CN110781332A - Clustering method of daily load curve of electric residential users based on compound clustering algorithm - Google Patents

Abstract

The daily load curve clustering method of electric residential users based on compound clustering algorithm obtains the daily load data of electric residential users. The data contains P samples, and each sample has a data set matrix of Q time point attributes; The daily load data is preprocessed to obtain the initial cluster; the dimensionality reduction process is performed on the initial cluster to obtain the dimensionality reduction cluster; the clustering algorithm 1 is used to perform preliminary clustering on the dimensionality reduction cluster to obtain the initial cluster center; the clustering algorithm 2 is used to obtain the initial clustering center. The initial clustering centers obtained by clustering algorithm 1 are clustered, and the clustering effectiveness index is used to evaluate the clustering results, and finally M clustering centers are obtained; the obtained M clustering centers are used as the clustering algorithm. 2 is the initial clustering center, and the data is clustered to obtain user groups with similar behaviors. The invention clusters the huge and scattered daily load data into user groups with similar behaviors. The power enterprise managers can analyze the clustered user groups to better predict the peak and trough of electricity consumption, and provide a more reliable method for the management of power business.

Description

Clustering method of daily load curve of electric residential users based on compound clustering algorithm

technical field

The invention relates to the technical field of electricity consumption of electric residential users, in particular to a method for clustering daily load curves of electric residential users based on a compound clustering algorithm.

Background technique

With the rapid development of the power industry and the popularization of smart meters, it has become more convenient to obtain the electricity consumption of electricity households.

In the face of huge electricity consumption data, the existing data mining and analysis technology is used to conduct regular analysis and feature extraction on the daily load data of power users, so as to facilitate the power company to provide users with higher quality power supply services according to the electricity price policy. Among them, the subdivision of residential users' electricity consumption is an important aspect of the power company's provision of high-quality services. In the face of the increasing proportion of residential users' electricity load, the use of reasonable and efficient data clustering algorithms to analyze users can help power companies. Provide more reasonable and personalized power supply solutions according to the characteristics of users, so that users can get a better experience.

However, the single original clustering algorithm has low clustering efficiency and poor clustering effect. For example, the K-means algorithm selects the initial clustering center randomly, which makes it easy to use the data set with a large amount of sample data. The clustering results fall into local optimum. Unable to determine the optimal number of clusters, researchers need to test one by one, resulting in low clustering efficiency. Therefore, the potential laws and electricity consumption characteristics in the user's electricity consumption data cannot be well reflected, so that the power companies cannot provide good support for the clustering of residential users.

SUMMARY OF THE INVENTION

The present invention provides a method for clustering daily load curves of electric residential users based on a compound clustering algorithm. The method clusters the load curves according to the real-time data collected in the smart meter of the electricity load, and then will have the same electricity consumption behavior. users get together.

The technical scheme adopted in the present invention is:

The clustering method of daily load curve of electric residential users based on compound clustering algorithm includes the following steps:

Step 1: Obtain the daily load data of electric residential users, the data contains P samples, and each sample has a data set matrix of Q time point attributes;

Step 2: Preprocess the daily load data of electric residential users to obtain the initial cluster;

Step 3: Perform dimensionality reduction processing on the initial cluster to obtain a dimensionality reduction cluster;

Step 4: Use clustering algorithm 1 to perform preliminary clustering on the dimensionality reduction cluster to obtain the initial cluster center;

Step 5: Use clustering algorithm 2 to cluster the initial clustering centers obtained by clustering algorithm 1, and use the clustering validity index to evaluate the clustering results, and finally obtain M clustering centers;

Step 6: Use the M clustering centers obtained in step 5 as the initial clustering centers of clustering algorithm 2 to cluster the data to obtain user groups with similar behaviors, and conduct behavioral characteristics for the obtained user groups with similar behaviors. analyze.

In the step 1, for P samples, each sample has a daily load data set of electric residential users with Q time-point attributes, which specifically includes:

P samples are resident user samples. Residents’ lives are mainly affected by factors such as seasonal changes, temperature changes, income levels, air conditioners, and the ownership rate of electric cookers. Unused factors will lead to different daily load curves; Q is each time point of the day. For the power consumption at this time point collected by the smart meter, the value of Q is determined according to the time interval at which the smart meter collects data.

In the step 2, the preprocessing includes missing value processing, data standardization, and data regularization processing;

Missing value processing, delete data with more missing values, and complete data with fewer missing values;

Data normalization, converting the original data linearization method to the range of [0, 1];

For data regularization, each attribute is subtracted from the mean corresponding to the attribute, and then divided by the corresponding variance of the attribute.

In the step 3, the dimensionality reduction process adopts PCA (Principal Component Analysis), that is, the principal component analysis method; the obtained dimensionality reduction cluster is a data set matrix of p samples and each sample has q attributes.

In the step 4, the clustering algorithm 1 is used to perform preliminary clustering on the dimensionality reduction clusters to obtain user groups with similar behaviors, which specifically includes: using the Mean-shift algorithm to cluster p samples in the data set into N categories, wherein, N is a positive integer.

In the step 5, the cluster centers obtained by the clustering algorithm 1 are clustered by using the clustering algorithm 2, and the clustering result is evaluated by using the clustering validity index, which specifically includes: using the K-means algorithm to obtain the mean-shift algorithm. In the range of the number of clusters N, cluster [2, N] respectively, where N is a positive integer, and use the Calinski-Harabasz (CH) index to evaluate the clustering results , select the result with the largest CH value, and finally get M cluster centers, where M is a positive integer in [2, N].

In the step 6, the obtained M cluster centers are used as the initial cluster centers of the K-means algorithm, and each sample in the data set, that is, each user or each record, is clustered, and finally M classes are obtained. User.

The present invention is a method for clustering the daily load curve of electric residential users based on a compound clustering algorithm. The daily load data of electric residential users is taken as the analysis object, and through multiple algorithm processes such as data preprocessing, data dimension reduction, and feature clustering, etc. Among them, the feature clustering algorithm is preferably a combination of the Mean-shift algorithm and the K-means algorithm. Cluster the huge and scattered daily load data into user groups with similar behaviors. By analyzing the clustered user groups, the managers of power enterprises can better predict the peaks and troughs of electricity consumption, provide more reliable methods for the management of power business, and provide better services for power customers.

Description of drawings

FIG. 1 is a flow chart of Embodiment 1 of the method of the present invention.

FIG. 2 is a flow chart of Embodiment 2 of the method of the present invention.

Detailed ways

Example 1:

The clustering method of daily load curve of electric residential users based on compound clustering algorithm includes the following steps:

Step 1: Obtain the daily load data of electric residential users, the data contains P samples, and each sample has a data set matrix of Q time point attributes;

Step 2: Preprocess the daily load data of electric residential users to obtain the initial cluster;

Step 3: Perform dimensionality reduction processing on the initial cluster to obtain a dimensionality reduction cluster;

Step 4: Use clustering algorithm 1 to perform preliminary clustering on the dimensionality reduction clusters to obtain N initial cluster centers;

Step 5: Use clustering algorithm 2 to cluster the initial clustering centers obtained by clustering algorithm 1, and use the clustering validity index to evaluate the clustering results, and finally obtain M clustering centers;

Step 6: Use the M clustering centers obtained in step 5 as the initial clustering centers of clustering algorithm 2 to cluster the data to obtain user groups with similar behaviors, and conduct behavioral characteristics for the obtained user groups with similar behaviors. analyze.

Example 2:

The clustering method of daily load curve of electric residential users based on compound clustering algorithm includes the following steps:

First, the daily load data of electric residential users is obtained, which contains a data set matrix with P samples and Q time point attributes for each sample.

In general, the data set passed by the power grid company's marketing system includes tens of thousands or more samples, each of which is a residential electricity user. With the popularization of smart meters, it has become very difficult to count the daily load data of each user's residential users. easy.

Then, perform preprocessing on the acquired daily load data of electric residential users to obtain an initial cluster, wherein, in this embodiment, the preprocessing process includes performing missing value processing, data standardization, and data regularization on the electric residential user daily load power consumption data As well as data dimensionality reduction, after the above processing, the initial cluster obtained is a dataset matrix of p samples and each sample has q attributes.

Specifically, the missing value processing is to delete the samples with few valid values, and complete the missing values of the samples with many valid values. Of course, when deleting attributes with few valid values, redundant attributes can be deleted together.

In the process of deleting samples, if n samples are deleted, p samples remain, where p=P-n. In addition, there are various ways to supplement the missing values. In this application, the average of the existing validity is taken as the filling value of the missing value. Those skilled in the art can choose other supplementary methods according to their non-uniformity and will not affect the subsequent analysis process.

The data normalization is specifically, the method of linearizing the original data is converted to the range of [0, 1], and the calculation formula of the maximum-minimum normalization is:

The method realizes equal scaling of the original data, wherein X _norm is the normalized data, X is the original data, and X _max and X _min are the maximum and minimum values of the original data set, respectively.

Data regularization is performed by subtracting the mean corresponding to the attribute from each attribute, and then dividing by the corresponding variance of the attribute. After standardization and regularization, the data of each attribute are clustered around 0 and the variance is 1, that is, the obtained sample data has zero mean and unit variance.

The data dimensionality reduction is specifically, using PCA (Principal Component Analysis), that is, principal component analysis method, to reduce the dimensionality of the data set, and obtain the processed dimensionality reduction cluster R.

In the process of dimensionality reduction, if the number of dimensionality reduction is q, then the dimensionality reduction cluster R after dimensionality reduction is a dataset matrix with p samples and each sample has q attributes.

Afterwards, the clustering algorithm is used to cluster the data in the data set R, and the user groups with similar electricity consumption behaviors are obtained, which includes, first, the Mean-shift algorithm is used to cluster the p samples in the data set into N categories , where N is a positive integer, and then the K-means algorithm is used to cluster the N cluster centers obtained by the Mean-shift algorithm, and within the range of the number of clusters N, cluster [2, N] respectively, where N is a positive integer, and the Calinski-Harabasz (CH) index is used to evaluate the clustering results, select the result with the largest CH value, and finally get M cluster centers, where M is a positive integer in [2, N], and finally, The obtained M cluster centers are used as the initial cluster centers of the K-means algorithm to perform clustering on each sample in the data set, that is, each user or each record, and finally obtain M classes of users.

In this embodiment, the specific process of the Mean-shift algorithm includes:

First, find any sample i from the data set, calculate the mean shift vector of the sample point and change the current center point position; then, shift the window and recalculate the probability density; finally converge to the maximum value of the probability density, Mean-shift Process the next object in the dataset R.

The more similar or equal the data attribute values in the same class, the greater the density of samples in this class. Each user group with similar behavior is called a class, and finally multiple similar classes are obtained, each class has its central sample point, and the user groups are named as class 1, class 2..., class N in turn.

In this embodiment, the specific process of the K-means algorithm includes:

The first step is to record the N central sample points as X={x ₁ , x ₂ ,...,x _N }, and find k points Y={y ₁ , y ₂ ,... ,y _k } as the cluster center, where k belongs to [2, N];

The second step is to calculate the distance from each point in cluster X to the k cluster center points in Y, and classify it into the class corresponding to the cluster center point with the smallest distance;

The third step is to recalculate each cluster center;

The fourth step, repeat the second and third steps until the position of the cluster center does not change;

The fifth step is to calculate the Calinski-Harabasz (CH) index corresponding to the k value;

The sixth step is to use for each k value from 2 to N, repeat the first step to the fifth step, select the k value corresponding to the maximum value of the Calinski-Harabasz (CH) index as K, and the corresponding cluster center as Z ={Z ₁ ,Z ₂ ,...,Z _K };

The seventh step is to calculate the distance from each point in the data set R to the K cluster center points in Z, and divide it into the class corresponding to the cluster center point with the smallest distance;

The eighth step, repeat the seventh and third steps until the position of the cluster center does not change;

The ninth step, output the clustering results.

In this embodiment, the specific calculation process of the Calinski-Harabasz (CH) indicator is as follows:

Among them, g represents the number of clusters, h represents the current class, trB(h) represents the trace of the inter-class dispersion matrix, and trW(h) represents the trace of the intra-class dispersion matrix. The larger the CH, the closer the class itself is, and the more dispersed the classes are, that is, the better the clustering results.

Claims (7)

1. The electric power resident user daily load curve clustering method based on the composite clustering algorithm is characterized by comprising the following steps of:

step 1: acquiring daily load data of power residents, wherein the data comprises P samples, and each sample has Q data set matrixes of time point attributes;

step 2: preprocessing daily load data of power resident users to obtain an initial cluster;

and step 3: performing dimensionality reduction on the initial cluster to obtain a dimensionality reduction cluster;

and 4, step 4: carrying out primary clustering on the dimensionality reduction cluster by adopting a clustering algorithm 1 to obtain an initial clustering center;

and 5: clustering the initial clustering centers obtained by the clustering algorithm 1 by adopting a clustering algorithm 2, and evaluating clustering results by using a clustering effectiveness index to finally obtain M clustering centers;

step 6: and 5, clustering the data by using the M clustering centers obtained in the step 5 as initial clustering centers of the clustering algorithm 2 to obtain user groups with similar behaviors.

2. The electric power resident user daily load curve clustering method based on the composite clustering algorithm as claimed in claim 1, wherein: in step 1, for P samples, each sample has Q time point attributes of the daily load data sets of the electric power residents, including: the P samples are resident user samples, the resident life is mainly influenced by factors such as seasonal changes, air temperature changes, people receiving level, air conditioning and electric cooking ownership, and different daily load curves can be caused by different factors; q is the power consumption at each time point every day collected by the intelligent electric meter at the time point, and the value of Q is determined according to the time interval of data collection of the intelligent electric meter.

3. The electric power resident user daily load curve clustering method based on the composite clustering algorithm as claimed in claim 1, wherein: in the step 2, the preprocessing comprises missing value processing, data standardization and data regularization processing;

missing value processing, namely deleting data with more missing values and completing data with less missing values;

data standardization, namely converting a method of raw data linearization into a range of [0, 1 ];

and (4) carrying out data regularization treatment, namely subtracting the mean value corresponding to each attribute from each attribute, and then dividing the mean value by the variance corresponding to the attribute.

4. The electric power resident user daily load curve clustering method based on the composite clustering algorithm as claimed in claim 1, wherein: in the step 3, pca (principal Component analysis), namely a principal Component analysis method, is adopted for the dimension reduction; the obtained dimensionality reduction cluster is a data set matrix with p samples and q attributes in each sample.

5. The electric power resident user daily load curve clustering method based on the composite clustering algorithm as claimed in claim 1, wherein: in the step 4, the dimensionality reduction cluster is subjected to preliminary clustering by adopting a clustering algorithm 1 to obtain a user group with similar behaviors, and the method specifically comprises the step of clustering p samples in a data set into N types by adopting a Mean-shift algorithm, wherein N is a positive integer.

6. The electric power resident user daily load curve clustering method based on the composite clustering algorithm as claimed in claim 1, wherein: in the step 5, clustering is performed on the clustering centers obtained by the clustering algorithm 1 by using the clustering algorithm 2, and clustering results are evaluated by using a clustering validity index, which specifically comprises: clustering N clustering centers obtained by the Mean-shift algorithm by adopting a K-means algorithm, respectively clustering [2 and N ] within a clustering number N range, wherein N is a positive integer, evaluating a clustering result by using a Calinski-Harabasz (CH) index, selecting a result with the largest CH value, and finally obtaining M clustering centers, wherein M is the positive integer in [2 and N ].

7. The electric power resident user daily load curve clustering method based on the composite clustering algorithm as claimed in claim 1, wherein: in the step 6, the obtained M clustering centers are used as initial clustering centers of the K-means algorithm, each sample in the data set, namely each user or each record, is clustered, and finally M classes of users are obtained.

Images