KR20230103171A - Method and Apparatus for Predicting Demand for Public Goods Using Machine Learning Models - Google Patents

Abstract

A method and apparatus for predicting demand for public goods using a machine learning model are disclosed. The demand prediction method of the present invention analyzes the demand pattern of users of public goods, predicts the user's rental/return pattern through an artificial intelligence-based algorithm, and arranges public goods according to the predicted rental/return pattern. In addition, the demand prediction method more accurately predicts and provides user rental/return patterns by feeding back newly obtained user rental/return patterns through arrangement request information.

Description

Method and Apparatus for Predicting Demand for Public Goods Using Machine Learning Models}

The present invention relates to a method and apparatus for predicting demand for public goods using a machine learning model, and more specifically, by using information about users' use of public goods, training a machine learning model, and using the machine learning model It is about the technology of predicting the demand for public goods.

As the sharing economy expands to various fields, sharing platforms are emerging by combining the sharing economy with various public goods (eg, means of transportation). Public bicycles, which can be said to be the most representative of transportation sharing platforms, are short-distance public transportation equipped with an operating system that rents and returns from the state or local government, and is a service that allows individuals to use bicycles at low cost or free of charge in a short period of time. am.

Recently, due to COVID-19, it has been found that the use rate of using public goods has increased compared to the use rate of using public transportation such as buses and subways. In addition, when looking at each time zone, it was found that the usage rate for commuting to and leaving work on weekdays increased. It is analyzed that public goods have been recognized as a representative means of transportation that can be safely used than buses and subways in the COVID-19 situation. According to these analysis results, the number of users using sharing platforms for public goods is also steadily increasing.

However, in the case of public goods, it may frequently occur that it is difficult for users to use public goods at a desired time. For example, if there are no pre-arranged bicycles at the time the user wants to use or the place the user wants to use, it may take a long time to additionally place them. In order to solve the user's inconvenience, the user may request additional placement of public goods from a desired rental place through a smartphone application. However, it is inconvenient for users to wait indefinitely because not only immediate placement is not possible, but also information about additional placement time and time required for placement is not provided.

Accordingly, there is a need for a technology that collects information about users' use of public goods in advance, predicts the user's demand for using public goods, and enables the public goods to be appropriately supplied.

The present invention performs a machine learning-based data analysis algorithm in consideration of the access environment related to the day of the week, usage time, location of the rental office, etc., which affect the rental and return of public goods by users, thereby helping users rent and return public goods. Methods and devices for analyzing and learning patterns are provided.

The present invention predicts demand for public goods by analyzing and learning rental/return patterns from users' use information on public goods through a data analysis algorithm, and supplies public goods to rental offices according to demand so that users can use public goods without inconvenience. It provides a method and apparatus to enable.

The present invention increases the accessibility of public goods by adjusting the arrangement time at which public goods are supplied to rental stations in real time by feeding back the rental/return patterns of users according to the time of each rental station based on the insufficient supply information on public goods collected in the database. It provides a method and apparatus for doing so.

Demand forecasting method according to an embodiment of the present invention comprises the steps of: generating public goods usage information having a specific time period by adjusting the sync of time for public goods usage information of users received from a wireless terminal; Classifying public goods use information having a specific time period in response to each item in consideration of an access environment affecting rental and return of public goods; and determining rental/return patterns of users related to demand for public goods by applying the classified public goods use information to a machine learning model.

In the step of generating public goods usage information having a specific time period according to an embodiment of the present invention, the synchronization of time for rental information and return information included in users' public goods usage information for public goods placed in a rental office is adjusted to specify a specific time period. Public goods use information having a time period may be generated.

Public goods use information according to an embodiment of the present invention is a result of rental or return by a user for a specific period at each rental station where the public goods are placed, the usage time of the public goods, the place where the public goods are used, and the information of the public goods. It may include details including any one or more of the types.

In the step of classifying public goods use information having a specific time period according to an embodiment of the present invention, the specific time period is set in response to at least one item among season, holiday, rental station, day of the week, and time zone based on the access environment. It is possible to classify public goods use information.

In the step of determining the rental/return patterns of users according to an embodiment of the present invention, the classified public goods use information is applied to a machine learning model using a machine learning-based data analysis algorithm, and each item uses the public goods for time t. identifying the number of users; Generalizing a rental/return pattern appearing to users by analyzing the number of users identified for each item and the mutual relationship between each item; and determining rental/return patterns of users related to the demand for the public goods from the generalized rental/return patterns of users.

The step of generalizing the rental/return pattern appearing to users according to an embodiment of the present invention is the step of analyzing a specific condition affecting the identification of the number of users identified for each item by using the correlation between each item. ; and generalizing a rental/return pattern that appears to users based on the specific conditions; can include

In the step of determining rental/return patterns of users according to an embodiment of the present invention, when arrangement request information due to lack of supply of public goods is received from a wireless terminal, the number of users according to time for each rental station is based on the arrangement request information. The rental/return patterns of the determined users may be re-determined by feeding back the rental/return patterns.

In the step of determining rental/return patterns of users according to an embodiment of the present invention, rental/return patterns of users related to demand for public goods that can be supplied to each rental office for each time slot of the next day may be determined.

In the demand forecasting device including a processor according to an embodiment of the present invention, the processor generates public goods usage information having a specific time period by adjusting the sync of time for public goods usage information of users received from a wireless terminal. In response to each item in consideration of the access environment affecting the rental and return of public goods, public goods use information having a specific time period is classified, and the classified public goods use information is applied to a machine learning model to determine the demand for public goods. It is possible to determine the rental/return pattern of users.

The processor according to an embodiment of the present invention may generate public goods usage information having a specific time period by controlling a sync of time for rental information and return information included in users' public goods usage information for public goods placed in a rental office. there is.

The processor according to an embodiment of the present invention may classify public goods use information having a specific time period in response to at least one item among seasons, holidays, rental locations, days of the week, and time zones based on the access environment.

The processor according to an embodiment of the present invention identifies the number of users who have used the public good for time t for each item by applying the classified public good use information to a machine learning model using a machine learning-based data analysis algorithm, and The number of users identified by item and the mutual relationship between each item are analyzed to generalize the rental/return pattern that appears to users, and from the generalized rental/return pattern of users, the rental/return pattern of users related to the demand for the public goods. can determine

The processor according to an embodiment of the present invention analyzes a specific condition affecting the identification of the number of users identified for each item by using the correlation between each item, and the rental/rental / Return patterns can be generalized.

When the processor according to an embodiment of the present invention receives arrangement request information due to lack of supply of public goods from a wireless terminal, based on the arrangement request information, the processor according to an embodiment of the present invention feeds back the rental/return patterns of users according to time for each rental station, and determines the number of users determined. Rental/return patterns can be re-determined.

The processor according to an embodiment of the present invention may determine rental/return patterns of users regarding demand for public goods that can be supplied to each rental office for each time slot of the next day.

In the method for predicting demand for public goods according to an embodiment of the present invention, identifying public goods use information of users; determining parameters modeling the public good use information using a trained machine learning model; modeling the public good use information according to the parameters; and predicting a demand for the public good based on the modeling result. Demand forecasting method comprising a.

Public goods use information according to an embodiment of the present invention is a result of rental or return by a user for a specific period at each rental station where the public goods are placed, the usage time of the public goods, the place where the public goods are used, and the information of the public goods. It may include details including any one or more of the types.

A parameter according to an embodiment of the present invention may refer to a parameter required for modeling the public good use information with a discrete probability distribution.

The demand forecasting method according to an embodiment of the present invention further includes identifying the number of users who have used the public good for a certain period of time based on the public good use information, and the determining of the parameters includes the user using the public good. A parameter modeling the number of s as a discrete probability distribution can be determined.

In the modeling step according to an embodiment of the present invention, the public good use information may be modeled using the parameters to follow a discrete probability distribution.

According to one embodiment of the present invention, by performing a machine learning-based data analysis algorithm in consideration of the access environment related to the day of use, usage time, location of the rental office, etc. that affect the rental and return of public goods by users, You can analyze and learn the rental/return patterns of users.

According to one embodiment of the present invention, by predicting demand for public goods by analyzing and learning rental/return patterns from users' use information on public goods through a data analysis algorithm, public goods are supplied to rental offices according to demand to use public goods. Users can use public goods without any inconvenience.

According to one embodiment of the present invention, based on the arrangement request information due to lack of supply of public goods collected in the database, the arrangement time at which public goods are supplied to the rental station is reduced by feeding back the rental/return patterns of users over time for each rental station. It can be adjusted in real time to increase the accessibility of public goods.

According to one embodiment of the present invention, by using public goods usage information having a 24-hour period in synchronization with the time at which the arrangement of public goods is requested, it can be used to predict the demand for public goods on the next day.

1 is a diagram illustrating a demand forecasting device according to an embodiment of the present invention.
2 is a diagram illustrating an example of a process of collecting and transmitting data between a wireless terminal and a demand forecasting device according to an embodiment of the present invention.
3 is a diagram showing an example of a demand forecasting process by the demand forecasting device according to an embodiment of the present invention.
4 is a diagram showing an example of usage information according to an embodiment of the present invention.
5 is a flowchart illustrating a method of predicting the number of users of a public good according to an embodiment of the present invention.
6 is a graph showing experimental results for demand prediction according to an embodiment of the present invention.
7 is a graph showing experimental results for demand prediction according to an embodiment of the present invention.
8 is a flowchart illustrating a demand forecasting method according to an embodiment of the present invention.
9 is a flowchart illustrating a demand forecasting method according to another embodiment of the present invention.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. However, since various changes can be made to the embodiments, the scope of the patent application is not limited or limited by these embodiments. It should be understood that all changes, equivalents or substitutes to the embodiments are included within the scope of rights.

Terms used in the examples are used only for descriptive purposes and should not be construed as limiting. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as "include" or "have" are intended to designate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, but one or more other features It should be understood that the presence or addition of numbers, steps, operations, components, parts, or combinations thereof is not precluded.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art to which the embodiment belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and unless explicitly defined in the present application, they should not be interpreted in an ideal or excessively formal meaning. don't

In addition, in the description with reference to the accompanying drawings, the same reference numerals are given to the same components regardless of reference numerals, and overlapping descriptions thereof will be omitted. In describing the embodiment, if it is determined that a detailed description of a related known technology may unnecessarily obscure the gist of the embodiment, the detailed description will be omitted.

1 is a diagram showing a demand forecasting device 101 according to an embodiment of the present invention.

Referring to FIG. 1 , the present invention relates to a technique of training a machine learning model using public goods usage information of users of public goods and predicting demand for the public goods using the machine learning model. The demand forecasting device 101 according to an embodiment may include a processor. The processor of the demand forecasting device 101 may perform a demand forecasting method according to an embodiment.

More specifically, the demand forecasting device 101 may receive public goods usage information of users for public goods from users' wireless terminals 102 or a server of a sharing platform. For example, the demand forecasting device 101 may be a server of a shared platform. Alternatively, the demand forecasting device 101 may receive public goods use information of users of public goods from a server of a sharing platform.

The wireless terminal 102 may apply for or reserve a rental for using public goods disposed in the rental offices 106 . The wireless terminal 102 may apply for or reserve the public goods 103 through a sharing platform operated as a web or an app according to a user's request. For example, the wireless terminal 102 may apply or make a reservation to rent the public goods 103 disposed at one of the rental offices 106 through the demand prediction device 101 .

The demand forecasting device 101 may check whether public goods are actually rented at the rental office according to the application or reservation of the wireless terminal 102 . For example, when a rental completion button displayed on the screen of the wireless terminal 102 is clicked, the wireless terminal 102 may transmit a request message for the completion button to the demand forecasting device 101 . The demand forecasting device 101 may check whether public goods are rented by transmitting an ACK signal for the request message to the wireless terminal 102 .

The demand forecasting device 101 may store, in the database 105, public goods usage information and users' arrangement request information for the public goods 103 using the public goods 103 through the wireless terminal 102 by users for each place. . Here, the information on the use of public goods may include details of use of public goods as a result of rental and return by users for one day at each rental place. The details of use of public goods may include at least one of a usage time of the public goods, a place where the public goods are used, and a type of the public goods. For example, the demand forecasting device 101 may identify the number of users who use the public good 103 in a specific place for a certain period of time based on the public good use information 201 .

The wireless terminal 102 may transmit placement request information to the demand forecasting device 101 in real time through the placement request function of the application when the user cannot rent due to lack of public goods through the application. For example, the wireless terminal 102 may transmit, to the demand forecasting device 101, arrangement request information for requesting the arrangement of additional public goods in a time zone and place that the user intends to use. Since a request for placement of a public good may be requested when a public good is not present in a place desired by users, it may not include details of actual use of the public good.

For example, placement request information applied when a public good is not present in a place desired by users may be used to train a machine learning model. As another example, placement request information applied when a public good is not present in a place desired by users may be used for training of a machine learning model with a higher weight than information about users using the public good.

2 is a diagram illustrating an example of a process of collecting and transmitting data between a wireless terminal and a demand forecasting device according to an embodiment of the present invention.

Referring to FIG. 2 , the wireless terminal 102 may transmit public property use information 201 and arrangement request information 202 related to rental and return to the demand forecasting device 101 . When transmitting the public goods usage information 201, the wireless terminal 102 may periodically transmit the public goods usage information 201 to the demand forecasting device 101 according to a predetermined time. For example, the wireless terminal 102 may generate public property use information 201 including all details of a user renting or returning a public property through an application installed in the wireless terminal 102 . The wireless terminal 102 may periodically transmit public goods use information 201 to the demand forecasting device 101 .

In addition, when transmitting the arrangement request information 202, the wireless terminal 102 may transmit the arrangement request information 202 to the demand forecasting device 101 in real time as soon as requested by the user. As an example, the wireless terminal 102 may transmit the placement request information 202 to the demand forecasting device 101 in real time through the placement request function of the application when the user cannot rent due to lack of public goods through the application. there is.

The demand forecasting device 101 may store and manage public property use information 201 and arrangement request information 202 received from the wireless terminal 102 in the first database 503 . The demand forecasting device 101 may synchronize the time of the public goods use information 201 . The public goods use information 201 may be information indicating a result of rental and return by a user for a day at each rental place. In addition, the public goods use information 201 may include the number of users who rent or return public goods placed in the rental office. The demand forecasting device 101 may store and manage public goods use information in the second database 504 at a specific point in time.

For example, the demand forecasting device 101 synchronizes the time of public goods usage information 201 collected from 0:00 to 24:00 according to a 24-hour system, and thus public goods usage information having a 24-hour period corresponding to each rental station. (201) can be created. The demand forecasting device 101 transmits the public goods use information 201 generated at the end of the day (24:00) to the second database 504, so that it can be stored and managed in the second database 504. there is.

The demand forecasting device 101 may synchronize the time of the arrangement request information 202 stored in the first database 503 according to the 24-hour system, and store and manage the information in the second database 504. At this time, the demand forecasting device 101 may transmit the placement request information 202 to the second database 504 in real time.

The demand forecasting device 101 may analyze the public goods use information 201 stored in the second database 504 in consideration of the public goods access environment. The access environment may indicate a condition for determining whether it is easy for users to access each rental station. For example, the access environment may be environmental conditions such as season, day of the week, time zone, location of the rental office, and the like.

At this time, the demand forecasting device 101 can largely analyze the public goods use information using ① an analysis method using public goods use information or ② an analysis method using public goods use information and arrangement request information 202 .

① Analysis method using public goods use information

The demand forecasting device 101 may estimate a user's rental/return pattern for public goods according to time at each rental station by applying public goods use information to a machine learning-based data analysis algorithm in consideration of an access environment. For example, various algorithms such as artificial intelligence/machine learning and deep learning may be used for the data analysis algorithm. The demand forecasting device 101 may supply public goods to each rental station through a user's rental/return pattern. The demand forecasting device 101 may perform scheduling so that the supply of public goods is preferentially provided to rental offices in which supply of public goods is urgent. A rental office in which the supply of public goods is urgent may refer to a rental office in which demand for public goods is not available based on the usage status of public goods by time zone and the current time. Such an analysis method may be to identify the public goods that can be supplied by each rental office on the next day, that is, on the next day (0:00) of the end of the day (24:00).

② Analysis method using public goods use information and placement request information (202)

The demand forecasting device 101 may estimate a user's access pattern to a specific rental place through the placement request information 202 due to lack of public goods. A specific rental office may be a rental office requested to supply additional public goods because it is impossible to lend due to a lack of public goods. Here, the impossibility of renting public goods in a specific rental office may mean that the number of users using a specific rental office has increased more than usual, or that new users who have not used a specific rental office have been introduced.

Accordingly, when the placement request information 202 is received, the demand forecasting device 101 may newly estimate a user's access pattern to a specific rental place. The demand forecasting device 101 may use the newly estimated user's access pattern as feedback on public goods supply, and re-estimate the user's rental/return pattern for public goods for each rental station through the corresponding feedback. Such an analysis method may be used to identify dynamically changing demand for public goods in real time, and accordingly, more smoothly supply public goods.

Eventually, the demand forecasting device 101 infers and uses the user's rental/return pattern for public goods through analysis and feedback of the machine learning-based public goods use information and arrangement request information 202, thereby providing a more suitable public goods supply service for the user. can provide.

In addition, the demand forecasting device 101 Through the method proposed in the present invention, the public goods use information of users who use public goods is applied to a data analysis algorithm to use it as data to predict the dispatch time for supplying public goods, as well as to use it as feedback for the predicted dispatch time to provide more accurate information. Dispatch time can be provided.

The demand prediction device 101 is more extended in that it analyzes and predicts the user's access (demand) pattern in public goods of all sharing platforms by using the user's access (demand) pattern predicted from the placement request information as feedback. to be applied in the field.

3 is a diagram showing an example of a demand forecasting process by the demand forecasting device according to an embodiment of the present invention.

Referring to FIG. 3 , the demand prediction device 101 predicts the next day's demand 302 for a public good 304 using a trained machine learning model 301, and provides public good use information 202 for the next day. The machine learning model 301 may be re-trained based on

The demand 302 predicted by the demand forecasting device 101 may mean the demand 302 for the public good 304 at an arbitrary place and by time zone. As an example, referring to (a) of FIG. 3, as a result of predicting demand 302, at a specific time, the demand 302 for public goods in zone A is determined to be '1', and public goods in zone B Demand for food 302 is determined as '2', and demand for food in zone C is determined as '5', showing that public goods 304 are actually arranged according to demand 302 determined for each zone. it is a drawing

Referring to (b) of FIG. 3 , public goods use information on the public goods 304 may be collected for each zone. The collected public goods use information may include details 305 of users using the public goods 304 for each place. The demand forecasting device 101 may train the machine learning model 301 again based on the collected public goods use information.

For example, if the demand 302 is incorrectly predicted, the public good 304 may not be present at a place desired by users. Accordingly, the placement request information requested when the public goods 304 is not present in a place desired by users may be used to train the machine learning model 301 . That is, the demand forecasting device 101 may train the machine learning model 301 again based on the placement request information for the public good 304 .

For example, since the placement request information due to the absence of the public good 304 indicates the degree to which the demand 302 prediction is wrong, the demand forecasting device 101 determines a loss function based on the number of placement request information, and determines the loss The machine learning model 301 can be re-trained so that this is minimized.

4 is a diagram showing an example of usage information according to an embodiment of the present invention.

(a) of FIG. 4 may illustrate a process 401 in which users apply to a server of a sharing platform to use a public property in a specific place through a wireless terminal. FIG. 4(b) may illustrate a process 402 in which a user transmits a placement request to a server of a sharing platform through a wireless terminal when there is no public good in a place desired by users.

According to an embodiment, a wireless terminal may request one or more of public goods usage information using public goods by users and arrangement request information for public goods by users to a server of a sharing platform. According to the present invention, the database of the server of the sharing platform may be updated whenever users request the use of public goods in a specific place to the sharing platform through a wireless terminal. The server of the shared platform may be the demand forecasting device of the present invention.

5 is a flowchart illustrating a demand forecasting method according to an embodiment of the present invention.

The flow chart of FIG. 5 may be a process of predicting information on the number of users accessing public goods using a machine learning-based data analysis algorithm.

In step 501, the demand forecasting device may collect public goods use information through an application of a wireless terminal possessed by a user using each rental station.

In step 502, the demand forecasting device may generate public goods use information having a 24-hour period from information on users' use of public goods collected from wireless terminals for a certain period of time according to the 24-hour system. Public goods use information having a 24-hour period may include information about rental and return information for one day at each public goods rental office.

At this time, when the arrangement request information due to the lack of public goods is received in step 508, the demand forecasting device adds public goods supply request information according to the previously predicted pattern of previous users, that is, in sync with the time of the rental/return pattern. Public goods use information having a new 24-hour period is formed. The newly formed information on the use of public goods can be used in data analysis and prediction algorithms through a data classification process.

In step 503, the demand forecasting device may classify public goods use information having a 24-hour period for each item in consideration of an access environment that affects rental and return of public goods by users. In detail, among the factors affecting the user's rental and return, time factors may include a season, a day of the week, a holiday, a time zone, and the like. The location-dependent factor is the location of the rental office, which may represent a space where public goods can be used near the rental office. For example, the location of the rental office is located in a park, amusement park, etc., or the location of the rental office near a station can increase the user's accessibility, which can increase the number of rentals and returns of users according to the user's floating frequency . Accordingly, the demand forecasting device may classify public goods use information having a 24-hour period according to factors affecting rental and return of users.

In step 504, the demand forecasting device may apply the classified public goods use information to a machine learning-based data analysis algorithm to estimate a usage pattern of the public goods for each element. The demand forecasting device can predict rental/return patterns of users through a usage pattern data aggregation that is inferred based on the estimated usage patterns of public goods.

More specifically, the demand forecasting apparatus may train a machine learning model according to a data analysis algorithm by using public goods use information of users about public goods. The machine learning model may be a model that predicts parameters modeling public goods usage information based on users' public goods usage information.

Public goods utilization information can be modeled to follow a discrete probability distribution based on parameters. For example, the demand forecasting device may determine the number of users using a public good as a parameter modeling a discrete probability distribution. When the parameters are determined, the demand forecasting device may predict demand for public goods on the next day based on the modeled use information. That is, a parameter according to an embodiment of the present invention may mean a parameter required for modeling the use information 105 as a discrete probability distribution.

For example, the demand forecasting device may set the number of users who have used public goods for a certain period of time as a random variable, and determine a parameter such that a probability mass function for the random variable follows a Poisson distribution. For example, a probability mass function for the number of users using a public good during a certain period of time may follow a Poisson distribution as shown in Equation 1 below.

는 n이 푸아송 분포를 따르도록 모델링하는 파라미터를 의미할 수 있다. N(t)는 공공재를 이용한 사용자들의 수를 랜덤 변수로 나타낸 것이다. P_N(t)(n)는, 푸아송 분포를 따르는 확률 질량 함수일 수 있다. 본 발명은 파리미터가 결정되는 경우, 공공재를 이용한 사용자들의 수를 확률 질량 함수로 표현할 수 있다. In Equation 1, n may mean the number of users who use public goods for time t in a specific place.

may mean a parameter modeling so that n follows a Poisson distribution. N(t) represents the number of users using a public good as a random variable. P _N(t) (n) may be a probability mass function following a Poisson distribution. In the present invention, when a parameter is determined, the number of users using a public good may be expressed as a probability mass function.

As another example, the demand forecasting device may set the number of users who have used public goods for a certain period of time as a random variable, and determine a parameter such that a probability mass function for the random variable follows a Gaussian distribution. For example, a probability mass function for the number of users using a public good during a certain period of time may follow a Gaussian distribution as shown in Equation 2 below.

은 특정 장소에서 t시간 동안 공공재를 이용한 사용자들의 수를 의미할 수 있다.

는

이 가우시안 분포를 따르도록 모델링하는 파라미터를 의미할 수 있다. N(t)는 공공재를 이용한 사용자들의 수를 랜덤 변수로 나타낸 것이다. P_N(t)(

)는, 가우시안 분포를 따르는 확률 질량 함수일 수 있다. In Equation 2,

may mean the number of users who use public goods for time t in a specific place.

Is

This may mean a parameter that is modeled to follow the Gaussian distribution. N(t) represents the number of users using a public good as a random variable. _PN(t) (

) may be a probability mass function following a Gaussian distribution.

When the parameter is determined, the demand forecasting device may express the number of users using the public good as a probability mass function. The demand forecasting device may predict the number of users who will use the public good at a specific time based on the probability mass function.

For example, as the machine learning model, a Recurrent Neural Network (RNN) that analyzes continuous data may be used. For example, Long Short-Term Memory (LSTM) may be used. The type of machine learning model is not limited to the described examples, and any machine learning model that can be employed by a person skilled in the art may be used. The machine learning model may be trained to output parameters modeling public goods usage information based on the public goods usage information. For example, the demand forecasting device may identify the number of users who use a public good for time t in a specific place from public good use information.

For example, the input data of the machine learning model may be public good use information (eg, the number of users who used the public good for time t). For example, in the case of commuting time, the demand for public goods may be higher than other times. The number of users using public goods may vary by season, holiday, rental location, day of the week, and time according to the access environment.

Input data of the machine learning model may be public goods use information classified according to an access environment in order to analyze a demand pattern for public goods.

For example, input data of the machine learning model may be public goods use information collected in winter among the four seasons. In this case, the demand forecasting device may analyze a pattern of demand for public goods using public goods use information collected in winter. Alternatively, the input data of the machine learning model may be information on the use of public goods collected during work hours or work hours on weekdays in recent years by time zone. In this case, the demand prediction device may analyze a pattern of demand for public goods using public goods use information collected during work hours or work hours on weekdays.

Alternatively, the input data of the machine learning model may be information on the use of public goods collected for each rental station and for each different location. In this case, the demand prediction device may analyze a pattern of demand for public goods by using public goods use information collected for each place.

Also, the output data of the machine learning model may be a parameter modeling public goods use information. For example, the output data of the machine learning model may be a parameter that makes the number of users using a public good follow a Poisson distribution at time t.

Training of the machine learning model may be performed based on public goods use information collected up to now. Parameters determined according to public goods use information collected so far may be used as correct answer labels of the machine learning model.

The machine learning model may be additionally trained based on public goods use information newly collected every day. The machine learning model may determine parameters of the probability mass function for each time slot of the next day.

In step 505, the demand prediction device may perform a process of generalizing a rental/return pattern appearing to a user by analyzing a mutual relationship between classified public goods use information. The demand forecasting device may analyze a specific condition that affects identification of the number of users identified for each item by using a correlation between each item. The demand forecasting device can generalize rental/return patterns that appear to users under specific conditions. This can perform a generalization process using data by season, shared day data, and data by rental office.

① Seasonal data

In detail, when the demand forecasting device analyzes public goods use information classified as seasonal data, it may be found that the frequency of use of public goods increases in the number of users using public goods in spring and autumn. Analyzing in terms of space, it can be seen that the number of users at rental stations around amusement parks increases in spring and autumn, and the number of users using public goods around residential areas increases in summer and winter. Accordingly, the demand forecasting device may analyze the mutual relationship between information on the use of public goods for each item, and generalize the rental/return pattern of the user through this analysis.

② Shared day data

Demand forecasting device analyzes the public goods use information classified as shared day data, in the case of public holidays, in the case of public holidays, the number of users who rent public goods from rental offices located around the amusement park increases, and users rent/return for specific items. patterns can be identified.

③ Data by rental location

When the demand forecasting device analyzes the information on the use of public goods classified as data for each rental station, it can analyze the user's movement range through the maximum and minimum values of the user's rental and return values. The demand forecasting device may generalize the user's rental/return pattern according to the user's moving range.

As a result, the demand forecasting device can extract generalized rental/return patterns of users through a process of analyzing public property use information classified by season, holiday, and location of rental office.

In step 506, the demand forecasting device may analyze public goods use information using a method of ① analyzing public goods use information according to an access environment or ② a method of analyzing public goods use information and arrangement request information.

① Method of analyzing public goods use information according to access environment

The demand forecasting device classifies the public property use information collected for a day at each rental center by day and time, and runs a machine learning-based data analysis algorithm through the classified data. Predictable.

② How to analyze public goods use information and placement request information

The demand forecasting device may perform feedback and weight update using arrangement request information due to lack of supply of public goods. In detail, when it is impossible to rent public goods at a desired time and place due to insufficient supply of public goods, users may transmit arrangement request information for public goods to the demand forecasting device through an application of a wireless terminal. Here, the user's arrangement request information can be used as feedback of the user's public goods rental and return pattern inferred through a machine learning-based data analysis algorithm.

In addition, the placement request information can be used to update the weight of an algorithm that analyzes and predicts a user's pattern based on data classified according to the day of the week and time zone. In detail, the present invention creates a set by weighting the predicted result data of the two models in the process of aggregating the data inferred through the analysis and prediction algorithm by day of the week and by time of day based on one-day data (24-hour data) to be predicted. can At this time, the weight of each model will be used in a method of updating the data later when feedback comes in during the process of aggregating the already predicted 24-hour cycle data while increasing the weight of the model with more accurate prediction result data at the time the batch request information is collected. can

Through this, the present invention reflects the user's real-time feedback on the data predicted one day ago (because it is 24-hour cycle data, so the data predicted through the algorithm comes out one day before), and also utilizes real-time information of the user's tendency through modification of the predicted data. can

In step 507, the demand prediction device aggregates the generalized rental/return pattern of the user through the season, holiday, and location of the rental office, and the rental/return pattern of the user predicted for the day and time of the week. Rental/return patterns can be predicted.

In step 508, the demand forecasting device may collect arrangement request information transmitted due to lack of supply of public goods. The demand forecasting device may utilize the dynamically changing demand for public goods in real time by applying the arrangement request information collected in steps 502 and 506 .

As a result, the present invention can collect public goods use information and arrangement request information due to insufficient supply of public goods of a user who accesses or uses public goods through an application of a wireless terminal. The present invention can apply a machine learning-based data analysis algorithm after classifying the collected public goods use information and placement request information according to season, holiday, location, day of the week, and time of the number of public goods users. Thereafter, the present invention analyzes and predicts the user's rental/return pattern of the next day through a data analysis algorithm, and provides a service of supplying public goods in accordance with the predicted user's rental/return pattern.

In addition, the present invention can increase the success rate of accessing public goods of a sharing platform by users by receiving feedback on rental/return patterns of users predicted through arrangement request information due to insufficient supply of public goods. This can be applied to all sharing platform fields where a sharing economy is used.

According to the present invention, a certain arrangement time and a user's rental/return pattern according to the arrangement location can be improved to the arrangement time of public goods. In addition, the arrangement request information of the present invention can be used as feedback on the user's rental/return pattern analyzed from public property usage information having a 24-hour period, and can be modified by reflecting real-time in predicted data.

6 is a graph showing experimental results for demand prediction according to an embodiment of the present invention.

Each graph in FIG. 6 is a graph in which the predicted result obtained by modeling with Poisson distribution for each time period is compared with the number of actual users. FIG. 6(a) shows a graph of the predicted result obtained by defining the number of users of a public good as a random variable and modeling with a Poisson distribution in the morning and the actual number of users. FIG. 6(b) shows a graph of the number of users of a public good defined as a random variable and the predicted result obtained by modeling with a Poisson distribution during lunchtime and the number of actual users. FIG. 5(c) shows a graph of the number of users of a public good defined as a random variable in the evening and a result predicted by modeling with a Poisson distribution and the number of actual users. FIG. 6(d) shows a graph of the number of users of a public good defined as a random variable and the predicted result obtained by modeling with a Poisson distribution at dawn and the number of actual users.

In each graph, a solid line (Data) may be an actual distribution, and a dotted line (Prediction) may be a result modeled as a probability distribution. As another example, each graph may be a graph representing a result of predicting the number of vehicles on the highway.

7 is a graph showing experimental results for demand prediction according to an embodiment of the present invention.

7 is a graph illustrating a result of predicting demand with a machine learning model having a long short-term memory (LSTM) structure, considering that the pattern of the number of users changes over time.

7(a) is a graph showing the number of users of a public good defined as a random variable in the morning, the result predicted by modeling with a Poisson distribution, the number of actual users, and the modeling result predicted through machine learning. is indicated by 7(b) is a graph showing the number of users of a public good defined as a random variable during lunchtime, the result predicted by modeling with a Poisson distribution, the number of actual users, and the modeling result predicted through machine learning. is indicated by (c) of FIG. 7 is a graph showing the number of users of public goods as a random variable in the evening, the result predicted by modeling with a Poisson distribution, the number of actual users, and the modeling result predicted through machine learning. is indicated by 7(d) is a graph showing the number of users of a public good defined as a random variable at dawn, the result predicted by modeling with a Poisson distribution, the number of actual users, and the modeling result predicted through machine learning. is indicated by

Referring to (a)-(d) of FIG. 7 , a difference between a result of modeling based on actual use information and a result of modeling using a machine learning model may not be large. In each graph, the thickest solid line (ML) represents the distribution modeled through the machine learning model, the dotted line (Dist) represents the distribution modeled based on actual data, and the lightest solid line (Data) may be the actual distribution. . As another example, each graph may be a graph representing a result of predicting the number of vehicles on the highway.

8 is a flowchart illustrating a demand forecasting method according to an embodiment of the present invention.

In step 801, the demand forecasting device may collect public goods usage information of users received from the wireless terminal. The demand forecasting apparatus may generate public goods use information having a specific time period by adjusting a time sync for information on users' use of public goods. At this time, the demand forecasting device may generate public goods use information having a specific time period by controlling a synchronization of time for rental information and return information included in users' public goods use information for public goods disposed in the rental office.

In step 802, the demand forecasting device may collect placement request information from the wireless terminal through a placement request function of an application of the wireless terminal due to a situation in which a user cannot rent due to a lack of public goods.

In step 803, the demand forecasting device may classify public goods use information having a specific time period in response to each item in consideration of an access environment affecting rental and return of public goods. The demand forecasting device may classify public property use information having a specific time period in response to at least one item among seasons, holidays, rental locations, days of the week, and time zones based on the access environment.

The demand forecasting device may apply the classified public goods use information to a machine learning model. The demand forecasting device may identify the number of users using the public good for time t for each item by applying the classified public good use information to a machine learning model using a machine learning-based data analysis algorithm.

For example, the demand forecasting device has a unit of time of 1 hour, 2 hours?? The number of users who used public goods for n hours can be identified. In addition, the demand forecasting device may set a weekday commute time (am 07:00 to am 09:00) or a weekday leave time (pm 17:00 to pm 19:00). The demand forecasting device may identify the number of users who use public goods for t time set for each time period, that is, for 2 hours.

In step 804, the demand prediction device may determine rental/return patterns of users related to demand for public goods. The demand forecasting device may generalize rental/return patterns appearing to users by analyzing the number of users identified for each item and the mutual relationship between each item. In this case, the demand forecasting device may analyze a specific condition that affects identification of the number of users identified for each item by using a mutual relationship between each item. The demand forecasting device can generalize rental/return patterns that appear to users under specific conditions.

The demand forecasting device may determine rental/return patterns of users related to demand for public goods from generalized rental/return patterns of users. The demand forecasting device can determine rental/return patterns of users regarding the demand for public goods that can be supplied to each rental office for each time slot of the next day.

At this time, when the demand forecasting device receives arrangement request information due to lack of supply of public goods from the wireless terminal, based on the arrangement request information, the rental/return pattern of users determined by feeding back the rental/return patterns of users over time for each rental station. can be re-determined.

In step 805, the demand forecasting device may supply public goods to each rental station through the rental/return pattern of the user.

9 is a flowchart illustrating a demand forecasting method according to another embodiment of the present invention.

In step 901, the demand forecasting device may identify public goods use information on public goods of users. Public goods use information may include any one or more of users' usage details of public goods and users' requests for placement of public goods. The details of use of public goods may include at least one of a usage time of the public goods, a place where the public goods are used, and a type of the public goods.

The demand forecasting device may identify the number of users who have used the public good for a certain period of time based on the public good use information. For example, the demand forecasting device may identify public good use information for a common period, such as a specific season or commuting, based on the public good use information.

In step 902, the demand forecasting device may determine parameters for modeling public goods use information by using the trained machine learning model. The parameter may refer to a parameter required for modeling public good use information with a discrete probability distribution. Public goods utilization information can be modeled to follow a discrete probability distribution based on parameters. For example, the demand forecasting device may determine a parameter for modeling the number of users using a public good as a discrete probability distribution. For example, the demand forecasting device may set the number of users who have used a public good for a certain period of time as a random variable, and determine a parameter such that a probability mass function for the random variable follows a Poisson distribution.

In step 903, the demand forecasting device may model public goods use information according to parameters. In step 904, the demand prediction device may predict the demand for public goods based on the modeling result. When the parameters are determined, the demand forecasting apparatus may predict demand for public goods on the next day based on the modeled public goods use information.

And, the demand forecasting device may perform the following processes in order to initially train the machine learning model. The demand forecasting device may identify public goods usage information about users' public goods. The demand forecasting device may determine parameters for modeling public goods use information. The demand forecasting device may train a machine learning model to output parameters based on public goods use information.

The demand forecasting apparatus may train the machine learning model using a difference between parameters output by the machine learning model and parameters determined by actual public goods use information. For example, the parameters determined by the public good use information may be parameters that allow the number of users who use the public good to follow a discrete probability distribution (eg, Poisson distribution, Gaussian distribution) for a certain period of time.

Meanwhile, the method according to the present invention is written as a program that can be executed on a computer and can be implemented in various recording media such as magnetic storage media, optical reading media, and digital storage media.

Implementations of the various techniques described herein may be implemented in digital electronic circuitry, or in computer hardware, firmware, software, or combinations thereof. Implementations may be a computer program product, i.e., an information carrier, e.g., a machine-readable storage, for processing by, or for controlling, the operation of a data processing apparatus, e.g., a programmable processor, computer, or plurality of computers. It can be implemented as a computer program tangibly embodied in a device (computer readable medium) or a radio signal. A computer program, such as the computer program(s) described above, may be written in any form of programming language, including compiled or interpreted languages, and may be written as a stand-alone program or in a module, component, subroutine, or computing environment. It can be deployed in any form, including as other units suitable for the use of. A computer program can be deployed to be processed on one computer or multiple computers at one site or distributed across multiple sites and interconnected by a communication network.

Processors suitable for processing a computer program include, by way of example, both general and special purpose microprocessors, and any one or more processors of any kind of digital computer. Generally, a processor will receive instructions and data from read only memory or random access memory or both. Elements of a computer may include at least one processor that executes instructions and one or more memory devices that store instructions and data. In general, a computer may include, receive data from, send data to, or both, one or more mass storage devices that store data, such as magnetic, magneto-optical disks, or optical disks. It can also be combined to become. Information carriers suitable for embodying computer program instructions and data include, for example, semiconductor memory devices, for example, magnetic media such as hard disks, floppy disks and magnetic tapes, compact disk read only memory (CD-ROM) ), optical media such as DVD (Digital Video Disk), magneto-optical media such as floptical disks, ROM (Read Only Memory), RAM (RAM) , Random Access Memory), flash memory, EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), and the like. The processor and memory may be supplemented by, or included in, special purpose logic circuitry.

In addition, computer readable media may be any available media that can be accessed by a computer, and may include both computer storage media and transmission media.

Although this specification contains many specific implementation details, they should not be construed as limiting on the scope of any invention or what is claimed, but rather as a description of features that may be unique to a particular embodiment of a particular invention. It should be understood. Certain features that are described in this specification in the context of separate embodiments may also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable subcombination. Further, while features may operate in particular combinations and are initially depicted as such claimed, one or more features from a claimed combination may in some cases be excluded from that combination, and the claimed combination is a subcombination. or sub-combination variations.

Similarly, while actions are depicted in the drawings in a particular order, it should not be construed as requiring that those actions be performed in the specific order shown or in the sequential order, or that all depicted actions must be performed to obtain desired results. In certain cases, multitasking and parallel processing can be advantageous. Further, the separation of various device components in the embodiments described above should not be understood as requiring such separation in all embodiments, and the program components and devices described may generally be integrated together into a single software product or packaged into multiple software products. You have to understand that you can.

On the other hand, the embodiments of the present invention disclosed in this specification and drawings are only presented as specific examples to aid understanding, and are not intended to limit the scope of the present invention. In addition to the embodiments disclosed herein, it is obvious to those skilled in the art that other modified examples based on the technical idea of the present invention can be implemented.

101: demand forecasting device
102: wireless terminal
103: public goods
104: one rental place
105: database
106: rental offices

Claims (20)

generating public goods usage information having a specific time period by adjusting time sync for public goods usage information of users received from a wireless terminal;
Classifying public goods use information having a specific time period in response to each item in consideration of an access environment affecting rental and return of public goods; and
determining rental/return patterns of users related to demand for public goods by applying the classified public goods use information to a machine learning model;
Demand forecasting method comprising a. According to claim 1,
The step of generating public goods use information having a specific time period,
A demand forecasting method for generating public goods usage information having a specific time period by adjusting time synchronization for rental information and return information included in users' public goods usage information for public goods placed in rental stations. According to claim 1,
The public goods use information,
As a result of rental or return by a user for a specific period at each rental place where the public goods are placed, the usage time of the public goods, the place where the public goods are used, and the type of the public goods, including any one or more of the details. , a demand forecasting method. According to claim 1,
The step of classifying public goods use information having a specific time period,
A demand prediction method of classifying public goods use information having a specific time period in response to at least one item among seasons, holidays, rental locations, days of the week, and time zones based on the access environment. According to claim 1,
The step of determining the rental/return pattern of the users,
identifying the number of users using public goods for time t for each item by applying the classified public goods use information to a machine learning model using a machine learning-based data analysis algorithm;
Generalizing a rental/return pattern appearing to users by analyzing the number of users identified for each item and the mutual relationship between each item; and
determining rental/return patterns of users related to demand for the public goods from the generalized rental/return patterns of users;
Demand forecasting method comprising a. According to claim 5,
The step of generalizing the rental/return pattern appearing to the users,
analyzing a specific condition affecting identification of the number of users identified for each item by using the mutual relationship between each item; and
Generalizing a rental/return pattern that appears to users based on the specific conditions;
Demand forecasting method comprising a. According to claim 1,
The step of determining the rental/return pattern of the users,
When receiving arrangement request information due to lack of supply of public goods from the wireless terminal,
A demand prediction method for re-determining the rental/return patterns of users determined by feeding back the rental/return patterns of users over time for each rental station based on the arrangement request information. According to claim 1,
The step of determining the rental/return pattern of the users,
A demand forecasting method that determines rental/return patterns of users regarding the demand for public goods that can be supplied to each rental office by time slot on the next day. In the demand forecasting device including a processor,
the processor,
Adjusting the sync of time for the public goods usage information of users received from the wireless terminal to generate public goods usage information having a specific time period;
Classify public goods use information with a specific time period in response to each item in consideration of the access environment that affects the rental and return of public goods,
Demand forecasting device for determining rental/return patterns of users related to demand for public goods by applying the classified public goods use information to a machine learning model. According to claim 9,
the processor,
A demand forecasting device for generating public goods usage information having a specific time period by controlling time synchronization of rental information and return information included in users' public goods usage information for public goods placed in a rental office. According to claim 9,
the processor,
A demand forecasting device that classifies public goods use information having a specific time period in response to at least one item among season, holiday, rental station, day of the week, and time zone based on the access environment. According to claim 9,
the processor,
Applying the classified public goods use information to a machine learning model using a machine learning-based data analysis algorithm to identify the number of users who used public goods for time t for each item,
The number of users identified for each item and the mutual relationship between each item are analyzed to generalize the rental/return pattern that appears to users,
Demand forecasting device for determining rental/return patterns of users related to the demand for the public goods from the generalized rental/return patterns of users. According to claim 12,
the processor,
A demand forecasting device that analyzes a specific condition affecting identification of the number of users identified for each item by using the mutual relationship between each item, and generalizes a rental/return pattern appearing to users according to the specific condition. According to claim 9,
the processor,
When receiving arrangement request information due to lack of supply of public goods from the wireless terminal,
A demand forecasting device for re-determining the rental/return patterns of users determined by feeding back the rental/return patterns of users over time for each rental station based on the arrangement request information. According to claim 9,
the processor,
Demand forecasting device that determines rental/return patterns of users regarding the demand for public goods that can be supplied to each rental office by time slot on the next day. In the method of predicting demand for public goods,
identifying public goods use information for public goods of users;
determining parameters modeling the public good use information using a trained machine learning model;
modeling the public good use information according to the parameters; and
Predicting the demand for the public good based on the modeling result
Demand forecasting method comprising a. According to claim 16,
The public goods use information,
As a result of rental or return by the user for a specific period at each rental place where the public goods are placed,
Demand forecasting method comprising details including at least one of a use time of the public good, a place where the public good is used, and a type of the public good. According to claim 16,
The parameter is
Demand prediction method, which means a parameter required for modeling the public good use information with a discrete probability distribution. According to claim 18,
Further comprising identifying the number of users who have used the public good for a certain period of time based on the public good use information,
To determine the parameters,
A method for predicting demand for determining a parameter for modeling the number of users using the public good as a discrete probability distribution. According to claim 16,
The modeling step is
Demand forecasting method of modeling using the parameters so that the public goods use information follows a discrete probability distribution.

Images