KR20210130395A - METHOD FOR ANALYZING RESIDENTIAL LOAD CHARACTERISTICS USING k-MEANS CLUSTERING AND PRINCIPAL COMPONENT ANALYSIS - Google Patents

Abstract

The present invention relates to a residential load characteristic analyzing method using k-means clustering and principal component analysis. The method comprises the following steps of: classifying, by a control unit of a residential load characteristic analysis apparatus, residential users through k-means clustering; extracting, by the control unit, a principal consumer representing each cluster by applying principal component analysis to a classified cluster; and analyzing, by the control unit, characteristics of each cluster by using questionnaire responses before collection of data of the extracted principal consumer.

Description

Residential load characteristic analysis method using k-means clustering and principal component analysis

The present invention relates to a residential load characteristic analysis method using k-means clustering and principal component analysis. More specifically, after classifying residential users through k-means clustering, principal component analysis is applied to the classified clusters to apply the principal component analysis to each cluster. It relates to a residential load characteristic analysis method using k-means clustering and principal component analysis, which extracts major consumers who can represent

Recently, the power system has been developed based on the demand side, and various demand response programs are applied to utilize demand resources.

At this time, the residential load is 22% of the total load in the United States in 2015 and 20% in Ontario, Canada in 2016, showing a significant proportion.

Therefore, in order to utilize the residential load demand resource, it is essential to design an appropriate residential load demand response program for residential consumers.

In designing such a residential load demand response program, it is necessary to reflect the characteristics of the residential consumer and the load included in the residential environment (eg, home appliances, etc.), so an understanding of this must be preceded.

At this time, many studies using clustering and principal component analysis have been conducted to analyze the characteristics of residential consumer load, and most studies perform dimensionality reduction of the entire data in advance using principal component analysis before k-means clustering. That is, k-means clustering is performed after principal component analysis.

However, when k-means clustering is performed after principal component analysis as described above, since the effect of principal component analysis on individual clusters cannot be confirmed, there is a problem in that the load characteristic analysis for each cluster is somewhat inaccurate. In addition, when the questionnaire response data is used to analyze the load characteristics for each group, there is a problem in that it is difficult to conduct an additional questionnaire when a new user appears.

Therefore, after classifying residential users through k-means clustering, principal component analysis is applied to the classified clusters to extract major consumers who can represent each cluster, and a method for performing residential load characteristic analysis for each cluster. This is a necessary situation.

The background technology of the present invention is disclosed in Korean Patent Publication No. 10-2017-0053193 (published on May 16, 2017, a data compression system and method based on k-means clustering for a wireless image sensor network).

According to one aspect of the present invention, the present invention was created to solve the above problems, and after classifying residential users through k-means clustering, principal component analysis is applied to the classified clusters to represent each cluster. The purpose of this is to provide a residential load characteristic analysis method using k-means clustering and principal component analysis, which extracts the main consumers who can .

A residential load characteristic analysis method using k-means clustering and principal component analysis according to an aspect of the present invention comprises: classifying, by a control unit of a residential load characteristic analysis apparatus, residential users through k-means clustering; extracting, by the control unit, principal component analysis to the classified clusters, which can represent each cluster; and analyzing, by the control unit, the characteristics of each group by using the questionnaire responses of the extracted main consumers prior to data collection.

)을 K 개의 군집으로 분류하는 비지도 기계학습 기법이며, 상기 데이터 집합은 복수의 군집 중 가장 거리가 짧고 SSE(sum of squared error) 값이 작은 군집에 배정되는 것을 특징으로 한다.In the present invention, the k-means clustering is a multidimensional N data set (

) is an unsupervised machine learning technique for classifying K clusters, and the data set is characterized in that the data set is assigned to a cluster having the shortest distance and a small sum of squared error (SSE) value among a plurality of clusters.

In the present invention, when the center of the cluster is recalculated based on the assigned data, the controller reclassifies the data set based on the recalculated center of the cluster, and there is no change in the classified cluster The clustering process is repeated until , and the clustering process is characterized in that the SSE value, which is a cost function, is minimized.

In the present invention, in the k-means clustering, classification is performed with a given number of clusters regardless of the optimal number of clusters, and the optimal number of clusters is selected using the 'Elbow method' to determine the appropriate optimal number of clusters in advance, In the graph showing the value of the cost function calculated according to the total number of clusters (K), if the total number of clusters (K) is increased by more than a specified number of clusters (K), the marginal utility of the decrease in SSE occurs, causing the graph to bend. It is characterized in that the number of clusters is selected as the optimal number of clusters.

사이의 상호 관계를 반영한 주성분(

)으로 하여금 기존 데이터의 차원을 감소시켜 간결하게 나타내는 것임을 특징으로 한다.In the present invention, in the principal component analysis, the control unit is a variable using Equation 2 below.

The principal component that reflects the interrelationship between

) to reduce the dimension of the existing data and represent it concisely.

(Equation 2)

은 loading vector로서 주성분(

)을 이루는 기존 변수의 계수이며, 이 값이 클수록 해당 변수

가 주성분(

)에 더 중요한 변수임을 나타낸다.here

is the main component (

) is the coefficient of the existing variable that makes up the

is the main component (

) is a more important variable.

)의

값을 내림차 정렬하고, 기준 등수(

) 이내

를 하나라도 가지는 소비자인 변수(

)를 전부 채택하되, 상기 기준 등수(

)는 아래의 수학식 3을 이용하여 산출하는 것을 특징으로 한다.In the present invention, in the step of extracting the main consumers who can represent each cluster, the control unit, in order to extract only the main consumers representing the characteristics of each cluster, the main component (

)of

Sort values in descending order,

) Within

A variable that is a consumer with at least one (

), but the above standard rank (

) is calculated using Equation 3 below.

(Equation 3)

)는

군집 내 주성분의 기준 등수이며,

는

군집의 총 주성분 개수이고,

는 해당 주성분으로 설명할 수 있는 전체 분산의 비율로서, 이를 반영하여 기준 등수를 산출하고, 누적합(

)은

크기를 고려하지 않고 단순 차등을 두어 기준 등수를 계산했을 때 추출되는 최대 소비자 수이다. Here, the standard rank (

)Is

It is the standard rank of the principal component in the cluster,

Is

is the total number of principal components in the cluster,

is the ratio of the total variance that can be explained by the corresponding principal component, and reflects this to calculate the standard rank, and

)silver

This is the maximum number of consumers extracted when the standard rank is calculated based on a simple difference without considering the size.

In the present invention, in the step of analyzing the characteristics of each cluster using the questionnaire responses at the time before the data collection of the extracted main consumers, the questionnaire questions and the response options for the questionnaire response are pre-specified in the form of a table do it with

According to one aspect of the present invention, after classifying residential users through k-means clustering, the present invention applies principal component analysis to the classified clusters to extract main consumers who can represent each cluster, and residential load characteristics Through the analysis method, it is possible to perform an analysis of the residential load characteristics by cluster.

1 is a flowchart for explaining a residential load characteristic analysis method using k-means clustering and principal component analysis according to an embodiment of the present invention.
2 is a graph illustrating a cost function value calculated according to the total number of clusters (K) in FIG. 1 .
3 is a graph showing power usage data of consumers classified by the optimal number of clusters in FIG. 1 .
4 is an exemplary diagram showing an average pattern graph for each cluster consisting of consumers extracted by principal component analysis in FIG. 1 .
FIG. 5 is an exemplary view in which only the survey response ratio in which the difference by group is prominent among the question responses in FIG. 1 is shown as a heat map.

Hereinafter, an embodiment of a residential load characteristic analysis method using k-means clustering and principal component analysis according to the present invention will be described with reference to the accompanying drawings.

In this process, the thickness of the lines or the size of the components shown in the drawings may be exaggerated for clarity and convenience of explanation. In addition, the terms to be described later are terms defined in consideration of functions in the present invention, which may vary according to the intention or custom of the user or operator. Therefore, definitions of these terms should be made based on the content throughout this specification.

Hereinafter, the residential load characteristic analysis method using k-means clustering and principal component analysis according to an embodiment of the present invention is performed through a residential load characteristic analysis device (not shown), and although not shown in the drawings, the residential load The characteristic analysis apparatus (not shown) may be implemented as a computer or a server including a control unit (eg, CPU, MPU, etc.).

1 is a flowchart for explaining a residential load characteristic analysis method using k-means clustering and principal component analysis according to an embodiment of the present invention. classify (S101), and then apply principal component analysis to the classified cluster (S102), extract major consumers that can represent each cluster (S103), and respond to a questionnaire at the time before data collection of the extracted consumers can be used to analyze the characteristics of each cluster.

를 K 개의 군집으로 분류하는 비지도 기계학습 기법이며, 아래 수학식 1은 비용 함수인 SSE(sum of squared error)로서 분류할 데이터와 각 군집의 중심 사이의 유클리드 거리를 의미한다. In this case, the k-means clustering is a multidimensional N data set.

is an unsupervised machine learning technique that classifies , into K clusters. Equation 1 below is a cost function, sum of squared error (SSE), which means the Euclidean distance between the data to be classified and the center of each cluster.

More specifically, the data set is assigned to the cluster with the shortest distance and the smallest SSE value among clusters. At this time, the cluster centroid is recalculated based on the assigned data, the data set is reclassified based on the recalculated cluster centroid, and the process is repeated until there is no change in the classified cluster. Finally, the clustering process proceeds in the direction of minimizing the SSE value, which is a cost function.

는 군집(

)의 중심이며,

는 데이터와 군집 사이의 유클리드 거리를 의미하며, SSE 값이 작을수록 군집이 잘 분류되었다고 판단된다.here

is the cluster (

) is the center of

is the Euclidean distance between the data and the cluster, and the smaller the SSE value, the better the cluster is classified.

Since the k-means clustering proceeds with classification with the given number of clusters regardless of the number of optimal clusters, it is important to determine the optimal number of clusters in advance. to select the optimal number of clusters.

2 is a graph showing a cost function value calculated according to the total number of clusters (K) in FIG. 1, and when the total number of clusters (K) is increased by a certain number or more, the limit of the reduction amount of the sum of squared error (SSE) When utility occurs, the graph is greatly bent, and the number of clusters in the bent portion is selected as the optimal number of clusters.

3 is a graph showing power usage data of consumers classified by the optimal number of clusters in FIG. 1 .

3 (a) shows the monthly average power usage pattern of all consumers, and FIG. 3 (b) shows the average pattern of the overall power usage, and the x-axis and y-axis are 30-minute unit time and power usage ( kWh).

At this time, the reason why the average patterns are similar to each other in (b) of FIG. 3 is that consumers who do not show comparatively group-specific characteristics are grouped together, and in this case, principal component analysis is used to extract the main consumers.

사이의 상호 관계를 반영한 주성분(

)으로 하여금 기존 데이터의 차원을 감소시켜 보다 간결하게 나타내는 기법이다.The principal component analysis is a variable using Equation 2 below.

The principal component that reflects the interrelationship between

) to reduce the dimension of the existing data and represent it more concisely.

은 loading vector로서 주성분(

)을 이루는 기존 변수의 계수이며, 이 값이 클수록 해당 변수

가 주성분(

)에 더 중요한 변수임을 나타낸다.here

is the main component (

) is the coefficient of the existing variable that makes up the

is the main component (

) is a more important variable.

On the other hand, for load characteristic analysis, only the main consumers representing the characteristics of each cluster are extracted.

)의

값을 내림차 정렬하고, 기준 등수(

) 이내

를 하나라도 가지는 소비자인 변수(

)를 전부 채택한다.For example, in order to extract only the main consumers representing the characteristics of each cluster, the main component (

)of

Sort values in descending order,

) Within

A variable that is a consumer with at least one (

) are all accepted.

)는 아래의 수학식 3을 이용하여 산출할 수 있다.The above standard rank (

) can be calculated using Equation 3 below.

)는

군집 내 주성분의 기준 등수이며,

는

군집의 총 주성분 개수이고,

는 해당 주성분으로 설명할 수 있는 전체 분산의 비율로서, 이를 반영하여 기준 등수를 산출하고, 누적합(

)은

크기를 고려하지 않고 단순 차등을 두어 기준 등수를 계산했을 때 추출되는 최대 소비자 수이다. Here, the standard rank (

)Is

It is the standard rank of the principal component in the cluster,

Is

is the total number of principal components in the cluster,

is the ratio of the total variance that can be explained by the corresponding principal component, and reflects this to calculate the standard rank, and

)silver

This is the maximum number of consumers extracted when the standard rank is calculated based on a simple difference without considering the size.

Table 1 below is a table comparing the number of consumers (before principal component analysis) and the number of consumers extracted by principal component analysis according to this embodiment.

As shown in Table 1, since the number of consumers extracted by the principal component analysis has only a small number of consumers compared to other clusters, a separate consumer extraction process in Cluster 6 is not performed.

association One 2 3 4 5 6 gun existing 1021 535 624 1313 277 14 3784 After principal component analysis 75 25 51 57 32 254

4 is an exemplary diagram showing an average pattern graph for each cluster consisting of consumers extracted by principal component analysis in FIG. 1 .

Referring to FIG. 4 , compared to the average pattern of power usage by cluster of FIG. 3 in which only k-mean clustering is performed, the average pattern graph for each cluster that has undergone principal component analysis has a relatively different pattern from other cluster patterns. can be visually confirmed. This is because only major consumers are extracted through principal component analysis to indicate the characteristics of the cluster, and relatively insignificant consumers are removed from the cluster.

For reference, after principal component analysis, the Euclidean distance is used as a similarity index to confirm that the number of consumers is more characterized than before and is distinguished from other cluster patterns.

After normalizing the average power usage for each cluster using the min-max normalization of Equation 4 below, the Euclidean distance is calculated and similarity is compared.

는 정규화 이후 군집별 평균 전력 사용량,

는 정규화 이전 군집별 평균 전력 사용량이며,

,

는 각각 평균 전력 사용량의 최소값과 최대값을 의미한다.here

is the average power usage per cluster after normalization,

is the average power usage per cluster before normalization,

,

denotes the minimum and maximum values of the average power consumption, respectively.

Table 2 below is a table showing the Euclidean distance for the average wattage pattern between each cluster before and after the principal component analysis. The higher the value, the greater the difference between the average wattage patterns, indicating that the similarity decreases. The average distance increased from 0.778 before principal component analysis to 1.2058 after principal component analysis, confirming a decrease in the similarity between clusters due to pattern characterization for each cluster.

association Before principal component analysis (Fig. 3 (b)) After principal component analysis (Fig. 4) One 2 3 4 5 6 One 2 3 4 5 6 One - 0.608 1.007 0.347 0.408 0.781 - 0.710 1.719 0.665 1.134 1.432 2 0.608 - 1.474 0.676 0.884 1.335 0.710 - 2.059 0.699 1.388 1.746 3 1.007 1.474 - 0.821 0.644 0.826 1.719 2.059 - 1.491 0.879 1.080 4 0.347 0.676 0.821 - 0.304 0.863 0.665 0.699 1.491 - 1.005 1.429 5 0.408 0.884 0.644 0.304 - 0.688 1.134 1.388 0.879 1.005 - 0.651 6 0.781 1.335 0.826 0.863 0.688 - 1.432 1.746 1.080 1.429 0.651 -

In addition, the characteristics of each cluster can be analyzed using the questionnaire responses at the time before the data collection of the consumers extracted by the principal component analysis.

In this case, the questionnaire questions and response options for analyzing the characteristics of each cluster are shown in Table 3 below.

Table 3 below is divided into questions indicating the characteristics of residential consumers or load characteristics in the home.

Residential Consumer Characteristics Characteristics of loads in the home 1) Family composition
I live alone
All people in my home are over 15
Both adults and children under 15
2) Chief income earner
an employee
self-employed (w/ or w/o employees)
Retired
Unemployed (seeking work or not)
carer : looking after relative family
3) Number of bedrooms: 1/2/3/4/5++
4) Internet access: Y/N 1) Heating resource
Electricity / gas / oil
solid fuel / renewable
2) Heat water resource
Electricity/gas/central heating
oil / solid fuel
3) Number of Appliances: 1/2/3++
tumble dryer / dish washer
electric heater/stand alone freezer
TV (less than or greater than 21 inch)
Desktop/laptop/Games consoles etc.

At this time, when analyzing the characteristics of residential consumers, the proportion of consumers in the corresponding cluster who responded to each item option with respect to the number of consumers included in the corresponding cluster is calculated and used.

For example, among consumers of cluster 1 in FIG. 5 , a consumer who responded to the 'I live alone' option corresponds to 44% of all consumers in cluster 1. In addition, in the case of the heating and hot water fuel survey in the load characteristics within the home, since heating or hot water can be used using multiple resources, consumers who use resources are considered. Also, in Table 3, “3) Number of Appliances” is first The average number of units used for each device is calculated for each device, and the ratio of the cluster to the total average number of units used is calculated.

FIG. 5 is an exemplary diagram in which only the survey response ratio in which the group-specific difference among the question responses is prominent is shown as a heat map in FIG. 1 .

Referring to FIG. 5 , a row is a response of each questionnaire and a column is a cluster, and a cluster with a high response rate is displayed in a dark color.

In FIG. 5, in Cluster 1, the proportion of households with children under 15 years old is low, the number of “Number of appliances” also has a high proportion of single people with a high proportion of 0 or 1, and since it is a single household, the average household appliance holding amount is It is very small, and the ratio of electricity consumption of heating resource and hot water resource is also low.

In FIG. 5, cluster 2 has a relatively large number of household appliances, and a high ratio of power consumption for heating and hot water.

In FIG. 5, cluster 3 has a large number of household appliances such as TVs, desktops, game consoles, etc., and a high proportion of Internet access, and the proportion of responses that answered that there are children under 15 with cluster 2 is similar, but the cluster 3, the daytime electricity consumption is higher, suggesting that the household has a child who does not attend school.

In FIG. 5 , clusters 4 and 5 are clusters having a high retirement ratio, but the proportion of electric fuel used in cluster 5 is larger.

In FIG. 5, in Cluster 6, the proportion of simple employees and single self-employed (Self employed(w/o employees)) is low, while the proportion of self employed(w/emplyees) with employees is very large, and the power consumption during the day is particularly high. This is because they are self-employed at home, which means they own a lot of home appliances due to self-employment and record the highest average usage.

As described above, in this embodiment, after k-means clustering, only important users are extracted through principal component analysis and characteristics of each cluster can be analyzed. It shows the characteristics of clusters that are more likely to participate in the demand response program because of their ease of movement.

For example, the application of demand response in clusters 2, 3, and 5 seems to be relatively effective, but since the proportion of electric fuel in cluster 3 is low, the participation rate in clusters 2 and 5 is expected to be higher, and in cluster 6, electricity consumption due to self-employment Since the time is fixed, it can be seen that although the amount of power used to move the load is large, it is relatively difficult to move the load. In addition, depending on which cluster a new consumer is classified into, consumer characteristics can be inferred without additional survey response data. It has the effect of being able to do it.

As described above, the present invention has been described with reference to the embodiment shown in the drawings, but this is merely exemplary, and various modifications and equivalent other embodiments are possible therefrom by those of ordinary skill in the art. will understand the point. Therefore, the technical protection scope of the present invention should be defined by the following claims. Also, the implementations described herein may be implemented as, for example, a method or process, an apparatus, a software program, a data stream, or a signal. Although discussed only in the context of a single form of implementation (eg, only as a method), implementations of the discussed features may also be implemented in other forms (eg, in an apparatus or a program). The apparatus may be implemented in suitable hardware, software and firmware, and the like. A method may be implemented in an apparatus such as, for example, a processor, which generally refers to a computer, a microprocessor, a processing device, including an integrated circuit or programmable logic device, or the like. Processors also include communication devices such as computers, cell phones, portable/personal digital assistants ("PDAs") and other devices that facilitate communication of information between end-users.

Claims (7)

classifying, by the control unit of the residential load characteristic analysis apparatus, residential users through k-means clustering;
extracting, by the control unit, principal component analysis to the classified clusters, which can represent each cluster; and
and analyzing, by the control unit, the characteristics of each cluster using the questionnaire responses at the time before the data collection of the extracted main consumers.
The method of claim 1, wherein the k-means clustering is
Multidimensional N data sets (

) is an unsupervised machine learning technique that classifies K clusters,
The data set is a residential load characteristic analysis method using k-means clustering and principal component analysis, characterized in that the data set is assigned to a cluster having the shortest distance and a small sum of squared error (SSE) value among a plurality of clusters.
3. The method of claim 2,
The center of the cluster is recalculated based on the assigned data,
the control unit,
Classifying the data set again based on the recalculated center of the cluster,
Repeat the clustering process until there is no change in the classified cluster,
The clustering process is a residential load characteristic analysis method using k-means clustering and principal component analysis, characterized in that it proceeds in the direction of minimizing the SSE value, which is a cost function.
The method of claim 1, wherein the k-means clustering is
Classification proceeds with the given number of clusters regardless of the optimal number of clusters,
In order to determine the optimal number of clusters in advance, the optimal number of clusters is selected using the 'Elbow method',
In the graph showing the value of the cost function calculated according to the total number of clusters (K), if the total number of clusters (K) is increased by more than a specified number of clusters (K), the marginal utility of the decrease in SSE occurs, causing the graph to bend. A residential load characteristic analysis method using k-means clustering and principal component analysis, characterized in that the number of clusters is selected as the optimal number of clusters.
According to claim 1, wherein the principal component analysis,
the control unit,
Variables using Equation 2 below

The principal component that reflects the interrelationship between

) by reducing the dimension of the existing data and representing it concisely. A residential load characteristic analysis method using k-means clustering and principal component analysis, characterized in that.
(Equation 2)

here

is the main component (

) is the coefficient of the existing variable that makes up the

is the main component (

) is a more important variable.
According to claim 1, wherein in the step of extracting the main consumers who can represent each cluster,
the control unit,
In order to extract only the main consumers representing the characteristics of each cluster,
chief ingredient(

)of

Sort values in descending order,

) Within

A variable that is a consumer with at least one (

) are accepted, but
The above standard rank (

) is a residential load characteristic analysis method using k-means clustering and principal component analysis, characterized in that it is calculated using Equation 3 below.
(Equation 3)

Here, the standard rank (

)Is

It is the standard rank of the principal component in the cluster,

Is

is the total number of principal components in the cluster,

is the ratio of the total variance that can be explained by the corresponding principal component, reflecting this to calculate the standard rank, and

)silver

This is the maximum number of consumers extracted when the standard rank is calculated based on a simple difference without considering the size.
The method of claim 1,
In the step of analyzing the characteristics of each cluster using the questionnaire responses at the time before the data collection of the extracted main consumers,
The questionnaire questions and response options for the questionnaire response are,
A residential load characteristic analysis method using k-means clustering and principal component analysis, characterized in that it is specified in advance in a table form.

Images