CN111915116A - Electric power resident user classification method based on K-means clustering - Google Patents

Abstract

The invention discloses a K-means clustering-based electricity residential user classification method, comprising the following steps: determining regional electricity residential user load data; preprocessing the data; Classify the data; based on the clustering results, analyze the load characteristics of each type of electricity residential users; analyze the regional power system load characteristics; compare and analyze the power system load curve and each type of electricity residential user load curve to determine the electricity consumption type of each type of electricity residential user . The invention adopts the K-means clustering algorithm to measure the similarity of the power consumption load data of the electric residential users, and perform cluster analysis on the users, so that the characteristics of the power consumption load curve of the users can be well measured and distinguished, and the user load can be realized. The clustering of curves and the analysis of load characteristics, and the K-means clustering algorithm is scalable and efficient in the clustering analysis of electric residential users' loads.

Description

A Classification Method of Electricity Residential Users Based on K-means Clustering

technical field

The invention relates to the technical field of power system analysis, in particular to a K-means clustering-based method for classifying electricity residential users.

Background technique

Under the mode of opening up the electricity sales side, there will be diversified electricity sellers in the electricity market, and they will compete in the electricity retail business. By subdividing users in the power market, and understanding the load characteristics of different types of users in detail, as well as the differentiated needs of different users, the electricity sellers in the electricity market can scientifically arrange electricity purchase and sales plans, and formulate diversified and diversified The electricity price package system provides users with more options, which is beneficial to the main body of electricity sales to attract more users and seize the electricity sales market.

At present, the method theory for realizing user segmentation based on clustering method has been relatively mature, such as fuzzy clustering method and Kohonen neural network (KNN) clustering method. However, the fuzzy clustering method is susceptible to the interference of subjective factors, resulting in local differences in the clustering results, and the algorithm is relatively complex; the practical application of the KNN clustering method shows that it cannot meet the requirements of load forecasting clustering for the identification of curve shape details; With the deepening of the reform of the electricity sales side, the residential users in the electricity market have reached a considerable scale, and it is no longer feasible to carry out one-to-one load characteristic analysis on these users. It is very necessary to distinguish, and there is no technology involved in this field at present.

SUMMARY OF THE INVENTION

In order to solve the above-mentioned technical problems, the technical solution adopted in the present invention is to provide a method for classifying electricity residential users based on K-means clustering, including the following steps:

Determine the load data of regional electricity residential users;

preprocessing the data;

Classify the preprocessed data according to the K-means clustering algorithm;

Based on the clustering results, analyze the load characteristics of each type of electricity residential users; analyze the regional power system load characteristics;

Compare and analyze the load curve of the power system and the load curve of each type of electricity residential user, and determine the electricity consumption type of each type of electricity residential user.

In the above method, it also includes pushing a reasonable electricity sales plan according to the electricity consumption type of each type of electricity residential user.

In the above method, the data preprocessing includes the following steps:

A1. Delete invalid records in load data; including negative data and 0;

A2. Delete the data with more continuous missing;

A3. Handling of individual missing values;

For the first and last data gaps, the trend ratio method is used to fill in;

For the gaps in the intermediate data, the non-neighboring mean generation method is used for filling.

In the above method, the deletion of continuously missing data is specifically:

With a time interval of 15 minutes, there are 96 load values per day, and samples with consecutively missing more than 30 load values are deleted.

In the above method, the described data is classified according to the K-means clustering algorithm; including the following steps:

Normalize the preprocessed load data and form a load data sample set;

Set the number of clusters, and classify the load data sample set according to the K-means clustering algorithm.

In the above method, the analyzing the load characteristics of electric residential users includes:

Based on the clustering results, the load characteristics of each type of users are analyzed, and the daily load characteristic curve is used to describe the load characteristics of electric residential users.

In the above method, the normalization processing of the preprocessed load data includes:

The preprocessed load data is normalized by the extreme value method, and the specific formula is as follows:

In the formula, X ^* is the value after normalization, X is the load data after preprocessing, and _Xmax is the maximum value in the load data after preprocessing.

In the above method, the electricity consumption types of the electricity residential users include: peak-peak electricity users, partial peak-peak electricity users and stable electricity users.

In the above method, the number of clusters K is 3-10.

In the above method, the number of clusters K is preferably 5.

The invention adopts the K-means clustering algorithm to measure the similarity of the power consumption load data of the electric residential users, and perform cluster analysis on the users, so as to measure and distinguish the characteristics of the power consumption load curve of the users well, and realize the user load. Curve clustering and load characteristic analysis, and the K-means clustering algorithm is scalable and efficient in the clustering analysis of electric residential users' loads.

Description of drawings

Fig. 1 is the flow chart provided by the present invention;

Fig. 2 is a schematic diagram of a data preprocessing process flow provided by the present invention;

Fig. 3 is the first kind of load cluster center curve diagram in the case provided by the present invention;

Fig. 4 is the second kind of load cluster center curve diagram in the case provided by the present invention;

Fig. 5 is the third type of load clustering center curve diagram in the case provided by the present invention;

Fig. 6 is the fourth type of load clustering center curve diagram in the case provided by the present invention;

Fig. 7 is the fifth kind of load clustering center curve diagram in the case provided by the present invention;

FIG. 8 is a daily load characteristic curve diagram of the regional power system in one week in the case provided by the present invention.

Detailed ways

The invention adopts the K-means clustering method to efficiently realize the subdivision of electricity residential users, and analyzes the load characteristics of different types of users in detail. Package; on the other hand, it can guide electricity residential users to use electricity reasonably, such as avoiding electricity consumption during peak system load, so as to optimize the electricity consumption behavior of electricity residential users and reduce the fluctuation of system load. The present invention will be described in detail below with reference to the specific embodiments and the accompanying drawings.

As shown in FIG. 1 , the present invention provides a method for classifying electricity residential users based on K-means clustering, which is characterized in that it includes the following steps:

S1. Determine the load data of regional electricity residential users;

When collecting load data, try to ensure data integrity and time continuity.

S2. Preprocess the data;

The method of data preprocessing is as follows, and the specific process is shown in Figure 2:

During data preprocessing, the lack or error of load data will affect the correctness of the clustering results. Therefore, the load data should be preprocessed before clustering. The methods are as follows:

(1) Delete invalid records in load data

The load value required in this embodiment is the forward active load, so the collected data are not lower than 0. First, delete the data with negative values in the table; secondly, in order to make the clustering results representative, delete the data with all 0s;

(2) Delete too many consecutive missing data

With a time interval of 15 minutes, there are 96 load values per day, and samples with consecutively missing more than 30 load values are deleted.

(3) Handling of individual missing values

Before clustering, the data needs to be supplemented and corrected. In this embodiment, for the data gaps at the beginning and end, the trend ratio method is used to fill in; for the intermediate data gaps, the non-neighboring mean generation method is used to fill in. The so-called non-neighboring mean generation method, that is, for a sequence that is not equidistant in time, or a sequence that is equidistant, but has holes after removing outliers, the data on both sides of the hole is averaged to construct new data to fill the gap. hole.

S3. Classify the preprocessed data according to the K-means clustering algorithm; specifically, the following steps are included:

S31. Normalize the preprocessed data to form a load data sample set D={x ₁ ,x ₂ ,...,x _m };

Since the loads of different residential electricity users are quite different, in order to better group users with similar electricity consumption behaviors into one group and ensure the effectiveness of this data training, it is necessary to normalize the preprocessed data. , and then analyze. This embodiment adopts the extreme value method to normalize the preprocessed data. This method can realize the proportional expansion and contraction of the original data. The specific formula is as follows:

In the formula, X ^* is the value after normalization, X is the load data after preprocessing, and _Xmax is the maximum value in the load data after preprocessing. After the preprocessed load data is normalized, it will be limited to the interval [0,1].

S32. Set the number of clusters, and classify the load data sample set D={x ₁ ,x ₂ ,...,x _m } according to the K-means clustering algorithm;

K-means clustering is a relatively simple clustering algorithm in the clustering algorithm to perform related clustering on the load curve.

The number of clusters is set by the user. For the load data sample set, K points are randomly selected as the center point of the cluster; secondly, the sample point closest to the center point is found according to the relevant distance metric, and it is classified into the nearest cluster. , update the center of the cluster at the same time; perform the above operations iteratively until the data no longer changes or the center point of the cluster no longer changes, and the divided cluster C={C ₁ ,C ₂ ,...,C _k } is output at this time, And the loss function of clustering is as follows:

In the formula, μ _i is the mean vector of cluster C _i , and its specific expression is as follows:

In this embodiment, when using the K-means clustering algorithm to perform cluster analysis, the number of clusters needs to be specified first; if the number of clusters is too large, the shape of the load curve will be similar, and the representativeness of the classification will not be strong; If the number of clusters is too small, there will be less load characteristic results and incomplete analysis. Therefore, the number of clusters K ranges from 3 to 10; in order to make the classification more representative, after many experiments, it is found that when the number of When K=5, the classification result is more reasonable.

S4. Based on the clustering results, analyze the load characteristics of each type of electricity residential users; specifically include:

According to the clustering result in step S3, analyze the load characteristics of electric power residential users, and the adopted method is:

Usually, the analysis of the load characteristics of electricity residential users is realized by calculating the load characteristic index. The load characteristic index takes time as the unit, and different time limits can be divided into different load characteristic indexes, such as annual load characteristic index, monthly load characteristic index and daily load characteristic index, and can be further subdivided into curve types in different load characteristic indexes Metrics, comparative metrics, and descriptive metrics. In the embodiment, in order to obtain the result of the load characteristic analysis quickly and accurately, the curve-type index in the daily load characteristic index, such as the daily load characteristic curve, will be used to analyze the load characteristic of the electric residential user.

S5. Analyze the load characteristics of the regional power system; the method for analyzing the load characteristics of the regional power system in this embodiment is consistent with the analysis of the load characteristics of each type of electricity residential user in step S4, and the daily load characteristic curve is used to describe the load characteristics of the regional power system.

S6. Comparatively analyze the load curve of the power system and the load curve of each type of electricity residential user, and determine the electricity consumption type of each type of electricity residential user, including peak-consuming electricity users, some peak-consuming electricity users and stable electricity users.

Based on the above classification results, this embodiment designs differentiated electricity consumption packages and service strategies for different types of user groups under the market environment where the electricity sales side is open, so as to plan ahead for market-oriented electricity sales, specifically including:

S7. According to the electricity consumption type of each type of electricity residential user, push a reasonable electricity sales plan.

The present embodiment is described below through a specific case.

This case takes a certain area as an example. According to the idea of the above embodiment, firstly collect 2498 daily load curves of electric residential users, the sampling interval is 1 hour, and the daily load data contains 24 active power points in total; Data padding, removal of invalid data samples and normalization. After data processing, there are 30 invalid samples and 2468 valid samples in the sample data.

The normalized 2468 samples were input into the K-means clustering model. When using K-means clustering for cluster analysis, it is first necessary to specify a K value of 5 for cluster analysis, and the number of cases included in each cluster category is shown in Table 1.

Table 1. Number of cases in each cluster

Based on the clustering results, the determined cluster center curves and their corresponding differentiated power consumption characteristics of users are as follows.

(1) The number of users in the first category of results is 810, accounting for 32.8% of the valid samples. As shown in Figure 3, the daily load curve of this type of users is relatively stable, basically stable within 24 hours of the day, and the fluctuation range is not large.

(2) The number of users of the second type of results is 20, accounting for 0.8% of the valid samples. As shown in Figure 4, the daily load curve of such users is "two peaks and one valley", and the electricity load is low during the day and high at night.

(3) The number of users in the third category is 955, accounting for 38.7% of the valid samples. As shown in Figure 5, the daily load curve of such users is "two peaks, one level and one valley", and the peak electricity consumption occurs at around 11:00 and 20:00 respectively. The daily load curve of this type of user belongs to the typical peak load curve, and the proportion of electricity consumption during the trough period is small, so the basic load of this type of user has a large load regulation ability. Willingness to participate in peak shifting management.

(4) The number of users in the fourth category is 85, accounting for 3.4% of the valid samples. As shown in Figure 6, the daily load curve of this type of users is relatively stable, similar to the first type, and the fluctuation range is not large, but the difference from the first type is that the load shows a downward trend after eight o'clock.

(5) The number of users in the fifth category is 598, accounting for 24.3% of the number of valid samples. As shown in Figure 7, the daily load curve of this category of users is “one peak, one level and one valley”, and its load characteristics are similar to those of the third category of users. resemblance. This type of load has great potential for load regulation.

After the typical load curve of residential users is extracted, the power consumption characteristics of relative individuals can be compared with the power consumption characteristics of groups. This case is compared with the collected local regional total load curve, and a classification rule is proposed to facilitate the design of retail electricity price packages for future electricity sellers. This case mainly starts from measuring the guiding potential of the user's electricity consumption strategy, compares the user's load curve with the regional total load curve, and re-categorizes. For the main body of electricity sales, this classification method is helpful for the classification management of users, realizes the optimization of the power load structure, and saves the operation cost.

When carrying out load analysis on the power system, the daily load of the power system in the region within a week is extracted to draw a weekly load curve, as shown in Figure 8. It can be seen that the shape of the daily load curve in this region is in the form of "three peaks and one valley". It can show three peaks in the morning, middle and evening. The morning peak appears at 9:00-11:00 and lasts for a short time; the afternoon peak appears between 14:00-18:00 and lasts for a long time; The peak appears between 19:00-21:00 and the duration is shorter; the trough appears between 1:00-6:00 and the duration is longer.

By comparing and analyzing the daily load characteristic curve of residents and the daily load characteristic curve of the system, from the different perspectives of the individual curve shape and the group curve shape, the user load characteristic analysis considering the system load characteristic can better meet the needs of the main body of electricity sales. To formulate retail electricity price packages and provide basis for value-added services, residential users can be divided into three categories, namely, peak-consuming electricity users, partial peak-consuming electricity users, and stable electricity users.

(1) Peak power users: This type of users mainly corresponds to the third and fifth types of users mentioned above. The time interval of the peak period of the curve is highly consistent with the regional total load curve. The error is within 30 minutes. Compared with the afternoon peak, there is a big gap between the maximum value of the evening peak.

(2) Some peak-consuming electricity users: This type of users mainly corresponds to the second type of users. The peak time interval of the curve is basically consistent with the regional total load curve, and there is a behavior of avoiding peak electricity consumption relative to the overall load.

(3) Users of stable electricity consumption: This type of users mainly corresponds to the first and fourth types of users, and the fluctuation of electricity consumption in a day is not large.

The example results show that the K-means clustering algorithm is used to measure the similarity of the electricity load data of the electric residential users, and the cluster analysis of the users can be used to measure and distinguish the characteristics of the electricity load curve of the users. The clustering and load characteristic analysis of the user load curve, and the K-means clustering algorithm has scalability and high efficiency in the cluster analysis of the electricity residential user load. In the market environment where the electricity sales side is open, it is of great practical significance to improve the competitiveness of enterprises to design differentiated electricity consumption packages and service strategies for different types of user groups, and plan ahead for market-oriented electricity sales.

The present invention is not limited to the above-mentioned best embodiment, and anyone should know that structural changes made under the inspiration of the present invention, and all technical solutions that are the same or similar to the present invention, fall within the protection scope of the present invention.

Claims (10)

1. A method for classifying users of electric power residents based on K-means clustering is characterized by comprising the following steps:

determining regional electric power resident user load data;

preprocessing the data;

classifying the preprocessed data according to a K-means clustering algorithm;

analyzing the load characteristics of each type of electric power resident users based on the clustering result; analyzing the load characteristics of the regional power system;

and comparing and analyzing the load curve of the power system and the load curve of each type of power residential users, and determining the power consumption type of each type of power residential users.

2. The method as claimed in claim 1, further comprising pushing a reasonable electricity selling plan according to the electricity consumption type of each type of electric residents.

3. The method of claim 1, wherein the pre-processing the data comprises:

a1, deleting invalid records in the load data; negative value data and 0;

a2, deleting data with more continuous missing;

a3, processing individual missing values;

for the vacancy of the head-end and the tail-end data, completing by a trend proportion method;

and for the vacancy of the intermediate data, completing the vacancy by a non-adjacent mean generation method.

4. The method according to claim 3, wherein the deleting of the data with a large number of consecutive deletions is specifically:

at 15min intervals, 96 load values per day, samples with continuously missing more than 30 load values were deleted.

5. The method of claim 1, wherein the processed data is classified according to a K-means clustering algorithm; the method comprises the following steps:

carrying out normalization processing on the preprocessed load data, and forming a load data sample set;

and setting the clustering number, and classifying the load data sample set according to a K-means clustering algorithm.

6. The method according to claim 1, wherein said analyzing electrical residential customer load characteristics comprises:

and analyzing the load characteristics of each type of users based on the clustering result, and describing the load characteristics of the electric power residential users by daily load characteristic curves.

7. The method of claim 5, wherein normalizing the pre-processed load data comprises:

carrying out normalization processing on the preprocessed load data by adopting an extreme method, wherein a specific formula is as follows:

in the formula, X^*Is normalized value, X is preprocessed load data, X_maxIs the maximum value in the preprocessed load data.

8. The method of claim 1 wherein said electrical residential consumer electricity usage pattern comprises: the peak-meeting power utilization type users, part of the peak-meeting power utilization type users and the stable power utilization type users.

9. The method according to claim 5, wherein the number of clusters K is 3-10.

10. Method according to claim 9, characterized in that the cluster number K is preferably 5.

Images