TW201914324A - Machine learning based time-dependent smart data pricing structure - Google Patents

Abstract

Conventional time-dependent charging mechanism merely considers usage behavior of user but not estimates future traffic usage, so that ISP operator cannot calculate a better price for user. Compared with the conventional mechanism, a machine learning based time-dependent data pricing structure is provided in the embodiments of the present invention. In the embodiments, a first and second traffic models would be established dynamically according to history data of user network usage, so as to estimate future traffic usage of each user according to data of transaction willingness and traffic model. Accordingly, ISP operator can provide a better time divided charging discount offer according to the future traffic usage, so that the targets for improving benefit and control network congestion would be achieved.

Description

Machine-based time-based smart billing architecture

The invention relates to a technology for network service and billing management, in particular to a time-based smart billing architecture based on machine learning.

The emergence of the Internet has completely changed the way people live and communicate. Further, the emergence of cloud services and big data makes people need to use network services all the time. For this reason, effective management of network resources and providing users with good network service quality have become very important issues. Currently, Internet Service Providers (ISPs) usually use Smart Data Pricing (SDP) mechanisms to set different pricing plans based on the usage of different users. Users are encouraged to adjust their Internet access mode to mitigate network congestion during peak hours and to provide users with a better Quality of Experience (QoE). Therefore, smart data billing has become a new and highly regarded research topic in the field of information communication.

Some scholars who have long studied smart data billing have proposed different data billing mechanisms, such as: Usage Pricing, Dynamic Pricing based on location/application/blocking, and two-sided billing (Two- Sided Pricing), Transaction Pricing. Although these billing methods can effectively improve network congestion in some applications, they do not consider the actual difference in network congestion at different times. In view of this, the mechanism of Time-dependent Smart Data Pricing (TDP) was proposed.

The core concept of the smart billing mechanism is that the user sets the price according to the usage of the network at different times, thereby facilitating the user to access the Internet during relatively relatively uncomfortable periods to balance the different time periods. Network usage. However, the above-mentioned smart charging mechanism still only uses the mathematical model to assume the user's usage habits. Therefore, it is difficult to accurately infer the degree of willingness of users to change their Internet access time (Willingness) according to the intelligent charging mechanism, which makes it impossible for ISPs to calculate the optimal pricing, and thus cannot effectively improve the user's change. The degree of willingness to access the Internet and the lack of network usage.

The well-known time-based smart billing mechanism only uses the mathematical model to assume the user's usage habits and does not use the concept of machine learning to estimate the user's future usage traffic, thus causing the ISP industry to fail to use the user's past and recent network usage behavior patterns. To predict traffic and calculate the best pricing. Therefore, the main object of the present invention is to provide a time-dependent smart charging architecture based on machine learning, which is capable of flexibly transmitting the first traffic pattern or the second traffic pattern according to historical data of the user using the network. In turn, the future usage flow of each user is estimated based on the data of the transfer intention and the flow pattern.

In order to achieve the above-described primary object of the present invention, the inventor of the present invention proposes an embodiment of the machine learning-based time-based smart charging architecture, which includes an electronic device, a database management system, and a data analysis and processing system. The electronic device records and uploads the user's network usage information. The database management system includes a first user database, a second user database, a transfer rule database, and a processing unit. The first user database stores the first network usage information of the user. The second user database stores the second network usage information of the user. The transfer rule database stores a transfer record of the user's network usage period, which is related to the network usage period recorded when the user uses the time-of-use pricing discount provided by the network service provider. The processing unit determines whether the user uses any of the pricing discounting schemes provided by the network service provider. If the user does not use any of the pricing discount schemes, the user's network usage information is regarded as the first network usage information and stored in the first user database. If the user uses the time-sharing pricing discount scheme, the user's network usage information is regarded as the second network usage information and stored in the second user database. The processing unit further calculates the transfer intention of each user according to the number of users and the number of transfer records of the network usage period. The data analysis and processing system includes a flow pattern update module, a transfer probability extraction module, and a future flow estimation module. The traffic pattern update module is configured to describe according to the first network usage information stored in the first user database and the second network usage information stored in the second user database and based on machine learning technology. At least one traffic pattern of each user's network usage habits. The transfer probability extraction module obtains the transfer intention of each user from the transfer rule database. The future traffic estimation module estimates the future usage traffic of each user based on the transfer intention and the traffic pattern.

In an embodiment of the invention, the time-based smart charging architecture further includes a pricing management system. The pricing management system calculates the profit and loss of the network service provider based on the future usage traffic and traffic patterns; and, based on the content of the time-sharing pricing discount scheme, the profit and loss, and the operating cost of the network service provider, the pricing management system can further Calculate the profitability of Internet service providers.

In this way, the ISP industry can set a suitable time-sharing pricing discount scheme according to the future usage flow, thereby achieving the purpose of improving profit and loss profitability and controlling network congestion.

The above described features and advantages of the invention will be apparent from the following description.

In order to more clearly describe a machine learning-based time-based smart billing architecture proposed by the present invention, the following will be described in detail with reference to the drawings.

First embodiment:

Please refer to FIG. 1, which is a block diagram showing a machine learning-based time-based smart charging architecture of the present invention. As shown in FIG. 1, the time-based smart billing architecture 1 of the present invention includes one or more electronic devices 3, a database management system 11, a data analysis and processing system 13, and a pricing management system 14.

In the embodiment of the present invention, the electronic device 3 used by each user 2 (for example, a communication module supporting a communication module such as a third-generation or fourth-generation mobile network) needs to load an application (Application, App) 121, the application 121 presents the user interface 12 through a display unit (eg, LCE, LED display, etc.) of the electronic device 3. It should be noted that the electronic device 3 with the networking capability referred to herein may be an electronic device such as a smart phone, a tablet computer, a smart watch, or a notebook computer.

In the case that the Internet Service Provider (ISP) does not provide any pricing discount to one of the users 2, the application 121 can record the user ID 2 during the period when the user 2 accesses the Internet. The plurality of network usage information of the user 2, and then the electronic device 3 uploads the network usage information to the database management system 11 for storage. Briefly, if the charging mechanism adopted by the ISP is a TIP (Time Independent Pricing) charging architecture, the network usage information collected by the application 121 includes the user identity (ID), the date and time of the Internet access. , network usage time, and network usage.

On the other hand, in the case where the ISP provides a time-sharing discounting scheme to the user 2, the application 121 can record the user's network usage information and upload it to the data during the user 2's access to the Internet. The library management system 11 stores it. Simply put, if the charging mechanism adopted by the ISP is a TIP (Time Dependent Pricing) charging architecture, the network usage information recorded by the application 121 will include the user identity (ID), the date of the Internet access, and Time, network usage time, network usage, and time-of-day pricing discounts. Time-of-use pricing discounts, for example, divide a day into 24 network usage periods, and make the pricing discount for each network usage period: .

The database management system 11 includes a first user database 111, a second user database 112, a transfer rule database 113, and a processing unit (for example, a CPU, a wafer, or a special controller, etc.). The processing unit determines whether the user 2 uses any of the pricing discounting schemes provided by the network service provider. For example, it is determined whether the network usage information of the user 2 includes the content of the time-of-day pricing discount or the user identity. If the user 2 does not use any pricing discount scheme (the network usage information does not have the content of the time-of-day pricing discount), the processing unit regards the user's network usage information as the first network usage information and stores it to the first The user database 111 can be referred to as the first user database 111 as a TIP database. If the user 2 uses the time-of-use pricing discount scheme (the network usage information has the content of the time-sharing discount), the processing unit regards the user's network usage information as the second network usage information and stores the information to the second user. The database 112 can be referred to as the second user database 112 as a TDP database.

It should be noted that, in other embodiments, the electronic device 3 may also determine whether to use any of the pricing discount schemes, and directly transmit the recorded network usage information to the first user database 111 or the second user database. 112.

In addition, in the case where the ISP provides a time-sharing discounting scheme to the user 2 (ie, based on the TDP charging architecture), when the user 2 wants to access the Internet, the application 121 promptly reminds the user that 2 can be divided. The time-of-day discount scheme provides access to the Internet during the preferential period, which encourages User 2 to use the Internet service during non-network congestion periods. Once the user 2 accepts the reminder message of the application 121 and finally uses the network service during the non-network congestion period, the transfer rule database 113 stores the plurality of network usage periods of the user 2 Transfer the record. Further, the processing unit of the database management system 11 can calculate the willingness of each user 2 according to the number of all users 2 and the number of transfer records of the network usage period, and store them in the transfer rule database 113. .

Referring to FIG. 1 , the data analysis and processing system 13 includes a flow pattern update module 131 , a transfer probability extraction module 132 , and a future flow estimation module 133 . These software modules are loaded and executed by processing units such as CPUs, chips, and the like. It should be noted that, according to the first network usage information stored in the first user database 111 and the second network usage information stored in the second user database 112, the traffic pattern update module 131 is At least one traffic pattern can be established to describe the usage habits of each user 2. Thus, as shown in FIG. 1, after the transfer probability extraction module 132 obtains the transfer intention of each user 2 from the transfer rule database 113, the future traffic estimation module 133 can be based on the transfer intention and the flow pattern. In turn, a future usage flow of each user 2 is estimated.

The process and manner of estimating the flow pattern and estimating the future usage flow will be explained in detail below with the aid of related diagrams.

The establishment of the first flow pattern and future flow patterns:

Please refer to Figure 2, which shows the estimation process of the user's traffic pattern. Among them, the tables (a), (b), and (c) in FIG. 2 show the time-based record table of the network usage of the user 2 under the TIP charging architecture. And, as shown in the schematic diagram of the time-sharing (segment) flow shown in the time period (d), the flow pattern update module 131 can infer that the user 2 is in the network according to the tables (a), (b), and (c). The average network usage for the time period "2" is 50. On the other hand, please refer to the schematic diagram of the transition of the user network usage period shown in FIG. The pricing discounts for each network usage period shown in Figure 3 are organized in the following table (1). Table 1)

According to the transfer record of the network usage period of the user 2 recorded in FIG. 3, the processing unit of the database management system 11 can establish the transfer record listed in the following table (2) and the transfer intention listed in the table (3). Table 2) table 3)

In the present invention, in particular, the flow pattern update module 131 can be configured to establish a first flow pattern ( Temp(u) ) for each user 2 in accordance with the following mathematical formulas (1) and (2). ………………………………….(1); , ……………..……………….(2);

The descriptions of the parameters or algebras contained in the above mathematical formulas (1) and (2) are organized in the following table (4). Table 4)

Through the above mathematical formulas (1) and (2), the traffic pattern update module 131 can automatically update the average usage (SJ) of the user 2 based on the TIP charging architecture for a certain network usage period (PJ). Further, after the user 2 starts to select the TDP charging structure, since the TDP database (that is, the second user database 112) may not have recorded the network traffic usage data related to the user 2, the future traffic estimation module 133, according to the following mathematical formulas (3), (4) and (5), the average usage pattern of each user 2 based on the TDP charging architecture for a certain network usage period (P _j ) can be established. .........................(3); ..............................................(4); .......................(5);

The descriptions of the parameters or algebras contained in the above mathematical formulas (3), (4) and (5) are organized in the following table (5). table 5)

It must be explained that in the initial stage of user 2's TDP billing architecture, the network usage traffic recorded by the TDP database is considered to be flowing out of the TIP billing architecture; therefore, the expected amount of outflow ( ) is used to indicate this part. It is conceivable to subtract the expected amount of effluent from the average usage (S _J ) ( After that, you can get the remaining expected amount ( ). On the other hand, it is also necessary to consider the expected inflow of user 2 during a certain network usage period (P _j ) based on the TDP architecture ( ). Simply put, the average usage of the user 2 based on the TDP billing architecture for a certain network usage period (P _j ) is versus Sum.

In the future traffic pattern of user 2 under the TDP billing architecture (ie, versus In the future, the pricing management system 14 can calculate the profit and loss of the ISP based on the future usage flow and traffic patterns; and, according to the content of the time-sharing pricing scheme (for example: The profit and loss, the operating costs of the ISP, and the benefits of the ISP's operations, the pricing management system 14 can further calculate the profit of the ISP. In the present invention, the pricing management system 14 performs the estimation of the profit according to the following mathematical formula (6). ..............(6);

The descriptions of the parameters or algebras contained in the above mathematical formula (6) are organized in the following table (6). Table (6)

(2) Establishment of the second flow pattern and future flow pattern:

The above description has clearly and completely described how to store the first network usage information based on the TIP database (ie, the first user database 111) (including: user identity (ID), internet access time, internet Access time period and network usage) complete the first traffic pattern ( Temp(u) ) and future traffic patterns (ie, versus The establishment of the sum; however, it must be noted that once more and more users 2 are transferred from the TIP billing architecture to the TDP billing architecture, the data in the TIP database is no longer updated, which may result in The traffic pattern established by the traffic pattern update module 131 cannot correctly describe the network usage habits of the user 2. With this in mind, the process and method of establishing the second flow pattern and future flow patterns will be explained in detail below with the aid of relevant diagrams.

Please refer to Figure 4, which shows the estimation process of the user's traffic pattern. Among them, the tables (a), (b), and (c) in FIG. 4 show the time-based record table of the network usage of the user 2 under the TDP architecture. Here, the traffic pattern update module 131 reversely obtains the time-sharing (segment) traffic as shown in the table (d) according to the record of the network usage of the user 2 under the TDP architecture. For example, according to the record of the usage of the second network usage period of the user's 2 under the TDP architecture on the 200th, 199th, and 198th days, the traffic pattern update module 131 reverses the user 2 The usage of the second network usage period under the TIP architecture.

In particular, the present invention further enables the traffic pattern update module 131 to establish a second traffic pattern according to the second network usage information ( The flow pattern update module 131 calculates the second flow pattern according to the following mathematical formulas (7), (8), and (9). ..................................(7); , ……………………….………….(8); ………….…………………….(9);

The descriptions of the parameters or algebras contained in the above mathematical formulas (7), (8) and (9) are organized in the following table (7). Table (7)

Simply put, the second flow pattern ( ) that is, according to the smart billing architecture 1 at the time of the invention, self-updating the first traffic pattern ( The result. Therefore, after obtaining the flow pattern, as described above for the function of the future traffic estimation module 133, the future traffic estimation module 133 can estimate each user according to the transfer intention and the flow pattern. The second flow pattern is used with future traffic. Finally, the pricing management system 14 can calculate the profit and loss and profit of the network service provider according to the future usage traffic and the traffic pattern.

Second embodiment:

Although the foregoing first embodiment can elastically transmit the first flow pattern ( Temp(u) ) or the second flow pattern ( The establishment of the future usage flow of each user 2 based on the data of the transfer intention and the flow pattern, so that the ISP industry can set a suitable time-price pricing discount scheme according to the future usage flow (for example) ), to achieve the purpose of controlling profit and loss and profit. However, in practical applications, the transfer record and transfer rules of the user's network usage period may not be easily obtained, and the future usage traffic of each user 2 based on the TDP charging architecture cannot be correctly estimated.

For the above reasons, the present invention further proposes a second embodiment of a time-dependent smart charging architecture based on machine learning. In particular, the future traffic estimation module 134 in the second embodiment further stores the second network usage information (including: user identity (ID), internet access) according to the TDP database (ie, the second user database 112). The date and time of the network access, the network usage period, the network usage, and the content of the time-of-day pricing discount, and then estimate and calculate the future usage traffic of each user 2.

It must be particularly noted that, unlike the aforementioned mathematical formulas (1) and (2) and the mathematical formulas (7), (8), and (9), the future flow estimation module 134 in the second embodiment performs future usage flow. The following steps are taken when estimating:

Step (1): setting the network usage traffic of a user to be tested in the future N network usage periods as ,and Indicates that the user to be tested is in any network usage period ( Network usage traffic; among them, ).

Step (2): establishing a plurality of user similarity matrices based on the second network usage information stored in the second user database 112.

Step (3): Selecting K neighboring users whose network usage habits are most similar to the user to be tested through the user similarity matrix.

Step (4): for any network usage period of the user to be tested ( The second network usage information of the K neighbor users is queried from the second user database 112.

Step (5): when the second network usage information indicates that any one of the K neighbor users is in the same network usage period as the user to be tested ( ) enjoy the same pricing discount ( At the time, the K neighbors are selected for the network usage during those network usage periods; among them, each neighbor user is selected for a total of L network usage data.

Step (6): averaging the L network usage of each neighbor user, and then averaging the K network average usage, that is, obtaining the user to be tested for any network usage period ( ) Network usage traffic.

Step (7): Repeat steps (4) to (6) until the estimation of the N network usage flows of the user to be tested is completed.

Please continue to refer to the following table (8A), table (8B), and table (8C), which records the network usage traffic record of the E-th user on the 98th-100th day; and, please also refer to the following table (9A) Tables (9B) and (9C) record the network usage traffic records of the F-th users on the 98th-100th day. Table (8A) Table (8B) Table (8C) Table (9A) Table (9B) Table (9C)

In step (2), based on the second network usage information stored by the second user database 112, the user similarity matrix of all users 2 is established according to the Euclid method or the Pearson method. For example, the foregoing table (8A), table (8B), and table (8C) are the network traffic records of the user E, and the network traffic records of the table (9A), the table (9B), and the table (9C) user F. And, its user similarity matrix is shown in the following table (10), wherein u ₁ , u ₂ , u ₃ , ..., u _{m in} Table (10) refer to the code of all users 2. Table (10)

Further, in order to estimate a specific user to be tested in any network usage period ( Network usage The second network usage information of the K neighbor users whose network usage habits are most similar to the user to be tested must be checked from the second user database 112. Mainly to check the network usage time of neighboring users ( Pricing discount with this network usage period ( Whether it is the same as this user to be tested. In the end, the average network usage of each of the neighbor users is averaged, and then the average usage of the K networks is averaged, that is, the user to be tested is obtained for any network usage period ( ) Network usage traffic. Moreover, according to the above manner, the network usage traffic of the user to be tested in the future N network usage periods can be estimated as .

The inventor of the present invention specifically names the smart charging architecture 1 as TDP-TR (Time-Dependent Pricing based on Transfer Rules) as described in the first embodiment, and the smart charging architecture described in the second embodiment. 1 is named TDP-KNN (Time-Dependent Pricing based on K Nearest Neighbors).

Based on the machine learning-based time-based smart billing architecture of the present invention shown in FIG. 1, engineers skilled in the art can find out according to the technical contents of the TDP-TR and TDP-KNN, the time of the present invention. According to the smart charging architecture, the TDP-TR technology can be used to estimate the future network traffic of the user 2 based on the TIP charging architecture and the TDP charging architecture. Meanwhile, after the user 2 uses the TDP charging architecture for a period of time, Directly use the technology of TDP-KNN to estimate the future network traffic of User 2. It is conceivable that the machine-based time-based smart charging architecture proposed by the present invention is sufficiently flexible in practical use.

Simulation verification:

In order to prove that the machine learning-based time-based smart billing architecture of the present invention can accurately predict traffic based on user history information, user past and recent network usage behaviors, and calculate the benefits of ISPs can be maximized. The best pricing, the inventor of this case pre-set the experimental parameters and rules contained in the following table (11), completed the relevant simulation experiments. Table (11)

Please refer to FIG. 5, which is a data graph showing the number of users relative to the (system) operation time. Also, please refer to Figure 6 at the same time, which shows a data graph of the number of users relative to the profit of the ISP. Among them, the data of Figure 5 shows that the system operation time required by the TDP-TR method is significantly less than the TDP-KNN method. At the same time, the data in Figure 6 shows that both the TDP-TR and TDP-KNN methods can bring considerable profit growth to the ISP industry as the number of users continues to increase; among them, the TDP-KNN method particularly leads to better profit growth.

Continuing to refer to FIG. 7, a data graph showing the number of divisions of the network usage period relative to the (system) operation time is shown. Also, please refer to Figure 8 at the same time, which shows a data graph of the number of segments used during the network usage period relative to the profit of the ISP. Among them, the data in Figure 7 shows that under the same number of users, the system operation time required by the TDP-TR method and the TDP-KNN method is extended as the number of divisions of the network usage period increases. In addition, the data in Figure 8 shows that under the same number of users, the profit brought by the TDP-TR method and the TDP-KNN method to the ISP is also increased as the number of segments of the network usage period increases.

Features and effects

Based on the above detailed description of the machine learning-based time-based smart charging architecture proposed by the embodiments of the present invention, it is believed that engineers and telecommunication operators familiar with network communication technologies can easily find that the present invention is applied to practical applications. It shows the following characteristics and effects:

The well-known time-based smart billing mechanism only uses the mathematical model to assume the user's usage habits and does not use the concept of machine learning to estimate the user's future usage traffic, thus causing the ISP industry to fail to use the user's past and recent network usage behavior patterns. To predict traffic and calculate the best pricing. It is better than the prior art that the machine learning-based time-based smart charging architecture proposed by the embodiment of the present invention can flexibly transmit the first traffic pattern or the second traffic type according to the historical data of the user using the network. The establishment of the sample, and then based on the data of the transfer intention and the flow pattern to estimate the future usage flow of each user 2, so that the ISP industry can set a suitable time-price pricing discount scheme according to the future usage flow (for example ), to achieve the purpose of controlling profit and loss and profit.

Through the implementation of the present invention, the ISP can not only alleviate the problem of network congestion or insufficient bandwidth, but also adaptively propose a suitable time-price pricing discount according to the network usage habits of different users 2. The way to attract users to change the time of using the network, to alleviate the effect of network congestion.

Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention, and any one of ordinary skill in the art can make some changes and refinements without departing from the spirit and scope of the present invention. The scope of the invention is defined by the scope of the appended claims.

1‧‧‧ hourly smart billing architecture

11‧‧‧Database Management System

12‧‧‧User interface

13‧‧‧Data Analysis and Processing System

14‧‧‧ Pricing Management System

111‧‧‧First User Database

112‧‧‧Second User Database

113‧‧‧Transfer rule database

2‧‧‧Users

3‧‧‧Electronic devices

121‧‧‧Application

131‧‧‧Flow pattern update module

132‧‧‧Transfer probability capture module

133‧‧‧Future Traffic Estimation Module

1 is a schematic diagram of a machine learning-based time-based smart charging architecture of the present invention; FIG. 2 is a flow chart showing an estimation process of a user's traffic pattern; and FIG. 3 is a schematic diagram showing a transition of a user network usage period. Figure 4 is a flow chart showing the flow pattern of the user; Figure 5 is a data graph showing the number of users relative to the (system) operation time; Figure 6 is a data graph showing the number of users relative to the profit of the ISP; Figure 7 The data curve showing the number of segments of the network usage period relative to the (system) operation time; Figure 8 is a data graph showing the number of segments of the network usage period relative to the profit of the ISP.

Claims (8)

A machine-based time-based smart charging architecture includes: a plurality of electronic devices respectively recording network usage information of one of a plurality of users, and separately uploading network usage information of the user; The management system includes: a first user database storing a plurality of first network usage information of the user; a second user database storing a plurality of second network usage information of the user; a database storing a transfer record of the plurality of network usage periods of the user, wherein the transfer record is related to a network recorded when the user uses a time-of-day pricing discount provided by an Internet service provider And a processing unit, determining whether the user uses any of the pricing discounting schemes provided by the network service provider, and if the user does not use any pricing discount scheme, the user's network is The road usage information is regarded as the first network usage information and stored in the first user database, and if the user uses the time-sharing pricing The solution, the user's network usage information is regarded as the second network usage information and stored in the second user database, and according to the number of the users and the transfer record of the network usage period The number of pens further calculates a transfer intention of each of the users; and a data analysis and processing system, comprising: a flow pattern update module, according to the first stored in the first user database a network usage information and/or the second network usage information stored in the second user database and based on machine learning technology, establishing a network describing usage habits of each of the users a flow rate pattern; a transfer probability extraction module, obtaining the transfer intention of the user from the transfer rule database; and a future flow estimation module, according to the transfer intention and the flow pattern, pushing Estimate a future usage flow for each of the users. The machine learning-based time-based smart charging architecture as described in claim 1, further comprising: a pricing management system, configured to calculate the network according to the future usage traffic and the traffic pattern One of the road service provider profits and losses; and, based on the content of the time-sharing pricing discount scheme, the profit and loss, and one of the operating costs of the network service provider, the pricing management system may further calculate the network One of the road service providers is profitable. The machine learning-based time-based smart charging architecture as described in claim 1, wherein the network usage information system includes: user identity, internet access date and time, network usage time, and Network usage. The machine learning-based time-based smart charging architecture as described in claim 3, wherein the network usage information further includes: the content of the time-sharing pricing discount. The machine learning-based time-based smart charging architecture of claim 1, wherein the traffic pattern update module establishes a first traffic pattern according to the first network usage information. The first flow pattern is represented by the following mathematical formula: ; , ; among them: Is the first flow pattern; u is a user; In the case that the network service provider does not provide any pricing discount scheme to a user, the network usage of the user during the Jth network usage period on the first day; Q The number of statistical days in which the user accesses the Internet; Based on the number of statistical days of accessing the Internet, the average network usage of the user during the Jth network usage period; N, I, Q, and J are integers. The machine learning-based time-based smart charging architecture as described in claim 1, wherein the data analysis and processing system further comprises: a future traffic estimation module, configured according to the second user data The second network usage information in the library estimates the future usage traffic of each of the users; and the future traffic estimation module performs the estimation of the future usage traffic. The following steps are as follows: (1) setting the network usage traffic of a user to be tested in the next N network usage periods, and Representing the network usage traffic of the user to be tested during any network usage period; (N) establishing a plurality of user similarity matrices based on the second network usage information stored in the second user database; (3) selecting a network through the user similarity matrix Using K neighboring users who are most similar to the user to be tested; (4) querying the second network corresponding to the K neighbor users from the second user database for any network usage period Using information; (5) when the second network usage information indicates that any of the K neighbor users and the user to be tested enjoy the same pricing discount scheme during the same network usage period, pick The network usage of the K neighboring users during the network usage period; wherein, each of the neighbor users is selected from a network usage amount of L, L is an integer; (6) The average network usage of each of the neighboring users is averaged, and then the average usage of the K networks is averaged, that is, the network usage traffic of the user to be tested during any of the network usage periods is obtained; 7) Repeat steps (4) through (6) until the completion of the Sensing a user's network usage of the N Collocation of traffic. The machine learning-based time-based smart charging architecture as described in claim 2, wherein the pricing management system performs the profit estimation by the following mathematical formula: ; among them: Refers to the parameter that takes the largest result; From , ,versus a combination of discounted prices for obtaining the best solution; a pricing discount combination of the time-sharing pricing discount scheme; Relating to the profit and loss of the network service provider caused by the time transfer behavior of the user; The cost of the network service provider; The benefit of the network service provider; ; a network usage amount of the j-th period before the user performs the time-lapse transfer behavior; a network usage amount of the j-th period after the user performs the time-lapse transfer behavior; The pricing discount at the jth time period using the time-sharing pricing discount scheme; j is an integer. The machine learning-based time-based smart charging architecture according to claim 5, wherein the traffic pattern update module further establishes a second traffic pattern according to the second network usage information. The second flow pattern is represented by the following mathematical formula: ; , ; ; among them: The second flow pattern; u is a user; Providing the time-of-sale pricing discount scheme to the network service provider In the case of any of the users, the network usage of the jth period in the i-th day; w is the total number of days the user accesses the Internet; Based on the total number of days of access to the Internet, the average network usage of the user during the jth network usage period; In the case that the network service provider does not provide the time-sharing pricing discount scheme, the network usage of the user during the j-th period of the i-th day; And the transfer intention of the user shifting from the jth time period to the kth time period; The transfer intention of the user shifting from the kth time period to the jth time period; i, k, and j are integers.