable according to the influence degree of the wind power-load variable on a preset index in the scene data, wherein the wind power-load variable comprises wind power and load of each node in the power grid;
performing dimensionality reduction processing on the wind power-load variable based on the weight of the wind power-load variable to obtain dimensionality reduction scene data;
and clustering the dimensionality reduction scene data based on the weighted Euclidean distance, and determining a plurality of typical wind power-load operation scenes.
Preferably, the weighting the wind power-load variable according to the degree of influence of the wind power-load variable on a preset index in the scene data includes:
constructing a variable sample based on wind power-load variables in scene data, wherein the dimensionality of the variable sample is the total number of wind power and loads;
obtaining a target value of a preset index corresponding to each variable sample, forming a training sample by each variable sample and the corresponding target value, and forming a training sample set by all the training samples;
and taking the variable sample in the training sample as an input value, taking a target value corresponding to the variable sample as an output value, and training the training sample set by adopting a machine learning algorithm to obtain the weight of each wind power-load variable in the variable sample.
Preferably, the performing the dimensionality reduction processing on the wind power-load variable based on the weight of the wind power-load variable to obtain dimensionality reduction scene data includes:
respectively comparing the weight value of each wind power-load variable with a preset weight threshold value;
and eliminating the wind power-load variable with the weight lower than the weight threshold value to obtain the dimension reduction scene data.
Preferably, the clustering the dimensionality reduction scene data based on the weighted euclidean distance to determine a plurality of typical wind power-load operation scenes includes:
randomly selecting a plurality of dimension reduction scene data from all dimension reduction scene data as an initial clustering center according to the number of preset typical scenes;
adjusting the clustering centers of all types based on the weighted Euclidean distance from all dimension reduction scene data to all clustering centers until the clustering is finished;
various clustering centers are used as a typical wind power-load operation scene.
Preferably, the weighted euclidean distance is calculated as follows:
in the formula pxFor the xth dimensionality-reduced scene data, pxFor the y-th dimensionality reduction scene data, d (p)x,py) Represents pxAnd pxWeighted Euclidean distance of, w'hThe weight of h-dimension variable of dimension-reduced scene data, N' is the dimension of dimension-reduced scene data, pxhFor the value of h-dimension variable of x-dimension reduced scene data, pyhAnd the value of the h-dimension variable of the y-dimension reduction scene data is obtained.
Preferably, the taking the scene data with the farthest clustering center corresponding to the scene as the extreme scene corresponding to the typical wind power-load operation scene includes:
and respectively selecting the scene data with the number farthest from the clustering center from the classes corresponding to the typical wind power-load operation scenes according to the number selected according to the preset requirements and according to the weighted Euclidean distance as a principle, wherein the number of the scene data is selected as the extreme scenes corresponding to the typical wind power-load operation scenes.
Based on the same invention concept, the application also provides a power grid operation extreme scene extraction system, which comprises: the system comprises a data acquisition module, a typical scene module and an extreme scene module;
the data acquisition module is used for acquiring time sequence data of wind power and load output of each node in the power grid in each hour in the horizontal year as scene data;
the typical scene module is used for clustering all scene data to obtain a plurality of typical wind power-load operation scenes considering time sequence relevance and geographic distribution characteristics;
and the extreme scene module is used for taking the scene data with the farthest clustering center corresponding to each typical wind power-load operation scene as the extreme scene corresponding to the typical wind power-load operation scene based on each typical wind power-load operation scene.
Preferably, the typical scene module includes: the system comprises a weighting unit, a dimension reduction unit and a typical scene unit;
the weighting unit is used for weighting the wind power-load variable according to the influence degree of the wind power-load variable on a preset index in the scene data, wherein the wind power-load variable comprises wind power and load of each node in the power grid;
the dimension reduction unit is used for carrying out dimension reduction processing on the wind power-load variable based on the weight of the wind power-load variable to obtain dimension reduction scene data;
the typical scene unit is used for clustering the dimensionality reduction scene data based on the weighted Euclidean distance and determining a plurality of typical wind power-load operation scenes.
Preferably, the weighting unit includes: the system comprises a variable sample subunit, a training sample set subunit and a weight subunit;
the variable sample subunit is used for constructing a variable sample based on the wind power-load variable in the scene data, and the dimensionality of the variable sample is the total number of wind power and loads;
the training sample set subunit is configured to obtain a target value of a preset index corresponding to each variable sample, combine each variable sample and the corresponding target value into a training sample, and combine all the training samples into a training sample set;
and the weight subunit is used for training the training sample set by using a machine learning algorithm by taking the variable sample in the training sample as an input value and taking the target value corresponding to the variable sample as an output value to obtain the weight of each wind power-load variable in the variable sample.
Compared with the closest prior art, the invention has the following beneficial effects:
the invention provides a method and a system for extracting extreme scenes of power grid operation, which comprise the following steps: acquiring time sequence data of wind power and load output of each node in a power grid in each hour in the horizontal year as scene data; clustering all scene data to obtain a plurality of typical wind power-load operation scenes considering time sequence relevance and geographic distribution characteristics; based on each typical wind power-load operation scene, taking the scene data with the farthest clustering center corresponding to the scene as the extreme scene corresponding to the typical wind power-load operation scene; according to the method, on the basis of considering the time sequence corresponding relation of wind power-load and the geographical distribution characteristics of the wind power-load, the scene point farthest from the clustering center is extracted by adopting a clustering method to serve as an extreme scene, the method can provide help for planners to quickly evaluate the operation safety level, and meanwhile, the rationality and the scientificity of the offline safety and stability analysis of the large-scale power grid are improved.
The invention further determines the variable weight according to the importance degree of the variable to the research problem, such as the preset index, and extracts the scene point which is farthest from the weighted Euclidean distance of the clustering center as the extreme scene by adopting the weighted clustering method, thereby improving the rationality of scene extraction.
Drawings
FIG. 1 is a schematic flow chart of a method for extracting extreme scenes in power grid operation according to the present invention;
FIG. 2 is a wind power-load scenario set consisting of two node wind powers and a node load according to the present invention;
FIG. 3 is a diagram of an embodiment of an extreme scene extraction method provided by the present invention;
FIG. 4 is a schematic view of a wind power access system topology according to the present invention;
fig. 5 shows a critical unit power angle difference under 5 extreme scenes corresponding to the typical scene 1;
fig. 6 shows a critical unit power angle difference under 5 extreme scenes corresponding to the typical scene 2;
FIG. 7 is a critical unit power angle difference under 5 extreme scenes corresponding to a typical scene 3;
FIG. 8 shows the power angle difference of the critical unit in the extreme scene and the maximum and minimum load scene;
FIG. 9 is a schematic diagram of a basic structure of a power grid operation extreme scene extraction system provided by the present invention;
fig. 10 is a detailed structural schematic diagram of a power grid operation extreme scene extraction system provided by the invention.
Detailed Description
The invention aims to provide a power grid operation extreme scene extraction method aiming at improving the reasonability and efficiency of scene analysis and helping planners and decision makers to perform rapid technical feasibility evaluation on a power grid planning scheme according to wind power and load information.
The following describes embodiments of the present invention in further detail with reference to the accompanying drawings.
Example 1:
the schematic flow diagram of the method for extracting the extreme scene of the power grid operation provided by the invention is shown in fig. 1, and the method comprises the following steps:
step 1: acquiring time sequence data of wind power and load output of each node in a power grid in each hour in the horizontal year as scene data;
step 2: clustering all scene data to obtain a plurality of typical wind power-load operation scenes considering time sequence relevance and geographic distribution characteristics;
and step 3: based on each typical wind power-load operation scene, the scene data with the farthest clustering center corresponding to the scene is used as the extreme scene corresponding to the typical wind power-load operation scene.
Clustering is a common method for acquiring a power grid operation scene, namely, original data are directly clustered on a time dimension to obtain a required operation scene, so that the time sequence relevance of the data is kept, and the extraction of an extreme operation scene can be established on the basis of acquiring a typical scene of a power grid by clustering. The selection of the clustering variables is important for clustering, different analysis problems are oriented, the used clustering variables are different, and meanwhile, the clustering variables have different weights and have larger influence on clustering. The students congregate with the sea and the Lei propose to utilize the key scene of improving the time sequence operation of the K-means clustering generation system to carry out the transient safety risk assessment of the large-scale power grid planning. The student Ruidong Liu et al proposes to scan the power grid time sequence operation scene based on a PSO-k-means clustering method so as to identify key scenes and carry out small signal stability analysis. The researches are beneficial to scientifically identifying the key scenes of the operation stability level of the power grid, but at present, researches on how to directly extract extreme scenes from a time sequence scene are not common, particularly, the selection of clustering variables and the analysis of the influence of the variables on research problems are relatively deficient when a clustering method is adopted, and the related technologies need to be further deepened. In the embodiment, the weighted clustering method is adopted to extract the scene point with the farthest weighted Euclidean distance from the clustering center as the extreme scene, so that the rationality of scene extraction is improved.
The invention provides a method for extracting an extreme scene of power grid operation based on weighted clustering, which comprises the following steps:
step 11: constructing a wind power-load combined operation scene considering time sequence relevance and geographic distribution characteristics;
step 11 specifically includes that the consideration of time sequence relevance refers to the establishment of a model and a scene set time and the time sequence relevance between wind power and loads, and the consideration of geographic distribution characteristics refers to the adoption of wind power output and loads of each node wind power plant and loads. The method comprises the steps of constructing a wind power-load combined operation scene considering time sequence relevance and geographic distribution characteristics, and adopting time sequence data of system node wind power and node load every hour in horizontal year.
Preferably, the scene construction method further includes directly processing the wind power-load original data to maintain the relevance between the data, and the wind power-load model can also be constructed based on the original data to generate more sample data.
The time sequence operation scene variable matrix for the research period is constructed as follows:
P=[p1,p2,···,pT]T∈RT×N (1)
pi=[pi1,pi2,···,piN]∈R1×N (2)
wherein each row of the matrix P represents a scene vector with a total of T scenes, generally in hours, 8760 scenes can be taken in a horizontal year, and the scene PiIn the method, N variables are included, N represents the wind power and load dimension of a node, and pijAnd representing the wind power output or load of the jth node in the ith time period, namely the wind power output or load of the jth node in the ith scene.
step 21: weighting the wind power-load variable in the scene, and constructing a dimension reduction matrix;
specifically, the step 21 is to weight the variables according to the influence degree of the scene variables on the research problem, and the step of constructing the dimensionality reduction matrix is to remove the variables with the weight lower than the threshold value according to the weight of the variables. The problem to be researched can be specifically refined to a preset index, such as the generated energy or the electric energy abandonment and the like.
Preferably, a machine learning algorithm is adopted, the influence degree of different variables on the research problem is identified according to the sample training model, and then the weight is given. The method comprises the following steps:
based on N-dimensional wind power and load variables, sample data is selected, and a variable sample set X ═ X is constructeda|xa∈RNWhere a is the sample number, xaFor vectors containing N-dimensional feature variables, x for each set of samplesaAnalysis was carried out to obtain xaThe target value of the corresponding research question is used as the target value tau of the sample, and the sample data and the target value of the sample form a training set [ X, tau]。
And (3) for the input variable X and the output variable T, training the training set by adopting a machine learning algorithm RReliff algorithm to obtain the weight and the sequence of each variable.
And setting a variable threshold, and directly removing the variables with the influence degree lower than the threshold, namely the wind power and load variables of the nodes with the weight lower than the threshold, so as to construct a dimension-reduced variable matrix with the dimension of N'.
Step 22: acquiring a plurality of typical wind power-load operation scenes by adopting a weighted clustering method;
Preferably, a common weighted clustering method-K-means method is adopted to perform clustering reduction on the time sequence scene to obtain a typical wind power-load operation scene.
(1) For the aforementioned dimension reduction matrix dataset P ═ Pi|pi∈RN'I 1, 2., T }, where N' is the dimensionality of the data objects and T is the number of data objects in the data set, and k initial clustering centers c are randomly selected1,c2,…,ck. k is generally the number of typical scenes to be selected.
(2) And respectively calculating the weighted Euclidean distance between each data object and k clustering centers, and distributing the data objects to the class to which the nearest clustering center belongs, wherein the weighted Euclidean distance is described as follows.
Any two data objects px,pyThe weighted euclidean distance of (a) is expressed as follows:
w 'of'hFor the weight of each dimension variable of the data object, N' is the number of variables, corresponding to the number of variables after dimensionality reduction, pxhIs the value of the h-dimension variable for the x-th time period.
(3) Recalculating the cluster centers, defined as follows:
in the formula NjIs the number of samples in class j, C1,C2,...,CkAs the center of the cluster c1,c2,…,ckA data set of the class to which it belongs.
(4) Calculating the distance between each cluster center and the last calculated cluster center, if the distance is less than a set threshold value,
ending, if the condition is not met, repeating the steps (2) - (4).
The resulting cluster centers are typical wind-load scenarios associated with the research problem.
And 3, based on each typical wind power-load operation scene, taking the scene data with the farthest clustering center corresponding to the scene as the extreme scene corresponding to the typical wind power-load operation scene, namely
Step 31: and selecting the edge points of each class as extreme scenes corresponding to the typical scene.
Specifically, the extreme scene is a high-risk severe scene which easily causes system instability, current violation, wind and light abandonment and load shedding in a time sequence scene.
Selecting the edge point of each class as the extreme scene corresponding to the typical scene means that selecting the edge point which is ranked at the highest weighted Euclidean distance from the typical scene (i.e. the cluster center) as the extreme scene corresponding to the typical scene, and the method is that on the basis of determining the typical wind power-load scene, in each class, the sample point which is farthest from the Euclidean distance from the class center is taken as the edge point, and the edge point can be regarded as the corresponding system extreme operating point in the typical scene, i.e. the extreme operating scene, and is expressed as follows:
where k is the number of clusters, ciAs cluster center, eiThe terminal scene corresponding to the ith typical scene.
Further, if n layers of edge points are taken from each class, that is, each typical scene corresponds to n extreme scenes, n × k extreme operating scenes are formed.
And m represents the m-th layer extreme scene corresponding to each typical scene, and m is 1,2 …, n.The ith layer extreme scene corresponds to the ith typical scene.
The extreme scenes are extracted by a clustering method, the number and the coverage range of the extreme scenes can comprise scenes with respectively more extreme wind power and loads, and can also comprise moments with more average wind power or loads, namely, the extreme scenes are not always on the outer contour of the data points, but also can be inside the data points, and the extreme scenes are related to the selected number of typical scenes.
Example 2:
specifically, the implementation process and effect of the present invention are described by taking the transient power angle stability as an example.
And (3) carrying out transient power angle stability example analysis by adopting a new England 10 machine 39 node system, and verifying the validity and the rationality of the extracted extreme scene. Selecting unreleased wind time series data of two node wind power plants W1 and W2 and a node load L1 of 8760 hours all year round, wherein a scene set constructed by wind power and load 8760 hour data is shown in FIG. 2, an extreme scene extraction process is shown in FIG. 3, and a wind power access system topology is shown in FIG. 4.
Randomly selecting 100 scenes from 8760 node wind power-load scenes to perform transient power angle stability simulation according to a set fault mode, calculating a transient stability target value to form a sample training set, performing sample training by using a RReliefF algorithm, identifying a leading variable influencing the transient power angle stability of the system, and weighting.
The transient stability target value TSI is:
in the formula ofmaxAnd taking the maximum power angle difference of any two synchronous machines after the system fails, wherein the larger the TSI is, the more stable the system is.
8760 scenes are clustered based on variable weighting to obtain 3 types of typical scenes, 5 layers of edge points of each type of typical scenes are taken as extreme scenes, and 15 extreme scenes are shared.
Comparing the typical scene and the critical set power angle difference curve of the extreme scene corresponding to the typical scene, as can be seen from fig. 5, 6 and 7, the first swing amplitudes of the critical set power angle difference curve of the 3 types of typical scenes are all smaller than those of the corresponding extreme scene, that is, the transient power angle stability of the extracted extreme scene is weaker than that of the corresponding typical scene.
Fig. 8 compares the critical set power angle difference between the extreme scenario and the scenario with the maximum load and the minimum load, and it can be seen from fig. 8 that the extracted extreme scenario is larger in first swing amplitude value of the critical set power angle difference than the scenario with the maximum load and the minimum load, that is, the transient stability is worse, and the validity of the extreme scenario extracted by the method provided by the present invention is illustrated by comparison.
Example 3:
based on the same inventive concept, the present application further provides a power grid operation extreme scene extraction system, the basic structure of which is shown in fig. 9, and the system comprises: the system comprises a data acquisition module, a typical scene module and an extreme scene module;
the data acquisition module is used for acquiring time sequence data of wind power and load output of each node in the power grid in each hour in the horizontal year as scene data;
the typical scene module is used for clustering all scene data to obtain a plurality of typical wind power-load operation scenes considering time sequence relevance and geographic distribution characteristics;
and the extreme scene module is used for taking the scene data with the farthest clustering center corresponding to the scene as the extreme scene corresponding to the typical wind power-load operation scene based on each typical wind power-load operation scene.
The detailed structure of the power grid operation extreme scene extraction system is shown in fig. 10.
Wherein, the typical scene module includes: the system comprises a weighting unit, a dimension reduction unit and a typical scene unit;
the weighting unit is used for weighting the wind power-load variable according to the influence degree of the wind power-load variable on the preset index in the scene data, and the wind power-load variable comprises wind power and load of each node in the power grid;
the dimension reduction unit is used for carrying out dimension reduction processing on the wind power-load variable based on the weight of the wind power-load variable to obtain dimension reduction scene data;
and the typical scene unit is used for clustering the dimension reduction scene data based on the weighted Euclidean distance and determining a plurality of typical wind power-load operation scenes.
Wherein, the empowerment unit includes: the system comprises a variable sample subunit, a training sample set subunit and a weight subunit;
the variable sample subunit is used for constructing a variable sample based on the wind power-load variable in the scene data, and the dimensionality of the variable sample is the total number of wind power and loads;
the training sample set subunit is used for acquiring a target value of each variable sample corresponding to a preset index, forming a training sample by each variable sample and the corresponding target value, and forming a training sample set by all the training samples;
and the weight subunit is used for training the training sample set by using a machine learning algorithm by taking the variable samples in the training samples as input values and the target values of the corresponding variable samples as output values to obtain the weight of each wind power-load variable in the variable samples.
The dimension reduction unit comprises a comparison subunit and a dimension reduction subunit;
the comparison subunit is used for respectively comparing the weight of each wind power-load variable with a preset weight threshold;
and the dimension reduction subunit is used for eliminating the wind power-load variable with the weight lower than the weight threshold value to obtain dimension reduction scene data.
Wherein, typical scene unit includes: an initialization subunit, a clustering subunit and a typical scene subunit;
the initialization subunit is used for randomly selecting a plurality of dimension reduction scene data from all the dimension reduction scene data as initial clustering centers according to the preset typical scene number;
the clustering subunit is used for adjusting the clustering centers of all types based on the weighted Euclidean distance from all the dimensionality reduction scene data to all the clustering centers until the clustering is completed;
and the typical scene subunit is used for taking various types of clustering centers as typical wind power-load operation scenes.
The extreme scene module is specifically configured to select, for each typical wind power-load operation scene, a number of pieces of scene data farthest from a cluster center from a class corresponding to the typical wind power-load operation scene according to a preset number required to be selected, and according to a weighted euclidean distance principle, select the number of pieces of scene data farthest from the cluster center as the extreme scene corresponding to the typical wind power-load operation scene.
Claims (10)
1. A power grid operation extreme scene extraction method is characterized by comprising the following steps:
acquiring time sequence data of wind power and load output of each node in a power grid in each hour in the horizontal year as scene data;
clustering all scene data to obtain a plurality of typical wind power-load operation scenes considering time sequence relevance and geographic distribution characteristics;
based on each typical wind power-load operation scene, taking the scene data with the farthest clustering center corresponding to the scene as the extreme scene corresponding to the typical wind power-load operation scene.
2. The method of claim 1, wherein the clustering all scene data to obtain a plurality of typical wind power-load operation scenes considering time sequence relevance and geographic distribution characteristics comprises:
weighting the wind power-load variable according to the influence degree of the wind power-load variable on a preset index in the scene data, wherein the wind power-load variable comprises wind power and load of each node in the power grid;
performing dimensionality reduction processing on the wind power-load variable based on the weight of the wind power-load variable to obtain dimensionality reduction scene data;
and clustering the dimensionality reduction scene data based on the weighted Euclidean distance, and determining a plurality of typical wind power-load operation scenes.
3. The method of claim 2, wherein the weighting the wind power-load variable according to the degree of influence of the wind power-load variable on the preset index in the scene data comprises:
constructing a variable sample based on wind power-load variables in scene data, wherein the dimensionality of the variable sample is the total number of wind power and loads;
obtaining a target value of a preset index corresponding to each variable sample, forming a training sample by each variable sample and the corresponding target value, and forming a training sample set by all the training samples;
and taking the variable sample in the training sample as an input value, taking a target value corresponding to the variable sample as an output value, and training the training sample set by adopting a machine learning algorithm to obtain the weight of each wind power-load variable in the variable sample.
4. The method of claim 2, wherein the performing the dimension reduction processing on the wind power-load variable based on the weight of the wind power-load variable to obtain dimension reduction scene data comprises:
respectively comparing the weight value of each wind power-load variable with a preset weight threshold value;
and eliminating the wind power-load variable with the weight lower than the weight threshold value to obtain the dimension reduction scene data.
5. The method of claim 2, wherein the clustering the reduced-dimension scene data based on the weighted euclidean distance to determine a plurality of typical wind-load operation scenes comprises:
randomly selecting a plurality of dimension reduction scene data from all dimension reduction scene data as an initial clustering center according to the number of preset typical scenes;
adjusting the clustering centers of all types based on the weighted Euclidean distance from all dimension reduction scene data to all clustering centers until the clustering is finished;
various clustering centers are used as a typical wind power-load operation scene.
6. The method of claim 5, wherein the weighted euclidean distance is calculated as follows:
in the formula pxFor the xth dimensionality-reduced scene data, pxFor the y-th dimensionality reduction scene data, d (p)x,py) Represents pxAnd pxWeighted Euclidean distance of, w'hThe weight of h-dimension variable of dimension-reduced scene data, N' is the dimension of dimension-reduced scene data, pxhFor the value of h-dimension variable of x-dimension reduced scene data, pyhAnd the value of the h-dimension variable of the y-dimension reduction scene data is obtained.
7. The method of claim 2, wherein the taking the scene data with the farthest clustering center corresponding to the scene as the extreme scene corresponding to the typical wind power-load operation scene comprises:
and respectively selecting the scene data with the number farthest from the clustering center from the classes corresponding to the typical wind power-load operation scenes according to the number selected according to the preset requirements and according to the weighted Euclidean distance as a principle, wherein the number of the scene data is selected as the extreme scenes corresponding to the typical wind power-load operation scenes.
8. An electric network operation extreme scene extraction system, characterized by comprising: the system comprises a data acquisition module, a typical scene module and an extreme scene module;
the data acquisition module is used for acquiring time sequence data of wind power and load output of each node in the power grid in each hour in the horizontal year as scene data;
the typical scene module is used for clustering all scene data to obtain a plurality of typical wind power-load operation scenes considering time sequence relevance and geographic distribution characteristics;
and the extreme scene module is used for taking the scene data with the farthest clustering center corresponding to each typical wind power-load operation scene as the extreme scene corresponding to the typical wind power-load operation scene based on each typical wind power-load operation scene.
9. The system of claim 8, wherein the typical scene module comprises: the system comprises a weighting unit, a dimension reduction unit and a typical scene unit;
the weighting unit is used for weighting the wind power-load variable according to the influence degree of the wind power-load variable on a preset index in the scene data, wherein the wind power-load variable comprises wind power and load of each node in the power grid;
the dimension reduction unit is used for carrying out dimension reduction processing on the wind power-load variable based on the weight of the wind power-load variable to obtain dimension reduction scene data;
the typical scene unit is used for clustering the dimensionality reduction scene data based on the weighted Euclidean distance and determining a plurality of typical wind power-load operation scenes.
10. The system of claim 9, wherein the empowerment unit comprises: the system comprises a variable sample subunit, a training sample set subunit and a weight subunit;
the variable sample subunit is used for constructing a variable sample based on the wind power-load variable in the scene data, and the dimensionality of the variable sample is the total number of wind power and loads;
the training sample set subunit is configured to obtain a target value of a preset index corresponding to each variable sample, combine each variable sample and the corresponding target value into a training sample, and combine all the training samples into a training sample set;
and the weight subunit is used for training the training sample set by using a machine learning algorithm by taking the variable sample in the training sample as an input value and taking the target value corresponding to the variable sample as an output value to obtain the weight of each wind power-load variable in the variable sample.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011078388.9A CN114418789A (en) | 2020-10-10 | 2020-10-10 | Power grid operation extreme scene extraction method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011078388.9A CN114418789A (en) | 2020-10-10 | 2020-10-10 | Power grid operation extreme scene extraction method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114418789A true CN114418789A (en) | 2022-04-29 |
Family
ID=81260384
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011078388.9A Pending CN114418789A (en) | 2020-10-10 | 2020-10-10 | Power grid operation extreme scene extraction method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114418789A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116523351A (en) * | 2023-07-03 | 2023-08-01 | 广东电网有限责任公司湛江供电局 | Source-load combined typical scene set generation method, system and equipment |
-
2020
- 2020-10-10 CN CN202011078388.9A patent/CN114418789A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116523351A (en) * | 2023-07-03 | 2023-08-01 | 广东电网有限责任公司湛江供电局 | Source-load combined typical scene set generation method, system and equipment |
CN116523351B (en) * | 2023-07-03 | 2023-09-22 | 广东电网有限责任公司湛江供电局 | Source-load combined typical scene set generation method, system and equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Wu et al. | A data mining approach combining $ K $-means clustering with bagging neural network for short-term wind power forecasting | |
Sun et al. | Data-driven representative day selection for investment decisions: A cost-oriented approach | |
Ozkan et al. | A novel wind power forecast model: Statistical hybrid wind power forecast technique (SHWIP) | |
Xie et al. | Determination of the optimum capacity and type of wind turbine generators in a power system considering reliability and cost | |
CN109670213B (en) | Multi-machine equivalent modeling method and device for wind power plant | |
CN106650982A (en) | Depth learning power prediction method based on multi-point NWP | |
CN110009141B (en) | Climbing event prediction method and system based on SDAE feature extraction and SVM classification model | |
CN112149264A (en) | Active power distribution network planning method based on improved Harris eagle optimization algorithm | |
CN113283179A (en) | Short-term load prediction method based on multi-target LSTM integrated network | |
CN115329899A (en) | Clustering equivalent model construction method, system, equipment and storage medium | |
Gualtieri | Improving investigation of wind turbine optimal site matching through the self-organizing maps | |
Du et al. | Impact of wind power scenario reduction techniques on stochastic unit commitment | |
CN115563885A (en) | Energy conversion system for energy storage battery of wind power plant | |
CN115795329A (en) | Power utilization abnormal behavior analysis method and device based on big data grid | |
CN114418789A (en) | Power grid operation extreme scene extraction method and system | |
Karimtabar et al. | Analysis and predicting electricity energy consumption using data mining techniques—A case study IR Iran—Mazandaran province | |
Wang et al. | Solar power prediction using weather type pair patterns | |
CN113536694A (en) | Robust optimization operation method, system and device of comprehensive energy system and storage medium | |
CN112508278A (en) | Multi-connected system load prediction method based on evidence regression multi-model | |
Guo et al. | Identification of power system dynamic signature using hierarchical clustering | |
CN115021269B (en) | Two-stage optimal power flow solving method based on data driving | |
Riahinia et al. | A neural network-based model for wind farm output in probabilistic studies of power systems | |
CN109978258A (en) | Multi-data source method for forecasting photovoltaic power generation quantity and system based on machine learning | |
Jahan et al. | Intelligent system for power load forecasting in off-grid platform | |
Liu et al. | A machine learning approach for fast future grid small-signal stability scanning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |