CN115861671A - Double-layer self-adaptive clustering method considering load characteristics and adjustable potential - Google Patents

Abstract

The invention provides a double-layer self-adaptive clustering method considering load characteristics and adjustable potential. The method provides an improved adaptive density peak algorithm for clustering daily load curves of power consumers at one time, realizes clustering of unbalanced load data sets under the influence of flexible resources and power consumption behaviors, and has adaptive capacity and strong robustness. And the method also provides the method for roughly positioning the power users with strong regulation potential by combining a principal component analysis algorithm with a self-organizing competitive neural network aiming at the secondary clustering of the regulation potential under the consideration of the influence factors of the power consumption behavior of the power users. And (4) correcting the primary clustering result by adopting BP neural network feedback according to the secondary clustering result to obtain a comprehensive clustering result, and evaluating the clustering effect by combining the coefficient and the Davis index. And finally, designing a most possible adjustment amount and a maximum adjustment amount to quantitatively calculate the adjustable potential of the target user. The method has the advantages that accurate evaluation and positioning are carried out on users, the adjustable potential is enabled, and data support is provided for load aggregators and power operators.

Description

Double-layer self-adaptive clustering method considering load characteristics and adjustable potential

Technical Field

The invention relates to user load demand response and intelligent measurement in the field of electric power, in particular to a double-layer self-adaptive clustering method considering load characteristics and adjustable potential.

Background

Due to the large access of new energy and flexible resources, the user load gradually becomes a generalized load, the user load comprises various types such as temperature control load, electric vehicles and distributed energy storage, and the difficulty of accurate calculation and cluster calculation of the user adjustable power is high simply from the equipment layer. Because the data of the power users obtained by advanced communication, measurement and data management technology can reflect the user behavior characteristics, potential users can be effectively positioned aiming at the accurate clustering of the power users, adjustable resources can be rapidly screened, and data support is provided for application scenes such as power market operation, power system scheduling and the like.

Most of the existing user load data clustering analysis methods only cluster user load data sets, and static clustering is performed by applying a traditional clustering algorithm according to a load curve of a user, but a large amount of renewable energy is accessed, and load fluctuation, randomness, load characteristic difference among users and the like cause large cluster shape difference and unbalanced distribution of the load data sets, but the existing clustering algorithm cannot be suitable for any cluster distribution condition and has no self-adaptive capacity; at present, a user load curve is mostly adopted for clustering power utilization modes of power users, but the fact that the differences of power utilization behaviors, attitudes, habits and the like of the users caused by multidimensional influence factors of the power utilization behaviors of the users are not considered, and the clustering adjustable potential of the users is difficult to accurately model and quantify; the existing load adjustable potential research aiming at power consumers can effectively realize the customer potential mining of demand response projects, but the load adjustable potential of the demand response projects is not specifically quantized, a specific calculation method is not provided, and the load adjustable potential research is not beneficial to the flexible and reasonable distribution of power energy.

Disclosure of Invention

In order to solve the challenges brought by participation of large-scale power users in power grid demand side scheduling, uneven user load corresponding conditions and serious user behavior characteristic differentiation, and urgently need to develop fine classification of the power users to achieve the purpose of high-quality interactive adjustment of the power users and a power grid, the invention provides a double-layer self-adaptive clustering method considering load characteristics and adjustable potential. On one hand, the method can better realize the clustering of the unbalanced load data set under the influence of flexible resources and power consumption behaviors by providing an improved adaptive density peak algorithm for once clustering the daily load curve of the power consumer, and has adaptive capacity and strong robustness. On the other hand, the method adopts the combination of a principal component analysis algorithm and a self-organizing competitive neural network, and roughly positions the power users with strong adjusting potential aiming at adjustable potential secondary clustering under the consideration of the influence factors of the power utilization behavior of the power users. And (4) correcting the primary clustering result by adopting BP neural network feedback according to the secondary clustering result to obtain a more accurate comprehensive clustering result, and evaluating the clustering effect by combining the coefficient and the Davis index. And finally, the adjustable potential of the target user is quantitatively calculated by the most possible adjustment amount and the maximum adjustment amount. The load clustering method not only can accurately evaluate and position users, but also can quantify the adjustable potential, provides effective data support for load aggregators and power operators, and solves the technical problems that when a traditional clustering algorithm is used for analyzing different load data sets, certain parameters need to be manually adjusted in each clustering process, the traditional clustering algorithm cannot be suitable for the situations of more load cluster shapes, and the computing speed is low and the robustness is poor.

The technical scheme adopted by the invention for solving the technical problems is as follows:

a double-layer self-adaptive clustering method considering user load characteristics and adjustable potential is characterized by comprising the following steps:

step 1: acquiring load data and user basic information data of a power user;

step 2: aiming at the daily load characteristic curve of the power consumer, an improved adaptive density peak algorithm is constructed to perform primary clustering analysis;

step 2.1: acquiring user power consumption data according to the intelligent electric meter, and further forming a power consumption load curve data set of each user;

step 2.2: data preprocessing is carried out on load characteristic data, wherein data loss and noise are caused by the influence of factors such as errors of measurement and communication, system faults and the like, and the data loss and noise comprise missing data filling, curve smoothing filtering and data normalization processing;

step 2.3: obtaining a parameter k in a self-adaptive manner according to the sample condition, and calculating by adopting a Natural Nearest Neighbor (3N) algorithm to obtain a Natural characteristic value sup _k K as KNN specifically includes:

step 2.3.1: initialization search index r =1, inverse neighbor set

Step 2.3.2: calculate each sample x _i KNN (K) _r (x _i )、RNN(x _i )；

Step 2.3.3: r = r +1;

step 2.3.4: when in use

So that->

Or let all->

X of _j When no longer changed, sup _k = r-1, output sup _k Otherwise, jumping to step 2.3.2;

step 2.4: determining the cluster center by utilizing an automatic cluster center selection mode, wherein the decision value of the cluster center is far greater than that of other samples, so that the decision value is subjected to descending order arrangement to generate a segmentation phenomenon, and the decision value is fitted by two straight lines in a segmentation manner, so that the segmentation point with the minimum fitting error is the optimal cluster center;

step 2.5: distributing the rest sample points according to a sample distribution strategy based on the weighted KNN graph;

step 2.6: the two evaluation indexes of the contour coefficient (SC) and the Davignon index (DBI) are combined to evaluate the accuracy of the clustering result;

and step 3: aiming at the big power data considering the influence factors of the power consumption behaviors of the users, a method based on the combination of principal component analysis and a self-organizing competitive neural network is provided for carrying out secondary cluster analysis on the power consumption patterns of the users;

step 3.1: carrying out PCA (principal component analysis) dimension reduction processing on the power data considering the influence factors of the power utilization behaviors of the users;

step 3.2: based on the PCA dimension reduction result, adopting a self-organizing competitive neural network to perform secondary clustering on the adjustable potential of the user;

and 4, step 4: the secondary clustering result is used as a training data set of the BP neural network, and the primary clustering result is fed back and corrected to obtain a comprehensive clustering result;

step 4.1: performing primary clustering on the user load data set to obtain K user groups with similar load characteristics;

step 4.2: performing secondary clustering on the user groups with the same type of user electrical load characteristics;

step 4.3: taking the result after the secondary clustering as a training set of a BP neural network, reversely correcting the primary clustering result, and utilizing the strong self-adaptive learning capability and nonlinear mapping capability of the BP neural network to enable the secondary clustering output to flexibly adjust the primary clustering result;

step 4.4: repeating the secondary clustering process according to the corrected primary clustering result to obtain a comprehensive clustering result finally considering the load characteristics and the adjustable potential of the user;

and 5: aiming at the comprehensive clustering result, a quantitative user adjustable potential calculation method is provided;

step 5.1: judging a cluster to which the target power load belongs according to the classification result and the power utilization mode to which the target power user regulation daily load baseline belongs;

step 5.2: calculating the flexible adjustable potential of the regulation time interval of the target user in the regulation day;

step 5.3: calculating the most possible adjusting potential delta P of the power consumer according to the calculation result _ii (t)＝x _t -x _m And maximum regulatory potential Δ P _ik (t)＝x _t -x _k 。

Compared with the prior art, the invention has the following beneficial effects:

(1) The method and the device comprehensively analyze the power utilization behaviors of the users, and consider information mining of user load data and information mining of electric power big data of user power utilization behavior influence factors, so that flexible interaction of user load characteristics and adjustable potential is realized, and a more accurate clustering effect is achieved.

(2) Aiming at various user load data with huge characteristic differences acquired by the intelligent electric meter, the improved adaptive density peak algorithm can better cluster the load curve of the power user, the improved adaptive density peak algorithm can better realize the clustering of unbalanced load data sets, and the improved adaptive density peak algorithm has adaptive capacity and simultaneously enhances the system robustness.

(3) Aiming at a data set which is acquired by a questionnaire and takes influence factors of the power consumption behavior of a user into consideration, the principal component analysis method is adopted to reduce the dimension of the data set, so that the operation time can be reduced under the requirement of accuracy. Based on the dimension reduction data set, the user adjustable potential secondary clustering is performed by using the self-organizing competitive neural network, and the algorithm is an unsupervised learning method and can better process the scene without clear clustering standard.

(4) The method constructs a double-layer clustering model based on a reverse regulation principle, and utilizes a feedback mechanism to reversely correct the result of the primary clustering by training the result of the secondary clustering through a BP neural network. And the strong self-adaptive learning capacity and the nonlinear mapping capacity of the BP neural network are adopted, the output of secondary clustering is flexibly utilized to adjust the primary clustering result, so that a more refined comprehensive clustering result is obtained, and decision support is provided for user energy scheduling.

(5) And quantitatively calculating the power users with strong power adjustable potential qualitatively determined after secondary clustering according to the comprehensive clustering result, dividing the power users into the most possible adjustment potential and the maximum adjustment potential, and specifically calculating the load potential of the power users to provide data reference and support for load aggregators and power operators.

The method is suitable for being applied as a double-layer self-adaptive clustering method considering load characteristics and adjustable potential.

Drawings

FIG. 1 is a diagram of a comprehensive analysis framework for a two-tier adaptive clustering method that takes into account load characteristics and scalability potential;

FIG. 2 is a flow chart of an algorithm for clustering a daily load curve data set of power by using an improved adaptive density peak algorithm in the method of the present invention;

FIG. 3 is a flow chart of a quadratic clustering algorithm considering the influence of the user electricity consumption behavior factor in the method of the present invention;

FIG. 4 is a flow chart of applying BP neural network feedback correction in the method of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. 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.

Because the user power load adjustable force comprehensive analysis relates to two modules of user power load data mining analysis and user power consumption behavior influence factors, a user load curve is formed according to load data collected by a user intelligent electric meter, the load characteristic is shown, and the load curve is insufficient information; carrying out multidimensional analysis aiming at the influence factors of the power utilization behavior of the power consumer so as to accurately identify the power utilization mode of the user, representing the schedulable potential of different users, and being sufficient information; and after qualitative clustering of the user load is carried out on the power user data sets in the two aspects, the adjustable potential of the user load with strong adjustable potential is accurately and quantitatively calculated. A power consumer double-layer clustering model based on a reverse correction principle is provided, firstly, aiming at a consumer load data set, an improved self-adaptive peak density algorithm is adopted to perform primary clustering of consumer loads, and 2 internal index evaluation clustering results of a profile coefficient (SC) and a Davis index (DBI) are provided. Then, on the basis of primary clustering, aiming at power data with user power consumption behavior influence under the same power consumption mode, secondary clustering with adjustable potential is carried out on the users, the users under each mode are roughly classified into power users with strong adjustable potential and power users with weak adjustable potential qualitatively, principal Component Analysis (PCA) is specifically applied to dimensionality reduction processing of the data set, and then self-organizing competitive neural network (SOM) is used for completing secondary clustering calculation on the user potential. Then, based on a reverse correction principle, the secondary result is used as a training data set of the BP neural network, the primary clustering result is corrected reversely, then the secondary clustering process is repeated, the final comprehensive clustering effect is calculated, and the clustering result is evaluated by using a mode of combining the profile coefficient and the Theisen index. And finally, aiming at the comprehensive clustering result of the power consumption of the user, and aiming at potential users, providing an adjustable potential quantitative calculation mode of the user load.

Referring to the attached figure 1 of the specification, a double-layer self-adaptive clustering method and system considering load characteristics and adjustable potential:

acquiring intelligent power load electricity consumption data and user basic information data according to an intelligent ammeter;

performing primary clustering of power utilization modes aiming at actual power utilization data of power consumers, and specifically adopting an improved adaptive density peak algorithm to perform clustering analysis aiming at daily load curves;

according to the preliminary calculation result, secondary clustering is carried out on the same type of load characteristic user groups by adopting the electric power big data considering the influence factors of the electric power consumption behaviors of the users, and the specific method is combination analysis of principal component analysis and a self-organizing competitive neural network.

And designing a specific algorithm to quantitatively calculate the adjustable potential of the specific time of the power consumer according to the operation result.

Referring to the attached figure 2 of the specification, an improved adaptive density peak value clustering (ISDPC) algorithm is provided for carrying out primary clustering on a user load data set by considering double-layer clustering of user load characteristics and power utilization modes, the improved adaptive density peak value clustering algorithm redefines a new density measurement method by using K-Nearest Neighbor (KNN) and a relative density idea, then a piecewise function is fitted in a decision diagram to obtain the optimal clustering number, and finally a weighted KNN diagram is constructed to improve a data set distribution clustering strategy. The method specifically comprises the following steps:

step 1: acquiring user power consumption data according to the intelligent electric meter, and further forming a power consumption load curve data set of each user;

step 2: the method aims at the problems of data loss, noise and the like caused by the influence of factors such as measurement and communication errors and system faults and the like in the data preprocessing of load characteristic data, and comprises missing data filling, curve smoothing filtering and data normalization processing.

And step 3: obtaining a parameter k in a self-adaptive manner according to the sample condition, and calculating by adopting a Natural Nearest Neighbor (3N) algorithm to obtain a Natural characteristic value sup _k K as KNN specifically includes:

firstly, the algorithm principle and the basic parameter calculation formula of the density peak algorithm are specifically as follows:

the DPC algorithm assumes that the local density at the center of each cluster is higher than the density of the surrounding neighboring points; assuming that the distance between centers of each cluster is long:

(1) Local density ρ _i . The calculation method comprises two calculation modes, namely a truncation core and a Gaussian core, which are respectively represented by formula (1) and formula (2):

in the formula: if x is less than 0, then χ (x) =1, otherwise, 0; d _i,j Is a sample point x _i And x _j The Euclidean distance therebetween; d _c The truncation distance is indicated.

(2) Minimum distance delta _i . Represents the sample point and is higher thanThe density and distance between the nearest sample points of the convergence are given by the formula:

(3) Decision graph

The cluster-like center is a sample point with higher local density and larger minimum distance at the same time, and a decision value gamma is calculated _i Determining the cluster center, wherein the formula is as follows:

γ _i ＝ρ _i ·δ _i (4)

and the ordinate of the decision diagram is the descending order of the gamma values, the abscissa is the order of the arrangement, the first c gammas are selected as cluster centers according to the decision diagram, and then the rest sample points are distributed.

Then, a local density calculation mode is improved aiming at the idea that the density peak algorithm is based on KNN and relative density, so that the truncation distance does not need to be set manually and the method is suitable for data samples with large cluster density difference, and the specific process is as follows:

definition 1 (inverse nearest neighbor): if the sample point x _j Is the sample point x _i One of the K-nearest neighbors of (1) is called the latter as the inverse neighbor of the former, and is marked as x _i ∈RNN(x _j )。

Definition 2 (natural steady state): in the process of searching the natural neighbors, when each data point has reverse neighbors or the number of the reverse neighbors is 0, the natural neighbor searching reaches a natural stable state.

Definition 3 (natural eigenvalues): the search times of the natural neighbor search process reaching the natural stable state are natural characteristic values, the average number of the adjacent nodes of the data sample is represented, and the average number is counted _k 。

Step 3.1: initialization search index r =1, inverse neighbor set

Step 3.2: calculate each sample x _i KNN of _r (x _i )、RNN(x _i )。

Step 3.3: r = r +1.

Step 3.4: when in use

So that->

Or let all->

X of _j When no longer changed, sup _k = r-1, output sup _k Otherwise, jump to step 3.2.

Then, the relative density of the sample points in the local range is calculated to reduce the problem that the sparse cluster center can not be found due to the overlarge distribution density difference of various clusters in the data sample, and the improved local density formula is as follows:

wherein, KNN _k (x _i ) Is a sum of x _i The set of k samples closest. The minimum distance and decision value are then calculated.

And 4, step 4: determining a cluster center v by utilizing an automatic cluster center selection mode, wherein the decision value of the cluster center is far greater than that of other samples, so that the decision values are subjected to descending order arrangement to generate a segmentation phenomenon, the decision values are piecewise fitted by using two straight lines, the segmentation point with the minimum fitting error is the optimal cluster center, and the specific calculation steps are as follows:

step 4.1: the decision values are arranged in descending order, with n points (i, ρ) _i ) Forming a data set S;

step 4.2: passing point S ₁ 、S _t And S ₁ 、S _n Make a straight line to obtain y ₁ ＝a ₁ x+b ₁ And y ₂ ＝a ₂ x+b ₂ The corresponding fit value is calculated for x =1,2.

Step 4.3: error in calculating fitDifference theta _t The formula is as follows:

step 4.4: let t = t +1, go to step 4.5 when t > n, otherwise return to step 4.2;

step 4.5: calculated to minimum theta _t At this time, t is the optimal cluster center number, and the sample points corresponding to the first t γ are the cluster center v = [ v ] s ₁ ,v ₂ ,...v _t ]。

And 5: distributing the rest sample points according to a sample distribution strategy based on the weighted KNN image, wherein the specific calculation steps are as follows:

step 5.1: a weighted KNN graph G = (V, E, W) was constructed. Node set V consists of all sample points, if x _j ∈KNN _k (x _i ) Then x _i And x _j There is an edge e between _i,j E is E, its integrated weight W _i,j The formula of (1) is:

W _i，j ＝d _i，j +(1-J _i，j ) (7)

in the formula (d) _i,j Measuring the similarity of the samples for the Euclidean distance between the sample points; j is a unit of _i,j ∈[0,1]The structural similarity between samples is represented, the more the public connection points of two nodes in the KNN graph are, the larger the coefficient is, and the more the structures of the two nodes are similar; Γ (x) _i ) Represents x _i Set of adjacent points. The composite weight of the edges expands the differences between different clusters.

Step 5.2: calculating the shortest paths between the centers of various clusters and other nodes in the weighted KNN graph by using a Dijkstra algorithm to obtain a shortest path matrix L, wherein the shorter the path between the node and the center of the cluster, the more similar the node is, and the calculation formula is as follows:

and 5.3, distributing the residual load sample points to the belonged cluster, wherein the specific formula is as follows:

step 6: evaluating the accuracy of a clustering result, wherein the clustering analysis of an actual daily load curve lacks external information, the clustering result is evaluated by adopting 2 internal indexes of a contour coefficient (SC) and a Davis-Sen index (DBI), and a specific calculation formula is as follows:

if the daily load curve data set X of the user is divided into t clusters C ₁ ,C ₂ ,...C _t SC Chinese DBI is defined as,

/>

in the formula, b (x) _i ) And x _i Average distance between sample points, a (x), not of a kind _i ) And x _i Is the average distance between a class of samples. b (x) _i ) The larger, a (x) _i ) The smaller the clustering result, the better.

The average distance from the cluster center of the ith class to the point in the class; m _i,j Is the euclidean distance of the class i, j centers. The smaller the DBI value, the better the clustering effect.

Referring to the attached figure 3 of the specification, the user adjustable potential secondary clustering based on PCA dimension reduction and self-organizing competitive neural network is carried out. On the basis of the primary clustering result, qualitatively dividing users into power users with stronger adjustability and power users with weaker adjustability, and specifically comprising the following steps:

step 1: carrying out PCA (principal component analysis) dimension reduction processing on the electric power data considering the influence factors of the power utilization behaviors of the users, wherein the specific solving steps are as follows:

step 1.1: inputting the big power data considering the influence factors of the power consumption behaviors of the users as a data set of secondary clustering, wherein the information contained in the multidimensional influence factors of the power consumption behaviors of the users is shown in a table 1,

step 1.2: the raw data matrix X is calculated according to the following formula:

wherein n represents the number of power consumers; the power utilization behavior influence factor data of each user has p indexes; x ₁ ,X ₂ ,...X _p A set of initially set random variables; a is _ip Is a coefficient, and satisfies

i＝1,2,...p；x ₁ ,x ₂ ,...x _p The corresponding values of each index are obtained; f ₁ ,F ₂ ,...F _p Are new uncorrelated variables.

Step 1.3: calculating the mean value

Step 1.4: calculating a covariance matrix;

step 1.5: computing the set of eigenvectors of the covariance matrix V = { V = } ₁ ,v ₂ ,...v _n And a set of eigenvalues U = { λ = } ₁ ,λ ₂ ,...λ _n }；

Step 1.6: the principal component is determined.

TABLE 1 multidimensional influence factors for user power consumption behavior

/>

/>

When the main components of the influence factors of the power utilization behavior of the user are extracted, the first k indexes with the characteristic value larger than 1 and the cumulative variance contribution rate larger than 90% are selected, wherein the cumulative contribution rate

And taking the data after dimensionality reduction as input data of secondary clustering.

Step 2: based on PCA dimension reduction results, adopting a self-organizing competitive neural network to perform secondary clustering on the adjustable potential of the user, and specifically comprising the following steps:

step 2.1: and (5) initializing. Inputting principal component analysis vector X subjected to PCA (principal component analysis) reduced feature dimension reduction into the SOM network, and normalizing the principal component analysis vector X with the network weight corresponding to each neuron of the competition layer to obtain

And W _j (j＝1,2,...,m)。

Step 2.2: the winning neuron is calculated. Each of the above

All W with competing layer _j (j =1, 2.. Said., m) comparing the similarity, selecting the network weight vector with the maximum similarity as the competition winning neuron, and recording as ^ based on the competition winning neuron>

Counting/or>

And/or>

The euclidean distance of (a) determines the similarity, i.e.:

further, the method can be obtained as follows:

the two formulas are combined to obtain:

step 2.3: and adjusting the network weight. The winning neuron has the right to adjust its weight vector, and its neuron output is 1, and the output of the rest neurons is 0, and the weight vector after adjustment is:

where α is the learning rate, α ∈ (0, 1), and gradually decreases during the learning process.

Step 2.4: and returning to the step 2.1 to continue training, stopping training when alpha is attenuated to a set value or 0, and outputting a training result.

Referring to the attached figure 4 of the specification, a double-layer clustering model for feedback correction of the BP neural network is constructed by considering double-layer clustering of user load characteristics and power utilization modes. Considering that a data set used by primary clustering is insufficient information, and a data set used by secondary clustering is sufficient information, correcting errors generated during primary clustering by means of a feedback mechanism, and simultaneously associating the power load characteristics of a user with adjustable potential to achieve a comprehensive clustering result considering the combination of the two, the specific implementation steps are as follows:

step 1: performing primary clustering on the user load data set to obtain K user groups with similar load characteristics;

step 2: performing secondary clustering on the user groups with the same type of user electrical load characteristics;

and step 3: taking the result after secondary clustering as a training set of a BP neural network, reversely correcting the primary clustering result, and utilizing the strong self-adaptive learning capability and the nonlinear mapping capability of the BP neural network to enable secondary clustering output to flexibly adjust the primary clustering result;

and 4, step 4: and repeating the secondary clustering process according to the corrected primary clustering result to obtain a comprehensive clustering result finally considering the user load characteristics and the adjustable potential.

In the embodiment, double-layer clustering of user load characteristics and power consumption modes is considered, and a quantitative calculation method for user load adjustment potential is provided, wherein after the adjustable potential of the comprehensive clustering result of the power users is qualitatively classified according to the method, the adjustable potential is quantitatively calculated for the power user load with stronger overall adjustment capacity, and the specific steps are as follows:

firstly, judging the electricity utilization mode of a target power user regulation daily load baseline according to a classification result, wherein a regulation daily cluster l is as follows:

wherein c is the number of clusters; x is a radical of a fluorine atom _t A load baseline of the t-th regulation day is generally a load average value of n working days before the regulation day, and n =5 is generally taken; mu.s _i Is the ith cluster center; xi shape _it Represents x _t Membership to the center of the ith class cluster.

Then, assume that the load regulation daily load curve belongs to the cluster C _m Cluster-like center is mu _m The load curve in the cluster is

Adjustment period t = [ t = _start ,t _end ]The load curve of the user is x _t Defining the flexible adjustable potential of t time period on a regulation day as delta P _im (t), m = i, i +1, i +2,.. K may be expressed as:

in the formula, mu _m A cluster center representing an mth cluster class that is not in the same cluster class as the baseline; x is the number of _t A load baseline for a regulation period; x is the number of _ψ The load curve with the farthest Euclidean distance in the same cluster with the load base line is obtained; delta P _im (t) flexible and adjustable potential for user load. Delta P _im (t) (m = i, i +1, i +2,.., k) represents the tunable potential in the user (k-i) in the different power usage modes.

Calculating the regulation day t according to the formula _d Flexible and adjustable potential under different power utilization modes of a time period is defined, and quantitative indexes for describing adjustable potential of a user are defined: most probable adjustment Δ P _ii (t), maximum adjustable quantity Δ P _ik (t) of (d). The most possible regulating quantity is a most possible realization adjustable capacity value for evaluating the user load, the specific calculation mode is a difference value between a baseline load of a regulation and control daily regulation period and the lowest load of a cluster to which the baseline load belongs, and the regulating mode is switching between power utilization modes of most similar users, so the indexes are defined as the most possible regulating quantity, and the specific calculation process is as follows:

if the daily load curve x is regulated _t Calculating the membership degree xi of the cluster to the center of the ith class cluster _ij The classification result is x _t ∈C _i Calculating a sample x _t With all samples x _j ∈C _i Euclidean distance of D _j ＝||x _t -x _j Obtaining a sample x with the maximum Euclidean distance from the regulation daily load curve _m Wherein

And during an adjustment period x _t -x _m Is greater than 0, wherein n is a sample set rejection curve x _t The number of curve bars remaining thereafter, so that in the regulation period t _d The most probable adjustment is Δ P _ii (t)＝x _t -x _m 。/>

The maximum adjustable quantity is the maximum value of the adjustable capacity which can be achieved by the evaluation user, the specific calculation mode is the difference value between the baseline load of the regulation and control daily regulation period and the cluster clustering center with the minimum load value, and the indexes are defined as the maximum adjustable quantity because the regulation mode changes the power consumption mode of the user to the maximum extent, and the specific calculation process is as follows:

sample x _t ∈C _i As the predicted load curve of the target user load regulation date t, in the regulation period t _d Inner calculation of cluster center μ _i Average load P of { i =1, 2.., k } _i ^ave Obtaining the minimum average load cluster number

During the regulation period t _d Inner maximum adjustable quantity Δ P _ik (t)＝x _t -x _k 。

According to the quantitative research aiming at the adjustable potential of the user layer, the load adjustable potential of the power user is preliminarily evaluated, data support is provided for further accurate calculation of adjustable power and analysis of an adjusting strategy, and a load aggregator and a power grid operator can also adopt the method to carry out mining screening on massive potential users and evaluation on the adjustable potential of regional load, so that the accurate mining efficiency of adjustable resources on the 'load' side is improved.

Finally, it should be noted that: while the present invention has been described in detail with reference to the foregoing embodiments, it should be apparent to those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein. The scheme in the embodiment of the application can be implemented by adopting various computer languages, such as object-oriented programming language Java and transliterated scripting language JavaScript.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

While the preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the present application.

It will be apparent to those skilled in the art that various changes and modifications may be made in the present application without departing from the spirit and scope of the application. Thus, if such modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include such modifications and variations as well.

Claims (5)

1. A double-layer self-adaptive clustering method considering user load characteristics and adjustable potential is characterized by comprising the following steps:

step 1: acquiring load data and user basic information data of a power user;

step 2: aiming at the daily load characteristic curve of the power consumer, an improved adaptive density peak algorithm is constructed to perform primary clustering analysis;

and step 3: aiming at the electric power big data considering the influence factors of the user power consumption behaviors, performing secondary cluster analysis on the user power consumption mode based on a method of combining principal component analysis and a self-organizing competitive neural network;

and 4, step 4: taking the secondary clustering result as a training data set of the BP neural network, and feeding back and correcting the primary clustering result to obtain a comprehensive clustering result;

and 5: and aiming at the comprehensive clustering result, a quantitative user adjustable potential calculation method is provided.

2. The method according to claim 1, wherein the step 2 comprises:

step 2.1: acquiring user power consumption data according to the intelligent electric meter to form a power consumption load curve data set of each user;

step 2.2: data preprocessing is carried out on load characteristic data, wherein data loss and noise are caused by the influence of factors such as errors of measurement and communication, system faults and the like, and the data loss and noise comprise missing data filling, curve smoothing filtering and data normalization processing;

step 2.3: obtaining a parameter k in a self-adaptive manner according to the sample condition, and calculating by adopting a Natural Nearest Neighbor (3N) algorithm to obtain a Natural characteristic value sup _k K as KNN specifically includes:

step 2.3.1: initialization search index r =1, inverse neighbor set

Step 2.3.2: calculate each sample x _i KNN of _r (x _i )、RNN(x _i )；

Step 2.3.3: r = r +1;

step 2.3.4: when in use

So that->

Or to have all +>

X of _j When no longer changed, sup _k = r-1, output sup _k Otherwise, jumping to step 2.3.2;

step 2.4: determining the cluster center by utilizing an automatic cluster center selection mode, wherein the decision value of the cluster center is far greater than that of other samples, so that the decision value is subjected to descending order arrangement to generate a segmentation phenomenon, and the decision value is fitted by two straight lines in a segmentation manner, so that the segmentation point with the minimum fitting error is the optimal cluster center;

step 2.5: distributing the rest sample points according to a sample distribution strategy based on the weighted KNN graph;

step 2.6: and the accuracy of the clustering result is evaluated by combining two evaluation indexes of the contour coefficient (SC) and the Davis-Sen index (DBI).

3. The method according to claim 1, wherein the step 3 comprises:

step 3.1: carrying out PCA (principal component analysis) dimension reduction processing on the power data considering the influence factors of the power utilization behaviors of the users;

step 3.2: and based on the PCA dimension reduction result, adopting a self-organizing competitive neural network to perform secondary clustering on the adjustable potential of the user.

4. The method according to claim 1, wherein the step 4 comprises:

step 4.1: performing primary clustering on the user load data set to obtain K user groups with similar load characteristics;

step 4.2: performing secondary clustering on the user groups with the same type of user electrical load characteristics;

step 4.3: taking the result after secondary clustering as a training set of a BP neural network, reversely correcting the primary clustering result, and utilizing the strong self-adaptive learning capability and the nonlinear mapping capability of the BP neural network to enable secondary clustering output to flexibly adjust the primary clustering result;

step 4.4: and repeating the secondary clustering process according to the corrected primary clustering result to obtain a comprehensive clustering result finally considering the user load characteristics and the adjustable potential.

5. The method according to claim 1, wherein the step 5 comprises:

step 5.1: judging the cluster to which the target power load belongs according to the classification result of the power consumption mode to which the target power user regulation daily load baseline belongs;

step 5.2: calculating the flexible adjustable potential of the regulation time interval of the target user in the regulation day;

step 5.3: calculating the most possible adjusting potential delta P of the power consumer according to the calculation result _ii (t)＝x _t -x _m And maximum regulatory potential Δ P _ik (t)＝x _t -x _k 。

Images