KR102284395B1 - Sales estimation system based on the amount of power usage and method thereof - Google Patents

Abstract

a data collection unit for obtaining power usage information of a store from the power usage measuring device, and obtaining store specification information and store sales information from an external input; a learning unit for machine learning the power usage information, the store specification information, and the store sales information; and an input unit for receiving the power usage information and the store specification information for a predetermined store from an external input. and a sales acquisition unit for inputting the power usage information and the store specification information for the predetermined store to the learning unit, and acquiring the store sales information for the predetermined store from the learning unit A prediction apparatus and method are disclosed.

Description

Electricity usage-based sales forecasting system and method

The present invention relates to a method and apparatus for predicting sales, and more particularly, to a method and apparatus for predicting sales of a store based on power usage.

A financial institution executes a loan based on the creditworthiness of the lender. When a financial institution executes a loan, a credit analysis that can confirm the possibility of default, such as the possibility of default by a lender, is an important management element of a financial institution. In order to objectively evaluate the creditworthiness of a lender, a financial institution utilizes business financial information including sales information of the lender. However, when it is difficult for a financial institution to obtain actual sales information of a lender due to the nature of the lender's business, it is difficult to accurately evaluate the creditworthiness of the lender. In particular, when a financial institution is an institution that lends a small amount of money and the lender is a small and medium-sized business operator, it is difficult for the micro-loan institution to secure the actual sales information of the small business operator, so it is difficult to accurately evaluate the creditworthiness of the small business operator.

Korean Patent Publication No. 10-2014-0018690

The present invention has been devised to respond to the above-described technical problem, and an object of the present invention is to substantially supplement various problems caused by limitations and disadvantages in the prior art, and sales of stores, etc. based on power consumption To provide a method and an apparatus for predicting , and to provide a computer-readable recording medium in which a program for executing the method is recorded.

According to an embodiment of the present invention, a sales forecasting method includes: acquiring power usage information of a store from a power usage measuring device by a data collection unit; acquiring store specification information and store sales information from an external input by the data collection unit; machine learning the power usage information, the store specification information, and the store sales information by a learning unit; receiving an identifier for a predetermined store from an external input by an input unit, and receiving the power usage information and the store specification information for the predetermined store from the data collection unit; inputting the power usage information and the store specification information for the predetermined store to the learning unit by the sales obtaining unit; and acquiring the store sales information for the predetermined store from the learning unit by the sales acquisition unit.

According to an embodiment of the present invention, the method further includes the step of estimating credit information for the predetermined store based on the store sales information for the predetermined store by a credit prediction unit.

According to an embodiment of the present invention, the power usage information is a measure of the power usage of a store periodically in a predetermined time unit.

According to an embodiment of the present invention, the store specification information includes a type of a store, a location of a store, an area of a store, and a type of home appliance used in the store.

According to an embodiment of the present invention, the step of machine learning includes calculating the effective business hours of a store in a predetermined time unit based on the power usage information.

According to an embodiment of the present invention, the step of machine learning may include calculating power usage per unit area of a shop based on the power usage information and the shop specification information; The power usage information is power usage per unit area.

According to an embodiment of the present invention, the step of machine learning may include implementing the power usage information of stores having the same or similar types of stores as a vector including 24 components for each time; cluster the same or similar vectors, and extract representative vectors representing each cluster; For each of the representative vectors, the effective business hours of the store are calculated in the predetermined time unit.

According to an embodiment of the present invention, the step of machine learning includes machine learning the power usage information, the store specification information, the effective business hours, and the store sales information.

According to an embodiment of the present invention, the method further includes calculating, by the learning unit, effective business hours for the predetermined store in a predetermined time unit based on the power usage information for the predetermined store.

According to an embodiment of the present invention, based on the power usage information, the store specification information, and the effective business hours for the predetermined store, predicting the store sales information for the predetermined store by the learning unit further steps.

In addition, according to an embodiment of the present invention, it includes a computer-readable recording medium in which a program for performing the method is recorded.

In addition, according to an embodiment of the present invention, the sales forecasting apparatus includes: a data collection unit for obtaining power usage information of a store from the power usage measuring device, and obtaining store specification information and store sales information from an external input; a learning unit configured to machine-lear the power usage information, the store specification information, and the store sales information; and an input unit for receiving an identifier for a predetermined store from an external input and receiving the power usage information and the store specification information for the predetermined store from the data collection unit. and a sales acquisition unit configured to input the power usage information and the store specification information for the predetermined store to the learning unit, and obtain the store sales information for the predetermined store from the learning unit.

According to the present invention, by providing a method and apparatus for estimating sales of stores, etc. based on power usage, even for loans where it is difficult for financial institutions to secure actual sales information, sales information can be predicted based on power usage, and accordingly creditworthiness can be estimated.

According to the present invention, by machine-learning power usage information, store specification information, and store sales information of stores belonging to the same or similar industry to a sales prediction device in advance, and a financial institution by inputting an identifier of a store to be evaluated in the sales prediction device , it is possible to obtain the learned sales information based on the power usage information and the store specification information of the evaluation target store.

Since the sales forecasting device according to the present invention calculates the effective business hours of a store based on power usage information, financial institutions can predict sales information based on the effective business hours even for loans where it is difficult to secure actual sales information, Credit rating can be estimated based on sales information. Therefore, according to the present invention, a financial institution can reduce the burden of work for determining creditworthiness when executing a loan, and can reduce costs by simplifying the credit review procedure.

1 is a block diagram of a sales forecasting apparatus according to an embodiment of the present invention.
2 is a representative vector illustration of power usage extracted to calculate effective business hours according to an embodiment of the present invention.
3 is an exemplary diagram illustrating effective business hours according to an embodiment of the present invention.
4 is a flowchart of a sales forecasting method according to an embodiment of the present invention.

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings. In the drawings, the same reference numerals refer to the same components, and the size of each component in the drawings may be exaggerated for clarity of description.

1 is a block diagram of a sales forecasting apparatus according to an embodiment of the present invention.

The sales prediction apparatus 100 according to an embodiment of the present invention includes a data collection unit 110 , a learning unit 130 , an input unit 150 , and a sales acquisition unit 170 .

The data collection unit 110 obtains the power usage information of the store from the power usage measuring device, and obtains the store specification information and the store sales information from an external input.

The power usage information is measured periodically by a predetermined time unit of power usage of a store. For example, the power usage measuring device periodically measures the power usage of a store every hour or 15 minutes. The data collection unit 110 acquires, from the power usage measuring device, information about the store's power usage, which the power usage measuring device periodically measures in a predetermined time unit. It is apparent to those skilled in the art that the power usage measuring device may be a separate device or the same device physically as the sales forecasting device 100 , and the operating entity of the power usage measuring device may be various operating entities including a power company.

The store specification information includes a type of store, a location of the store, an area of the store, and types of home appliances used in the store. When it is difficult to obtain detailed information of a store, average statistical information may be utilized. For example, the area of a store may be the average area of that type of store.

The learning unit 130 machine-learns the power usage information, the store specification information, and the store sales information for a store having store sales information. Although the learning unit 130 performs machine learning using an algorithm such as deep learning, it is obvious to those skilled in the art that the learning unit 130 may perform machine learning without being limited to a specific algorithm.

The learning unit 130 calculates the effective business hours of the store in units of a predetermined time based on the power usage information. For example, the learning unit 130 may calculate the effective business hours of the store in units of days or months. A detailed operation in which the learning unit 130 calculates the effective business hours of a store based on the power usage information will be described below in detail with reference to FIGS. 2 and 3 . The learning unit 130 machine-learns the power usage information, the store specification information, the effective business hours, and the store sales information.

The input unit 150 receives an identifier for a predetermined store from an external input, and receives the power usage information and the store specification information for the predetermined store from the data collection unit 110 based on the identifier. The predetermined store may be a store subject to evaluation by a financial institution.

The sales acquisition unit 170 inputs the power usage information and the store specification information for the predetermined store into the learning unit 130 , and obtains the store sales information for the predetermined store from the learning unit 130 . do. Specifically, the learning unit 130 calculates the effective business hours for the predetermined store in units of a predetermined time based on the power usage information for the predetermined store.

The learning unit 130 may further include a sales prediction unit. The sales prediction unit predicts the store sales information for the predetermined store based on the power usage information, the store specification information, and the effective business hours for the predetermined store.

The sales forecasting apparatus according to an embodiment of the present invention may further include a credit forecasting unit. The credit prediction unit estimates credit information for the predetermined store based on the store sales information for the predetermined store.

On the other hand, the learning unit 130 according to another embodiment of the present invention calculates the power usage per unit area of the shop based on the power usage information and the shop specification information. The learning unit 130 according to the present embodiment uses the power usage per unit area as the power usage information.

2 is a representative vector illustration of power usage extracted to calculate effective business hours according to an embodiment of the present invention.

The learning unit 130 calculates the effective business hours of the store in a predetermined time unit based on the power usage information.

The learning unit 130 represents power usage information of stores having the same or similar store type as a vector including 24 components for each time. Each component (hereinafter, 24 components) of the vector may be represented by normalizing power usage information for each hour by a value obtained by dividing the power usage information for each day by the total power usage of the day.

Equation 1 expresses a vector expressed by normalizing the hourly power usage information by a value divided by the total daily power usage.

The learning unit 130 clusters the same or similar vectors. The learning unit 130 clusters vectors using a K-means algorithm, etc., but it is obvious to those skilled in the art that the learning unit 130 is not limited to a specific algorithm and can use various clustering algorithms.

The learning unit 130 maintains a database including K representative vectors preprocessed to perform clustering. The K representative vectors represent the centroid vectors in each cluster. The learning unit 130 normalizes the time-specific power usage information and clusters the indicated vectors into K clusters composed of the same or similar vectors. The learning unit 130 divides n vectors, which are indicated by normalizing power usage information for each time, into K clusters that maximize the degree of cohesion between vectors in each cluster. To this end, specifically, the learning unit 130 clusters within each cluster in a manner that minimizes the variance of the distance difference from the center vector of the cluster.

Equation 2 represents a formula for grouping vectors s(t), which are represented by normalizing power usage information for each time.

시간 별 전력 사용량 정보를 정규화하여 나타낸 벡터

Vector representing the normalized power usage information for each hour

: 클러스터 i의 중심 벡터

: center vector of cluster i

: 클러스터 i

: cluster i

와 클러스터 내의 벡터와의 거리 차이의 분산인 E(s,i)를 최소화하는 방식으로 군집화하여 구성되며, 학습부(130)는 K 개의 클러스터로 군집화한다.In Equation 2, cluster i is the center vector

It is configured by clustering in a way that minimizes E(s,i), which is the variance of the distance difference between the and vectors in the cluster, and the learning unit 130 clusters them into K clusters.

According to an embodiment of the present invention, the learning unit 130 may additionally determine whether E(s,i) is equal to or less than a predetermined threshold value. In this case, the learner 130 may cluster cluster i into vectors having E(s,i) equal to or less than a predetermined threshold.

The learning unit 130 extracts representative vectors representing each cluster. The illustrated example shows a representative vector representing a specific group, expressing power usage information in units of 24 hours.

3 is an exemplary diagram illustrating effective business hours according to an embodiment of the present invention.

The learning unit 130 calculates the effective business hours of the store in a predetermined time unit for each of the representative vectors, such as the representative vector example shown in FIG. 2 . For example, the learning unit 130 calculates the effective business hours of the store for each of the representative vectors in various predetermined time units, such as by day of the week, by month, by morning/afternoon/evening time. As a representative vector of the illustrated example, the learning unit 130 may calculate the effective evening business hours of 8 hours from 16:00 to 24:00.

Through the calculation of the above-described effective business hours, the learning unit 130 machine-learns power usage information, store specification information, effective business hours, and store sales information for a store having store sales information.

In addition, the learning unit 130 calculates the effective business hours for the predetermined store in a predetermined time unit based on the power usage information for the predetermined store that is the evaluation target of the financial institution.

The learning unit 130 generates vectors representing the power usage information for the predetermined store. The learning unit 130 determines which cluster each of the vectors for the predetermined store belongs to among the clusters machine-learned by the learning unit 130, and a predetermined time unit for each of the vectors for the predetermined store Calculate the valid business hours of the store as

The learning unit 130 predicts the store sales information for the predetermined store based on the power usage information, the store specification information, and the effective business hours.

4 is a flowchart of a sales forecasting method according to an embodiment of the present invention.

In step 410 , the sales forecasting device 100 acquires the power usage information of the store from the power usage measuring device by the data collection unit 110 . The power usage information is measured periodically by a predetermined time unit of power usage of a store.

In step 420 , the sales forecasting apparatus 100 acquires store specification information and store sales information from an external input by the data collection unit 110 . The store specification information includes a type of store, a location of the store, an area of the store, and types of home appliances used in the store.

In step 430 , the sales prediction apparatus 100 machine-learns the power usage information, the store specification information, and the store sales information by the learning unit 130 for a store having store sales information. The sales forecasting apparatus 100 calculates the effective business hours of the store in a predetermined time unit based on the power usage information. The sales prediction apparatus 100 performs machine learning on the power usage information, the store specification information, the effective business hours, and the store sales information.

In step 440 , the sales prediction apparatus 100 receives an identifier for a predetermined store from an external input by the input unit 150 , and based on the identifier, the power usage information and the store specification information for the predetermined store is input from the data collection unit 110 . The predetermined store may be a store subject to evaluation by a financial institution.

In step 450 , the sales prediction apparatus 100 inputs the power usage information and the store specification information for the predetermined store to the learning unit 130 by the sales acquisition unit 170 . The learning unit 130 calculates the effective business hours for the predetermined store in units of a predetermined time based on the power usage information for the predetermined store. The learning unit 130 predicts the store sales information for the predetermined store based on the power usage information, the store specification information, and the effective business hours for the predetermined store.

In step 460 , the sales prediction apparatus 100 acquires the store sales information for the predetermined store from the learning unit 130 by the sales acquisition unit 170 .

Also, the sales prediction apparatus 100 estimates credit information for the predetermined store based on the store sales information for the predetermined store by the credit prediction unit.

On the other hand, the learning unit 130 according to another embodiment of the present invention calculates the power usage per unit area of the shop based on the power usage information and the shop specification information. The learning unit 130 according to the present embodiment uses the power usage per unit area as the power usage information.

Although preferred embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and equivalent other embodiments are possible. Accordingly, the true technical protection scope of the present invention should be defined by the appended claims.

For example, an apparatus according to an exemplary embodiment of the present invention may comprise a bus coupled to the respective units of the apparatus as shown, at least one processor coupled to the bus, the command, received and a memory coupled to the bus for storing a message or generated message and coupled to the at least one processor for performing instructions as described above.

In addition, the system according to the present invention can be implemented as computer-readable codes on a computer-readable recording medium. The computer-readable recording medium includes all types of recording devices in which data readable by a computer system is stored. The computer-readable recording medium includes a magnetic storage medium (eg, a ROM, a floppy disk, a hard disk, etc.) and an optical readable medium (eg, a CD-ROM, a DVD, etc.). In addition, the computer-readable recording medium is distributed in a network-connected computer system so that the computer-readable code can be stored and executed in a distributed manner.

Claims (21)

a data collection unit for obtaining power usage information of a store from the power usage measuring device, and obtaining store specification information and store sales information from an external input;
a learning unit for machine learning the power usage information, the store specification information, and the store sales information; and
an input unit for receiving an identifier for a predetermined store to be evaluated by a financial institution from an external input, and receiving the power usage information and the store specification information for the predetermined store from the data collection unit;
a sales acquisition unit inputting the power usage information and the store specification information for the predetermined store into the learning unit, and obtaining the store sales information for estimating a credit rating for the predetermined store from the learning unit; and
The learning unit normalizes the power usage information for each time calculated by Equation 1 below, and the vectors indicated by Equation 2 below are grouped into K clusters composed of the same vectors through Equation 2 below.
[Equation 1]

(here,

,

,

represents)
[Equation 2]

(here,

,

Vector representing the normalized power usage information for each hour,

: the centroid vector of cluster i,

: indicates cluster i) The method of claim 1,
and a credit prediction unit estimating credit information on the predetermined store, based on the store sales information for the predetermined store. The method of claim 1,
The power usage information is a sales forecasting device, characterized in that the store's power usage is periodically measured in a predetermined time unit. The method of claim 1,
The store specification information includes a type of store, a location of a store, an area of a store, and a type of home appliance used in the store. The method of claim 1,
The learning unit, based on the power usage information, sales forecasting apparatus, characterized in that for calculating the effective business hours of the store in a predetermined unit of time. 6. The method of claim 5,
the learning unit calculates power usage per unit area of a shop based on the power usage information and the shop specification information;
The power usage information is a sales forecasting device, characterized in that the power usage per unit area. The method of claim 5, wherein the learning unit
implementing the power usage information of stores having the same store type as a vector including 24 components for each time;
extracting representative vectors from among the vectors;
Sales forecasting apparatus, characterized in that for calculating the effective business hours of the store in the predetermined time unit for each of the representative vectors. 6. The method of claim 5,
The learning unit is a sales prediction device, characterized in that the machine learning the power usage information, the store specification information, the effective business hours, and the store sales information. The method of claim 1,
The learning unit, based on the power usage information for the predetermined store, sales forecasting apparatus, characterized in that for calculating the effective business hours for the predetermined store in a predetermined time unit. 10. The method of claim 9,
The learning unit further comprises a sales prediction unit for predicting the store sales information for the predetermined store, based on the power usage information for the predetermined store, the store specification information, and the effective business hours Sales forecasting device. acquiring power usage information of the store from the power usage measuring device by the data collection unit;
acquiring store specification information and store sales information from an external input by the data collection unit;
machine learning the power usage information, the store specification information, and the store sales information by a learning unit;
receiving, by an input unit, an identifier for a predetermined store to be evaluated by a financial institution from an external input, and receiving the power usage information and the store specification information for the predetermined store from the data collection unit;
inputting the power usage information and the store specification information for the predetermined store to the learning unit by the sales obtaining unit; and
Acquiring the store sales information for estimating the credit rating for the predetermined store by the sales acquisition unit from the learning unit;
The learning unit normalizes the power usage information for each hour calculated by the following Equation 1 below, and the vectors represented by the following Equation 2 are used to cluster the vectors into K clusters composed of the same vectors.
[Equation 1]

(here,

,

,

represents)
[Equation 2]

(here,

,

Vector representing the normalized power usage information for each hour,

: the centroid vector of cluster i,

: indicates cluster i) 12. The method of claim 11,
Based on the store sales information for the predetermined store by a credit prediction unit, the sales forecasting method further comprising the step of estimating credit information for the predetermined store. 12. The method of claim 11,
The power usage information is a sales forecasting method, characterized in that the store's power usage is periodically measured in a predetermined time unit. 12. The method of claim 11,
The store specification information includes a type of store, a location of a store, an area of a store, and a type of home appliance used in the store. 12. The method of claim 11,
The machine learning may include calculating an effective business hour of a store in a predetermined time unit based on the power usage information. 16. The method of claim 15,
The machine learning may include calculating power usage per unit area of a shop based on the power usage information and the shop specification information;
The power usage information is a sales forecasting method, characterized in that the power usage per unit area. 16. The method of claim 15, wherein the machine learning step
implementing the power usage information of stores having the same type of store as a vector including 24 components for each hour;
extracting representative vectors from among the vectors;
The sales forecasting method, characterized in that for each of the representative vectors, calculating the effective business hours of the store in the predetermined time unit. 16. The method of claim 15,
The machine learning may include machine learning the power usage information, the store specification information, the effective business hours, and the store sales information. 12. The method of claim 11,
Based on the power usage information for the predetermined store, the method for predicting sales further comprising the step of calculating the effective business hours for the predetermined store in a predetermined time unit by the learning unit. 20. The method of claim 19,
The method further comprising the step of predicting the store sales information for the predetermined store by the learning unit based on the power usage information, the store specification information, and the effective business hours for the predetermined store How to forecast sales. A computer-readable recording medium in which a program for performing the method according to any one of claims 11 to 20 is recorded.

Images