CN115759626A - Multi-resource dynamic clustering method and system for improving K-means + - Google Patents
Multi-resource dynamic clustering method and system for improving K-means + Download PDFInfo
- Publication number
- CN115759626A CN115759626A CN202211438073.XA CN202211438073A CN115759626A CN 115759626 A CN115759626 A CN 115759626A CN 202211438073 A CN202211438073 A CN 202211438073A CN 115759626 A CN115759626 A CN 115759626A
- Authority
- CN
- China
- Prior art keywords
- index
- clustering
- weight
- resource
- scheduling
- 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
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention relates to a data processing technology, in particular to a multi-resource dynamic clustering method and a multi-resource dynamic clustering system for improving K-means + +. Calculating clustering index weight by combining subjective and objective factors, and dynamically adjusting the weight based on real-time power grid scheduling requirements; and realizing multi-resource dynamic clustering by adopting a weight-improved K-means + + algorithm to obtain a dynamic clustering model of the controllable resource cluster so as to support the process that distributed resources participate in ADN (adaptive data network) optimization scheduling. According to the method, the real-time dynamic clustering model is built, the dimension of the small-capacity and large-quantity regulation and control resources is reduced into the large-capacity and small-quantity resource clusters, support is provided for high-proportion resources to participate in regulation and control, and compared with the traditional method, the method has the comprehensive advantages of being economical, reliable and easy to schedule in the aspect of meeting real-time regulation and control requirements.
Description
Technical Field
The invention belongs to the technical field of power resource optimization scheduling, and particularly relates to a K-means + + improved multi-resource dynamic clustering method and system.
Background
In recent years, distributed resources such as medium and small roof photovoltaic, small-capacity energy storage, controllable flexible load and the like are developed vigorously, so that distributed resources on the end side of a power grid tend to be changed into initiative and fragmentation gradually. Aiming at the scattered households with small capacity, large quantity and scattered positions, the scattered households are difficult to be used as independent main bodies to directly participate in the optimal scheduling of the active power distribution network, and the problems of decision variable dimension explosion, difficult convergence of solution results and the like can be caused if the independent solution is directly carried out, so that the problem that how to reasonably regulate and efficiently utilize controllable fragment resources with large quantity and small capacity in the active power distribution network is urgently needed to be solved by people.
Aiming at the current situation of grid connection of a large number of distributed controllable resources, resource clustering can effectively improve the utilization efficiency and flexibility of fragmented controllable power resources. Aggregation is an effective measure that can enable fragmented resources to participate in cooperative regulation, and is often applied to the field of power systems in recent years. The aggregation of the power resources is mainly to integrate distributed power units through an advanced technology to obtain aggregates with small quantity and large capacity, so that the utilization efficiency of electric energy and the deep regulation and control value of fragment resources are improved conveniently. Clustering is the derivation of aggregation, clustering and aggregation of fragment resources are realized based on comprehensive characteristic indexes considering multiple factors, and feasibility and resource utilization efficiency of participation of a large amount of fragment resources in scheduling are further improved.
Clustering algorithms have been widely applied to the integration and classification processes of various resources and data in power systems, but most of the conventional clustering algorithms have poor reliability and convergence when applied in the field of power systems. At present, research aiming at a clustering algorithm mainly focuses on the aspects of load curve classification, resource scene reduction, aggregation model building and the like, and research on distributed resource clustering serving for optimal scheduling of a power distribution network is very little, so that actual scheduling is lack of reference, and optimal scheduling of resources is not beneficial to adjustment.
Through the above analysis, the problems and defects of the prior art are as follows:
(1) In the prior art, the method is difficult to be used as an independent main body to directly participate in the optimization scheduling of the active power distribution network, and the problems of dimension explosion of decision variables, difficulty in convergence of solution results and the like can be caused if the method is directly solved.
(2) When most of traditional clustering algorithms are applied in the field of power systems, the reliability and the convergence are poor, the research on distributed resource clustering serving for optimal scheduling of a power distribution network is few, the actual scheduling is lack of reference, and the optimal scheduling of resources is not beneficial to adjustment.
Disclosure of Invention
Aiming at the problems in the background art, the invention provides a multi-resource dynamic clustering method and a multi-resource dynamic clustering system for improving a K-means + + algorithm.
In order to solve the technical problems, the invention adopts the following technical scheme: the multi-resource dynamic clustering method for improving the K-means + + algorithm comprises the following steps:
aiming at the active scheduling performance of the targets and resources served by the clustering, establishing a resource active scheduling index system, and calculating each index value in the index system;
calculating the clustering index weight of each index in the index system by combining subjective and objective factors, and dynamically adjusting the clustering index weight based on real-time power grid scheduling requirements;
and realizing dynamic clustering of distributed adjustable resources based on the improved K-means + + algorithm of the clustering index weight.
In the above multi-resource dynamic clustering method based on the improved K-means + + algorithm, the step of establishing a resource active scheduling index system based on the active scheduling performance of the resource for the target served by the clustering includes:
setting the controllability index, the economic index and the reliability index as a first-layer evaluation index of a resource active power scheduling index system;
based on the controllability index, taking the control capacity, the control accuracy, the control qualification rate and the response time as evaluation indexes, and constructing a second-layer evaluation index corresponding to the controllability index; constructing a second-layer evaluation index corresponding to the economic index by taking the unit scheduling cost as the evaluation index based on the economic index; and constructing a second-layer evaluation index corresponding to the reliability index by taking the power shortage time probability and the power shortage expected value as the evaluation indexes based on the reliability index.
In the multi-resource dynamic clustering method based on the improved K-means + + algorithm, the step of calculating the clustering index weight of each index in the index system by combining subjective and objective factors, and dynamically adjusting the clustering index weight based on real-time power grid scheduling requirements includes the following steps: quantifying the relative importance of each index in an index system by using an analytic hierarchy process, and determining the subjective weight of the related index; determining objective weight of each index system by using an entropy weight resisting method; and obtaining the comprehensive weight of the index system by adopting a linear weighting method based on the subjective weight and the objective weight.
In the above method for dynamically clustering multiple resources by using an improved K-means + + algorithm, the step of dynamically clustering distributed controllable resources by using the improved K-means + + algorithm based on the cluster index weight includes the following steps: clustering active scheduling indexes based on an index weight improved K-means + + algorithm comprises the following steps: clustering resources into K types, respectively establishing a multi-resource clustering model with active scheduling capability aiming at each aggregation cluster, and obtaining a clustering result based on the multi-resource clustering model; and carrying out validity check on the clustering result.
The system for improving the multi-resource dynamic clustering method of the K-means + + algorithm comprises a resource active scheduling index module, a resource active scheduling index calculation module and a resource active scheduling index calculation module, wherein the resource active scheduling index calculation module is used for establishing a resource active scheduling index system aiming at a target served by clustering and calculating each index value in the index system;
the weight adjusting module is used for calculating the clustering index weight of each index in the index system by combining subjective and objective factors and dynamically adjusting the clustering index weight based on real-time power grid scheduling requirements;
and the K-means + + module is used for realizing dynamic clustering of the distributed adjustable resources based on the improved K-means + + algorithm of the clustering index weight.
An electronic device, a computer-readable storage medium storing computer-executable instructions; and one or more processors coupled to the computer-readable storage medium and configured to execute the computer-executable instructions to cause the apparatus to perform the method of any of claims 1-4.
A readable storage medium storing computer-executable instructions that, when executed by a processor, configure the processor to perform the method of any one of claims 1-4.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides a multi-resource dynamic clustering method based on an improved K-means + + algorithm, which comprises the following steps: aiming at a target served by clustering, fully considering the active scheduling performance of resources, establishing a resource active scheduling index system, and calculating each index value in the index system; calculating clustering index weight by combining subjective and objective factors, and dynamically adjusting the weight based on real-time power grid scheduling requirements; based on the method, the dynamic clustering of multiple resources is realized by adopting a weight-improved K-means + + algorithm, and a dynamic clustering model of the adjustable resource cluster is obtained to support the process that distributed resources participate in ADN optimization scheduling.
According to the invention, a K-means + + clustering algorithm improved based on dynamic weights is applied to clustering of multiple resources, a real-time dynamic clustering model is built, the regulated and controlled resources with small capacity and large quantity are reduced into resource clusters with large capacity and small quantity, support is provided for high-proportion resource participation regulation and control, the feasibility and the resource utilization efficiency of participation and scheduling of a large amount of fragment resources are further improved, and the method has the comprehensive advantages of being more economical, reliable and easy to schedule in the aspect of meeting real-time regulation and control requirements compared with the traditional method.
The invention adopts an improved K-means + + algorithm to realize dynamic clustering of distributed resources so as to achieve better clustering effect.
The distributed algorithm has wide application in the aspect of optimal scheduling of a power system, but most of the current researches realize partition optimization according to different spatial and geographic positions, the research on virtual partitions based on different distributed resource benefit bodies is very little, and in addition, under the scene of high-proportion resource grid connection, a multi-resource clustering model needs to be considered in the distributed optimization process of an active power distribution network. Based on the clustering with the regulation and control sequence, the complexity of task allocation can be reduced, and the better economic advantage is taken into consideration, so that the scheduling result is more reasonable, and the virtual partition optimization based on the difference of distributed resource benefit main bodies under the scene of high-proportion resource grid connection is supported.
Drawings
FIG. 1 is a schematic flow chart of a multi-resource dynamic clustering method for improving a K-means + + algorithm according to an embodiment of the present invention;
fig. 2 is a resource active power scheduling comprehensive index system of a multi-resource dynamic clustering method of an improved K-means + + algorithm according to an embodiment of the present invention;
FIG. 3 is a schematic flow chart of the K-means + + algorithm considering dynamic weights according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of an IEEE33 node power distribution system according to an embodiment of a multi-resource dynamic clustering method for improving a K-means + + algorithm provided by the embodiment of the present invention;
fig. 5 is a graph showing a relationship between a clustering target and a clustering number in an embodiment of the multi-resource dynamic clustering method for improving the K-means + + algorithm according to the embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the following embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the embodiments and features of the embodiments may be combined with each other without conflict.
The present invention is further illustrated by the following examples, which are not to be construed as limiting the invention.
The technical problem to be solved by this embodiment is to realize reasonable regulation and control and efficient utilization of large-quantity small-capacity controllable fragment resources in an active power distribution network, and provide a reference method for actual scheduling to improve the problem that hierarchical analysis is too subjective and entropy weight method is not sufficient, so that the evaluation index weight is set more reasonably. In the embodiment, an evaluation index system is constructed, subjective weight is determined by using an analytic hierarchy process, objective weight is determined by using an inverse entropy weight method, comprehensive weight of evaluation indexes is obtained by using a linear weighting method based on the subjective weight and the objective weight, the weight is dynamically adjusted according to the real-time scheduling requirement of a power grid, indexes are normalized, a multi-resource clustering model considering active scheduling capability is established based on each aggregation cluster, and dynamic clustering is realized based on a K-means + + algorithm improved by the index weight.
The embodiment provides a multi-resource dynamic clustering method for improving a K-means + + algorithm, which comprises the following steps: aiming at the target served by the cluster, fully considering the active scheduling performance of the resources, establishing a resource active scheduling index system, and calculating each index value in the index system; calculating clustering index weight by combining subjective and objective factors, and dynamically adjusting the weight based on real-time power grid scheduling requirements; and (3) realizing multi-resource dynamic clustering by adopting a weight-improved K-means + + algorithm to obtain a dynamic clustering model of the adjustable resource cluster so as to support the process that distributed resources participate in ADN optimal scheduling. A K-means + + clustering algorithm based on dynamic weight improvement is applied to clustering of multiple resources, a real-time dynamic clustering model is built, the regulated and controlled resources with small capacity and large quantity are reduced into resource clusters with large capacity and small quantity, support is provided for high-proportion resource participation regulation, and compared with the traditional method, the method has the comprehensive advantages of being economical, reliable and easy to schedule in the aspect of meeting real-time regulation and control requirements.
The embodiment is realized by the following technical scheme, as shown in fig. 1, the multi-resource dynamic clustering method for improving K-means + + includes the following steps:
s1, aiming at a target served by clustering, fully considering the active scheduling performance of resources, establishing a resource active scheduling index system, and calculating each index value in the index system.
The resource active scheduling index system in S1 is shown in fig. 2, and the construction of the resource active scheduling index system includes: setting the controllability index, the economic index and the reliability index as a first-layer evaluation index of a resource active power scheduling index system;
s1.1, constructing a second-layer evaluation index corresponding to the controllability index by taking the control capacity, the control accuracy, the control qualified rate and the response time as evaluation indexes based on the controllability index;
s1.1.1 important index of the controllability index, wherein the regulation capacity reflects the regulation performance of the distributed unit, the index is defined according to the maximum and minimum output of the power generation unit, and the calculation formula is as follows:
in the formula (I), the compound is shown in the specification,
respectively representing the maximum and minimum output force of the distributed unit;
5363 the calculation formula of the regulation accuracy of S1.1.2 is:
in the formula, P t* 、P t Respectively representing the output and the actual output of the time t distribution unit under the regulation strategy, wherein N is the total time segment number, C ap Starting up capacity for a station, and T is the action duration time of the adjustable resources;
the calculation formula of the S1.1.3 regulation and control qualified rate is as follows:
wherein, B j The expression of (c) is:
in the formula, A sets data according to scheduling requirements;
s1.1.4 response time is the regulation and control response time of active power, in the actual operation process, because the difference of the power regulation speeds of inverters of different resources is large, the regulation and control response time of each distribution unit often has difference, and if the difference is not considered, the problem is brought to the accurate issuing of active regulation instructions, and the optimization process of the active power distribution network which participates in the active power distribution network is further influenced. Therefore, in the process of clustering a large number of distributed resources, the regulation response time of the active power is required to be used as a clustering index.
S1.2, constructing a second-layer evaluation index corresponding to the economic index by taking unit scheduling cost as the evaluation index based on the economic index;
the economic index is the unit regulation cost of different regulation units, and the unit energy abandoning cost c of the photovoltaic unit PV The energy storage unit and the flexible load are adjusted by the unit adjustment cost c ESS And c DR 。
And S1.3, constructing a second-layer evaluation index corresponding to the reliability index by taking the power shortage time probability and the power shortage expected value as the evaluation index based on the reliability index.
The calculation formula of the power shortage time probability in the S1.3.1 reliability index is as follows:
in the formula, N d To the number of incomplete response states in the total time period, D dk The duration time corresponds to the k-th incomplete response, and T is the action duration time of the controllable resources. The incomplete response comprises two conditions of failure response and partial response;
5363 the formula for calculating the expected value of electricity shortage of S1.3.2 is:
in the formula (I), the compound is shown in the specification,
respectively representing the kth partial unresponsiveness and partial response duration, η k The corresponding response degree of the k part; t is the duration of the action of the controllable resource, N d For incomplete response over the total periodNumber of times, P t* And representing the output of the distribution unit at the moment t under the regulation strategy.
S2, calculating the clustering index weight of each index in the index system by combining subjective and objective factors, and dynamically adjusting the clustering index weight based on real-time power grid scheduling requirements;
s2.1, from the perspective of expert experience, quantifying the relative importance of each index in an index system by using an analytic hierarchy process, and determining the subjective weight of the related index;
s2.1.1 constructs an evaluation index judgment matrix:
wherein n represents the number of indexes, a ij Represents index x i And index x j The importance of the comparison between, when i = j, a ij =1, when i ≠ j, a ij ＝1/a ji 。a ij The size of (A) is obtained from the comparison scale of 1-9 according to expert experience;
s2.1.2 for consistency check:
determining the consistency ratio C of the matrix A R Is defined as:
wherein, C I Is an index of the consistency of the data,
in the formula of max Judging the maximum eigenvalue of the matrix; r I A constant associated with the order n represents a random consistency index; satisfies C R And the judgment matrix is reasonable when the judgment matrix is less than or equal to 0.1.
S2.1.3 subjective weights were calculated using geometric averaging:
wherein the content of the first and second substances,
a subjective weight indicating the ith evaluation index,
the nth root of the product of each element of row i,
s2.2, determining the objective weight of each index system by using an anti-entropy weight method;
s2.2.1 evaluation index standardization treatment;
for each group of data containing index values, the ith index y in the jth group of data is calculated ij The occupied proportion is as follows:
in the formula, n and m represent the number of indexes and the number of evaluation objects, respectively, y ij Is the i index, p, in the jth data ij Is y ij The occupied proportion;
s2.2.2 calculates the inverse entropy value of the i index:
s2.2.3 calculates the entropy weight of the i index:
in the formula (I), the compound is shown in the specification,
an objective weight representing the ith evaluation index.
S2.3, finally, based on the subjective and objective weight calculation result, obtaining the comprehensive weight of the evaluation index by adopting a linear weighting method;
s2.3, comprehensively considering two weighting methods according to a linear superposition method, wherein the comprehensive weight calculation method comprises the following steps:
where α is the subjective and objective weight distribution coefficient, w i And a comprehensive weight representing the ith evaluation index.
S3, realizing dynamic clustering of distributed adjustable resources based on a K-means + + algorithm improved by clustering index weight; the resource active scheduling index system is constructed based on the purpose of enabling resources to participate in active optimal scheduling of the active power distribution network.
S3, clustering active scheduling indexes based on a clustering index weight improved K-means + + algorithm, respectively establishing a multi-resource clustering model considering active scheduling capability based on each clustering cluster, and checking a clustering result;
s3.1 clustering the active scheduling indexes based on the improved K-means + + algorithm of the index weight, wherein a flow diagram of the K-means + + algorithm considering the dynamic weight is shown in FIG. 3, and the specific flow is as follows:
s3.1.1 clustering index normalization, because the data of each index have a long difference, the index needs to be normalized respectively. Meanwhile, according to different index properties, the index is divided into a positive index and a reverse index, wherein the larger the positive index is, the better the positive index is, the smaller the reverse index is, the better the reverse index is, and the specific normalization process is as follows:
in the formula (I), the compound is shown in the specification,
respectively express quantitiesPositive and negative indexes after conversion, x ij Indicating index data before quantization, minx ij 、maxx ij Respectively representing the minimum value and the maximum value of the index data;
s3.1.2 determines the number of clusters and constructs a target function;
the elbow method is a simple and effective method for determining the clustering number. The K-means + + algorithm aims at realizing large Euclidean distance among clusters and small Euclidean distance in the clusters, the objective function can be expressed as the sum of sample distances in the clusters, and the calculation formula is as follows:
wherein K is the total cluster number of clusters, N m Represents the number of m cluster samples,
and x m* I samples and cluster centers of the m clusters are represented respectively;
with the increase of the number of clusters, the polymerization degree of the clusters gradually increases, and the polymerization target gradually decreases. When K reaches the real clustering number, the clustering degree return is sharply reduced, the corresponding number is determined as the optimal clustering number K corresponding to the maximum value of the variable quantity of the objective function D and similar to the elbow of a relation graph of D and K;
s3.1.3 dynamic selection of initial cluster center;
k-means + + is based on K-means, and the relative distance between centroids can be further increased by setting the probability that different positions are used as clustering centers, so that a better clustering effect is achieved. The step of selecting an initial cluster center comprises:
s3.1.3.1 randomly selects an initial clustering center;
s3.1.3.2 calculates the Euclidean distance D (x) between each point in the sample and the nearest cluster center, and the Euclidean distance D (x) is calculated by the following formula:
in the formula (I), the compound is shown in the specification,
a feature vector representing the ith distribution unit of the m clusters,
the feature vector of the selected cluster center representing the m clusters,
a jth clustering feature quantity representing the ith distribution unit feature vector of the m clusters,
representing the j-th clustering characteristic quantity of the characteristic vector of the clustering center selected by the m clusters.
S3.1.3.3 calculating the probability that each sample point is selected as the next cluster center; the formula chosen as the cluster center probability is:
wherein X is the sample set of all sample points;
s3.1.3.4 selecting the sample point corresponding to the maximum probability value as the next clustering center;
s3.1.3.5 repeats S3.1.3.2, S3.1.3.3 and S3.1.3.4 until the number of the selected cluster centers reaches the optimal cluster number K, and the initial cluster center selection is finished.
S3.1.4 traditional K-means + + algorithm does not consider the weight brought by the actual meaning of the index for the euclidean distance, on this basis, the embodiment introduces the weight reflecting the importance degree of the scheduling performance index into the K-means + + algorithm, and meets the real-time scheduling requirements for different clustering indexes, and the calculation formula of the weighted euclidean distance is as follows:
in the formula u i Feature vector, u, representing the ith distribution unit k Representing the feature vector of the selected cluster center, x ij J cluster feature quantity, x, representing i distribution unit feature vector kj J-th cluster feature quantity, omega, representing cluster center feature vector j And representing the comprehensive weight coefficient corresponding to each index, and comprehensively considering subjective and objective factors and a dynamic weight adjusting strategy in weight calculation.
S3.1.5 executing algorithm to cluster the active scheduling indexes;
executing a standard K-means algorithm, and calculating the weighted Euclidean distance from each characteristic quantity to the clustering center, wherein the calculation formula is as follows:
in the formula (I), the compound is shown in the specification,
a feature vector representing the ith distribution unit of the m clusters,
the feature vector of the selected cluster center representing the m clusters,
a jth cluster feature quantity representing the ith distribution unit feature vector of the m clusters,
representing the jth cluster characteristic quantity omega of the cluster center characteristic vector selected by the m clusters j And representing the comprehensive weight coefficient corresponding to each index, and comprehensively considering subjective and objective factors and a dynamic weight adjusting strategy in weight calculation.
In addition, for each feature vector, the feature vector is classified into the cluster corresponding to the cluster center with the minimum Euclidean distance. In each cluster, the update formula of the cluster center is as follows:
in the formula, N m Representing the number of eigenvectors, u, in the m clusters i Representing the feature vector of the ith distribution unit.
And (5) repeating S3.1.5 until the average error criterion function converges, finishing clustering, wherein the calculation formula of the average error criterion function is as follows:
in the formula (I), the compound is shown in the specification,
feature vector of selected cluster center, u, representing m clusters m* Representing the updated clustering center feature vector of the m clusters;
s3.2 the multi-resource clustering model considering the active scheduling capability is respectively established based on each aggregation cluster, dynamic clustering of distributed controllable resources is realized based on a dynamic weight improved K-means + + algorithm, massive resources are clustered into K types, the multi-resource clustering model considering the active scheduling capability is respectively established for each aggregation cluster, specifically, the multi-resource clustering model includes scheduling capacity, scheduling cost, scheduling accuracy and the like of the clustered clusters, dimension reduction of decision variables is realized, and feasibility of multi-resource participation in active power distribution network optimization scheduling is improved. The method comprises the following steps:
s3.2.1 aggregates the controllable performance of each cluster:
in the formula, N m Number of controllable units, P, for cluster m t mj,max 、P t mj,min Respectively represents the upper limit and the lower limit of the active output of the j unit in the m cluster at the time t, P t m,max 、P t m,min Representing the upper and lower limits of the active power output of the m clusters at time tAnd, D t A,mj 、D t QR,mj Respectively representing the regulation accuracy and the regulation qualification rate of the jth unit in the m clusters at the time t, D t A,m 、D t QR,m Respectively represents the mean value of the regulation accuracy and the qualification rate of the m clusters at the T moment, T t mj Is the regulation response time T of the j unit in the m cluster at the time T t mj Regulating and controlling the mean value of response time for m clusters at the t moment;
s3.2.2 aggregates the unit regulation and control cost of each cluster:
in the formula, C t mj Is the unit regulation and control cost, C, of the jth cell in the m-th cluster at time t t m The average value of the unit regulation cost of the m clusters at the time t;
s3.2.3 aggregates the reliability indexes of each cluster:
in the formula, L OLP,mj 、E ENS,mj Respectively representing the power shortage time probability and the power shortage expected value, L, of the j unit in the m clusters OLP,m 、E ENS,m Respectively representing the probability mean value of m cluster power shortage time and the expected value mean value of power shortage.
S3.2.4 polymerizes the upper and lower limits of energy storage and the maximum landslide climbing rate:
in the formula, E t mj,max 、E t mj,min Respectively represents the upper limit and the lower limit of the energy storage of the jth unit in the m clusters with the energy storage function at the time t, E t m,max 、E t m,min Respectively representing the energy storage at time tUpper and lower limit of energy storage of m clusters, r t m,up 、r t m,d o wn And the maximum climbing and the landslide rates of the m clusters with active climbing constraints at the moment t are represented respectively.
S3.3, the specific process of carrying out validity check on the clustering result comprises the following steps:
the clustering effect and quality are evaluated through a clustering effectiveness index, a clustering effect measuring method generally quantifies the compactness and the separation degree of clusters, and ensures that the compactness in the clusters is high enough and the separation degree between the clusters is high enough. The specific definition includes:
S3.3.1DBI index;
the DBI index is also called classification accuracy index, and reflects the maximum similarity of each cluster by the average distance between samples, and is specifically defined as:
wherein K is the number of clusters, d (u) m ) Is the average distance of each point within cluster m. d (u) k ) Is the average distance from each point in the cluster k to the cluster center, c m Cluster center values for the clusters m, c k The center value is clustered for cluster k. I is DBI The smaller the cluster size, the lower the similarity between the clusters and the better the clustering effect.
S3.3.2CH index;
the CH index comprehensively considers the compactness inside the cluster and the dispersity of different clusters, and is respectively represented by D and T:
wherein K is the number of clusters, N m Representing the number of eigenvectors in the m clusters, c m Cluster the center value, ω, for the cluster m m,i Indicating the membership of i points to the cluster within cluster m,
representing the centre point, x, of the entire data set i The ith sample data is represented, and a specific CH index calculation formula is as follows:
therefore, the larger the D is, the smaller the T is, the better the separation degree and the middle compactness of the clustering cluster are, and the better the clustering effect and quality are.
The embodiment also provides a system for improving a multi-resource dynamic clustering method of a K-means + + algorithm, and the multi-resource dynamic clustering system based on the improved K-means + + algorithm comprises:
the resource active scheduling index module is used for fully considering the active scheduling performance of the resources aiming at the target served by the clustering, establishing a resource active scheduling index system and calculating each index value in the index system;
the weight adjusting module is used for calculating the clustering index weight of each index in the index system by combining subjective and objective factors and dynamically adjusting the clustering index weight based on real-time power grid scheduling requirements;
and the K-means + + module is used for realizing dynamic clustering of the distributed adjustable resources based on a clustering index weight improved K-means + + algorithm.
Example 1
An IEEE-33 node power distribution network system shown in figure 4 is selected as a research object, and concentrated resources (photovoltaic power stations, concentrated energy storage and flexible loads) connected with each node are expanded into distributed resources (photovoltaic field clusters, energy storage clusters and flexible load clusters). Wherein the node 7 is connected to the grid photovoltaic field group 1, and the scale is 60; the node 20 is connected to the grid photovoltaic field group 2, and the scale is 45; the node 12 is connected to a grid energy storage cluster with the scale of 30; the node 28 is connected to the grid and can interrupt the flexible load cluster, and the size is 30. In addition, a controllable wind turbine generator set is incorporated at the node 29, and a gas turbine generator set is incorporated at the node 3.
In this embodiment, a resource index of a certain real-time scheduling period in a day is used as an analysis object, which is shown in table 1:
table 1 resource clustering index considering active scheduling capability
Taking the photovoltaic field group 1 as an example, the clustering analysis comprises the following steps:
1) Performing index normalization;
the selected index in the embodiment relates to a positive index and a negative index, wherein the positive index comprises: regulating and controlling accuracy, qualification rate and capacity; the inverse indexes are as follows: scheduling cost, power shortage time probability, power shortage expectation and regulation response time, and respectively carrying out normalization processing on the indexes based on the formulas (13) and (14) to obtain normalized data which can participate in clustering processing.
2) Selecting an optimal clustering number;
respectively calculating the numerical value of the clustering target D when the clustering number changes, drawing a curve that D changes along with the change of the clustering number, and solving the optimal clustering number of the photovoltaic field group 1 by using an inflection point method, as shown in fig. 5: FIG. 5 depicts a descending curve of unitized cluster targets as a function of cluster number. When the clustering number is 3, the slope change of the descending curve of the clustering target reaches the maximum value, and the return caused by the increase of the clustering number is rapidly reduced, wherein the clustering number is the optimal clustering number. That is, in the photovoltaic field group 1, the optimal clustering number of 60 distributed photovoltaic units is 3.
3) Dynamically weighting the indexes;
since the normalized index data needs to meet the actual significance of the index during clustering, it is necessary to weight the index. The invention calculates the comprehensive weight of each clustering index based on an AHP-entropy weight resisting method, and dynamically adjusts the weight coefficient according to the weighting strategy. The dynamic weighting strategy comprises the following steps: the controllability and the reliability of the resource participation scheduling are preferably considered in the peak time, and the economy of the resource scheduling is preferably considered in the valley time. And then respectively calculating the comprehensive index weights of different time periods.
(1) Peak period
First, subjective weights are analyzed, and a criterion layer judgment matrix is shown in table 2:
TABLE 2 criterion layer decision matrix
Wherein A, B, C respectively indicates the controllability, reliability and economic index of the standard layer. As can be seen, the degree of importance ranks: controllability = reliability > economy.
Next, an index layer decision matrix is introduced. Table 3 shows an adjustability determination matrix, in which A1, A2, A3, and A4 respectively indicate an adjustment accuracy, an adjustment qualification rate, an adjustment capacity, and a response time.
TABLE 3 Regulation judgment matrix
Table 4 is a reliability determination matrix:
TABLE 4 reliability determination matrix
Wherein, B1 and B2 refer to the power shortage time probability and the power shortage expected value, respectively. The economic index C only relates to the regulation and control cost of the controllable unit, the weight is 1, and a judgment matrix does not need to be constructed.
In addition, the objective right is realized by adopting an entropy weight resisting method based on real-time rolling historical data. Finally, the integrated weights are determined by the AHP-inverse entropy weight method, as shown in table 5:
TABLE 5 peak period active scheduling index synthetic weight
Therefore, the weight of the regulation accuracy rate and the power shortage time probability of the response index controllability and the reliability is greater than the weight of the economic index, and the requirements on the resource controllability and the reliability in peak time are met better.
(2) Flat and valley time period
Compared with the peak time period, the flat time period and the valley time period have relatively low requirements on regulation and control performance and reliability indexes and have relatively high requirements on economic indexes. The subjective weights are analyzed, and the criterion layer judgment matrix is shown in table 6:
TABLE 6 criterion layer decision matrix
Wherein A, B, C respectively refers to the controllability, reliability and economy of the standard layer. It can be seen that the importance ranks: economy > controllability = reliability, the index layer judgment matrix is the same as the peak period.
The integrated weights were determined by the AHP-inverse entropy weight method, as shown in Table 7:
TABLE 7 Integrated weights of active scheduling indexes in the mean and valley periods
Therefore, the weight of indexes such as the regulation accuracy, the qualification rate, the power shortage probability and the like is smaller than the economic index weight, and the requirement on the economy of the regulation resource in the flat and valley time is met better.
It should be noted that the present embodiment may be realized by hardware, software, or a combination of software and hardware. The hardware portion may be implemented using dedicated logic; the software portions may be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or specially designed hardware. Those skilled in the art will appreciate that the apparatus and methods described above may be implemented using computer executable instructions and/or embodied in processor control code, such code being provided on a carrier medium such as a disk, CD-or DVD-ROM, programmable memory such as read only memory (firmware), or a data carrier such as an optical or electronic signal carrier, for example. The apparatus and its modules of the present invention may be implemented by hardware circuits such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of hardware circuits and software, e.g., firmware.
The present embodiment achieves some positive effects in the development or use process, and indeed has great advantages compared to the prior art, and the following description is made in conjunction with data, graphs and the like of the test process.
4) Analyzing a clustering result;
and optimizing the Euclidean distance of the K-means + + based on the index dynamic weight, so that the clustering effect is more in line with the real-time requirement. The specific clustering results of the photovoltaic farm group 1 in a certain real-time scheduling period in a day are shown in table 8:
TABLE 8 clustering results of photovoltaic field group 1
As can be seen from table 8, the positive index sequence of active scheduling considering controllability, economy, and reliability is: cluster 1> cluster 2> cluster 3, with the inverse index ordering: cluster 3> cluster 2> cluster 1, which shows that cluster 1 has the best successful scheduling capability, followed by cluster 2, with cluster 3 being the worst. Therefore, the priority ranking of each cluster when participating in active scheduling is as follows: the cluster 1 is greater than the cluster 2 is greater than the cluster 3, the economy and the feasibility of the active scheduling of the participation of multiple resources can be improved within a certain regulation range, and the dimensionality reduction of the number of the resources is realized.
5) Comparing and analyzing clustering results of different clustering methods;
the K-means and K-means + + and the clustering result of the K-means + + algorithm with dynamic weights adopted in this embodiment are compared and verified respectively, as shown in table 9:
TABLE 9 comparative analysis of clustering effects of different clustering methods
Therefore, the clustering method adopted in the embodiment has the advantages of the DBI index and the CH index, which indicates that the clustering effect is improved.
Example 2
Based on the same inventive concept, the application also provides an electronic device, a computer readable storage medium storing computer executable instructions; and one or more processors coupled to the computer-readable storage medium and configured to execute the computer-executable instructions to cause the apparatus to perform the above-described methods.
Example 3
Based on the same inventive concept, the present application also provides a readable storage medium storing computer-executable instructions that, when executed by a processor, configure the processor to perform the above-described method.
While the invention has been described with reference to a preferred embodiment, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the spirit and scope of the invention.
Claims (7)
1. The multi-resource dynamic clustering method for improving the K-means + + algorithm is characterized by comprising the following steps of:
aiming at the active scheduling performance of the targets and resources served by the clustering, establishing a resource active scheduling index system, and calculating each index value in the index system;
calculating the clustering index weight of each index in the index system by combining subjective and objective factors, and dynamically adjusting the clustering index weight based on real-time power grid scheduling requirements;
and realizing dynamic clustering of distributed adjustable resources based on the improved K-means + + algorithm of the clustering index weight.
2. The method for dynamically clustering multiple resources by improving the K-means + + algorithm as claimed in claim 1, wherein the step of establishing an index system for active scheduling of resources based on the active scheduling performance of resources for the target served by the cluster comprises:
setting the controllability index, the economic index and the reliability index as a first-layer evaluation index of a resource active power scheduling index system;
based on the controllability index, taking the control capacity, the control accuracy, the control qualification rate and the response time as evaluation indexes, and constructing a second-layer evaluation index corresponding to the controllability index;
constructing a second-layer evaluation index corresponding to the economic index by taking the unit scheduling cost as the evaluation index based on the economic index;
and constructing a second-layer evaluation index corresponding to the reliability index by taking the power shortage time probability and the power shortage expected value as the evaluation indexes based on the reliability index.
3. The method for multi-resource dynamic clustering according to the improved K-means + + algorithm in claim 1, wherein the step of calculating the cluster index weight of each index in the index system in combination with objective and subjective factors, and dynamically adjusting the cluster index weight based on real-time grid scheduling requirements comprises the following steps:
quantifying the relative importance of each index in an index system by using an analytic hierarchy process, and determining the subjective weight of the related index;
determining objective weight of each index system by using an entropy weight resisting method;
and obtaining the comprehensive weight of the index system by adopting a linear weighting method based on the subjective weight and the objective weight.
4. The method for dynamically clustering multiple resources by improving the K-means + + algorithm according to claim 1, wherein the step of dynamically clustering distributed controllable resources by the improved K-means + + algorithm based on the cluster index weights comprises the following steps:
clustering active scheduling indexes by using an index weight improved K-means + + algorithm comprises the following steps:
clustering resources into K types, respectively establishing a multi-resource clustering model with active scheduling capability aiming at each aggregation cluster, and obtaining a clustering result based on the multi-resource clustering model;
and carrying out validity check on the clustering result.
5. The system for the multi-resource dynamic clustering method for the improved K-means + + algorithm as claimed in any one of claims 1 to 4, which comprises a resource active scheduling index module for establishing a resource active scheduling index system for the target served by the clustering, and calculating each index value in the index system;
the weight adjusting module is used for calculating the clustering index weight of each index in the index system by combining subjective and objective factors and dynamically adjusting the clustering index weight based on real-time power grid scheduling requirements;
and the K-means + + module is used for realizing dynamic clustering of the distributed adjustable resources based on the improved K-means + + algorithm of the clustering index weight.
6. An electronic device, characterized by a computer-readable storage medium storing computer-executable instructions; and one or more processors coupled to the computer-readable storage medium and configured to execute the computer-executable instructions to cause the apparatus to perform the method of any of claims 1-4.
7. A readable storage medium having stored thereon computer-executable instructions that, when executed by a processor, configure the processor to perform the method of any one of claims 1-4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211438073.XA CN115759626A (en) | 2022-11-16 | 2022-11-16 | Multi-resource dynamic clustering method and system for improving K-means + |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211438073.XA CN115759626A (en) | 2022-11-16 | 2022-11-16 | Multi-resource dynamic clustering method and system for improving K-means + |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115759626A true CN115759626A (en) | 2023-03-07 |
Family
ID=85372349
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211438073.XA Pending CN115759626A (en) | 2022-11-16 | 2022-11-16 | Multi-resource dynamic clustering method and system for improving K-means + |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115759626A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116108357A (en) * | 2023-04-11 | 2023-05-12 | 武汉大学 | Electrolytic aluminum FCM clustering method and system considering adjustment capability difference |
| CN116502111A (en) * | 2023-06-27 | 2023-07-28 | 武汉大学 | Distributed resource clustering method and system based on index weight improved K-means++ algorithm |
-
2022
- 2022-11-16 CN CN202211438073.XA patent/CN115759626A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116108357A (en) * | 2023-04-11 | 2023-05-12 | 武汉大学 | Electrolytic aluminum FCM clustering method and system considering adjustment capability difference |
| CN116108357B (en) * | 2023-04-11 | 2023-08-15 | 武汉大学 | Electrolytic aluminum FCM clustering method and system considering adjustment capability difference |
| CN116502111A (en) * | 2023-06-27 | 2023-07-28 | 武汉大学 | Distributed resource clustering method and system based on index weight improved K-means++ algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115759626A (en) | Multi-resource dynamic clustering method and system for improving K-means + | |
| CN110994694B (en) | Micro-grid source-charge-storage coordination optimization scheduling method considering differentiated demand response | |
| CN107069776B (en) | Energy storage look-ahead distributed control method for smooth microgrid tie line power | |
| CN109600178A (en) | The optimization method of energy consumption and time delay and minimum in a kind of edge calculations | |
| CN107482692B (en) | Active control method, device and system for wind power plant | |
| CN109378864B (en) | Source-network-load coordination optimization control method based on new energy consumption | |
| CN115117931A (en) | Power distribution network planning method and system considering electric vehicle flexibility and photovoltaic access | |
| CN115275983A (en) | Photovoltaic power fluctuation stabilizing control method based on distributed resource clustering | |
| CN115995838A (en) | Distributed energy storage and electric automobile oriented weighted clustering method, storage medium and electronic equipment | |
| CN111967733A (en) | Fuzzy comprehensive evaluation method for power peak regulation potential of aggregation group | |
| Zhu et al. | Research on energy storage capacity configuration for PV power plants using uncertainty analysis and its applications | |
| CN111509750B (en) | Power grid side energy storage system capacity configuration optimization method | |
| CN116302404B (en) | Resource decoupling data center-oriented server non-perception calculation scheduling method | |
| CN116914847A (en) | Aggregation complementation and optimal scheduling method for multiple generator sets in virtual power plant | |
| CN115758193A (en) | Distributed energy storage aggregation control method and device | |
| CN116090624A (en) | Fine granularity load segmentation prediction method | |
| CN110970939B (en) | Distributed energy cluster optimization method and system | |
| CN113765125A (en) | Cluster power distribution method for hydrogen energy storage system | |
| CN113991650B (en) | Light storage polymerization peak regulation method based on K-means + + algorithm | |
| CN111951123B (en) | Method and device for controlling electrical load, computer equipment and storage medium | |
| Jiang et al. | Intelligent Clustering Based Active Power Allocation Strategy for Wind Turbines | |
| CN117674197B (en) | Frequency adjustment method, storage medium and equipment using virtual power plant active support | |
| CN116502111A (en) | Distributed resource clustering method and system based on index weight improved K-means++ algorithm | |
| CN116659128B (en) | Energy-saving control method and system for solar energy combined heat pump | |
| CN113902291A (en) | Clustering feature-based temperature control equipment cluster control method and system |
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 |