JP7733038B2 - Analysis method using analysis control server, analysis control function server, program, and core system - Google Patents

Description

The present invention relates to technology for analysis control servers located in core systems of mobile communication networks. In particular, it relates to technology for NWDAF (Network Data Analytics Function).

Common to both 5G (5th Generation) and 4G (4th Generation) standards, mobile communication networks are conceptually composed of user equipment (UE), radio access networks (RAN), and core network systems.

Figure 1 is a functional configuration diagram of a core system in conventional technology.

The core system has a group of physical user plane network devices arranged on a virtualization platform of a group of logical control plane network devices. The control plane network devices accommodate a large number of user terminals via the RAN and transmit and receive control signals such as for establishing communications. The user plane network devices transmit and receive user data from user terminals via the RAN.
Furthermore, various NF (Network Function) servers are arranged in the core system to control the control plane network devices.

The core system is equipped with an NWDAF (analysis control server) that acquires and analyzes a wide variety of log data from various data sources, such as communication equipment. The NWDAF receives analysis requests from other NF consumers and returns the analysis results. This is standardized by the 3GPP (3rd Generation Partnership Project) (registered trademark) (see, for example, non-patent document 1).

According to FIG. 1, the system is configured with the following NF servers.
Application Function Server (NF Consumer) 10
Repository Function Server (NRF) 11
Analysis function servers (NWDAF) 121, 122, 123
The NF consumer 10 is an NF that is the source of an analysis request.
In response to a discovery request from an NF consumer, the NRF 11 selects one NWDAF from among a plurality of NWDAFs 12 and responds.
The NWDAF 12 holds a learning (AI) model according to the analysis item, executes inference processing in response to an analysis request, and returns the analysis results.
The control plane network devices and the user plane network devices serve as data resources, and serve as input data for the NWDAF to execute inference processing using a learning model.

Figure 2 is a sequence diagram of analysis by an analysis function server in conventional technology.

The sequence in FIG. 2 is based on Non-Patent Document 1.
(S01) Assume that the NF consumer 10 wishes to obtain an analysis result for analysis item α. At this time, the NF consumer 10 inquires of the NRF 11 about an NWDAF that can analyze analysis item α. The NF consumer 10 sends a discover request to the NRF 11.
The NF consumer 10 sets, in the discovery request, a Preferred Supported Analytics Delay, which is a parameter indicating the time required to obtain the analysis result for the analysis item from the NWDAF 12 (see, for example, Non-Patent Document 2).

(S02) The NRF 11 manages the Supported Analytics Delay (input data collection delay + inference delay) for each analysis item for each NWDAF 12.
Then, when the NRF 11 receives a discovery request from the NF consumer 10, it selects an NWDAF that can respond with the analysis result within the Preferred Supported Analytics Delay. For example, if the NWDAF 121 is selected as being able to respond, the NRF 11 returns its address to the NF consumer 10.

(S03) The NF consumer 10 sends an analysis request to the address of the NWDAF returned from the NRF 11. The analysis request includes the following information:
Analysis item (Analytics ID): Specification of analysis processing Analysis target period: Period requested as analysis result Desired waiting time: Waiting time for response of analysis processing

According to FIG. 2, it is assumed that the NF consumer 10 has sent an analysis request for analysis item α to the NWDAF 121 .
When the NWDAF 121 receives an analysis request and is unable to return the analysis results within the desired waiting time, it returns an error response to the NF consumer 10. At this time, the NWDAF 121 may set an analysis processing waiting time (revised waiting time) in the error response. This reduces the possibility of an error response being returned again when the NF consumer 10 sends an analysis request again.

(S04) When the NWDAF 121 receives an analysis request from the NF consumer 10, it must possess a learning model for analyzing analysis item α. If the NWDAF 121 does not possess a learning model for analysis item α, in the case of Figure 2, it sends a model acquisition request to the NWDAF 123 that possesses learning model C for analysis item α.
In response to this, the NWDAF 123 transmits the learning model C to the NWDAF 121 .
The time related to S24 is delayed as the "model acquisition time" in the NWDAF 121.

(S05) Next, the NWDAF 121 transmits a network data request to the data source to acquire input data for analysis of the analysis item α.
In response, the data source transmits network data to the NWDAF 121 .
The time related to S25 is delayed as the "data collection time" in the NWDAF 121.

(S06) Finally, the NWDAF 121 inputs the network data into the learning model, executes inference processing, and obtains the analysis results.
The time related to S26 is delayed as the "inference time" in the NWDAF 121.

(S07) Then, the NWDAF 121 returns the analysis results to the NF consumer 10.

3GPP TS 23.288 “Architecture enhancements for 5G System (5GS) to support network data analytics services,” [online], [Retrieved March 7, 2023], Internet <URL: https://portal.3gpp.org/desktopmodules/Specifications/SpecificationDetails.aspx?specificationId=3579> 3GPP TS 23.501 “System architecture for the 5G System (5GS)”, [online], [Retrieved March 7, 2023], Internet <URL: https://portal.3gpp.org/desktopmodules/Specifications/SpecificationDetails.aspx?specificationId=3144>

According to the conventional technology described above, the greater the variety of NWDAFs, the greater the processing load on the NRF for discovering those NWDAFs. The NRF needs to frequently update the Supported Analytics Delay for each NWDAF. Furthermore, the NRF needs to instantly select and respond to discovery requests received at approximately the same time from different NF consumers, which necessitates a reduction in the processing load. Furthermore, the NRF needs to monitor the processing load of each NWDAF to prevent inference processing from concentrating on a single NWDAF.

However, the NRF processing delays are caused by the processing load imposed by responding in real time to discovery requests received from multiple NF consumers. In addition, network load is also imposed by the frequent exchange of information between multiple NWDAFs. These may also contribute to processing delays for other NF servers.

In addition, as the model acquisition time required to acquire a learning model from another NWDAF is not included in the Supported Analytics Delay, there is a risk of unexpected processing delays occurring as a result of the NWDAF processing delay.

The present invention therefore aims to provide an analysis method, analysis control function server, program, and core system using an analysis control server that reduces the processing load on the repository function server and satisfies the desired waiting time for analysis requests from network function consumers as much as possible.

According to the present invention, there is provided an analysis method for a core system having a plurality of analysis function servers that analyzes predetermined analysis items using a learning model from network data, the method comprising:
an analysis control function server as a producer server having an inference time table that stores inference times corresponding to the length of an analysis request period for each analysis function server;
A first step in which an analysis control function server receives an analysis request from a consumer server, the analysis request including an analysis item, an analysis request period, and a desired waiting time;
a second step in which the analysis control function server uses an inference time table to search for a first analysis function server whose inference time corresponding to the length of the analysis request period is equal to or shorter than the desired waiting time;
a third step in which the analysis control function server transfers the analysis request to the first analysis function server;
The method is characterized by having a fourth step in which the first analysis function server executes inference processing for the analysis item using a learning model from the network data and transmits the analysis results to the consumer server.

According to another embodiment of the method for analyzing a core system of the present invention,
It is also preferable that the inference times in the inference time table are statistically estimated from inference times corresponding to the duration of past analysis request periods.

According to another embodiment of the method for analyzing a core system of the present invention,
the analysis control function server further includes a collection time table that stores, for each analysis function server, a data collection time for acquiring network data from a data source according to the length of the analysis request period;
In the second step, the analysis control function server selects as a search candidate a first analysis function server whose sum of the inference time and the data collection time according to the length of the analysis request period using the collection time table is equal to or less than the desired waiting time,
In the third step, the analysis control function server sends a network data request to the data source, requesting that the data source send network data for an analysis request period to the first analysis function server;
For the fourth step, the first analysis function server also preferably receives network data from a data source.

According to another embodiment of the method for analyzing a core system of the present invention,
It is also preferable that the data collection times in the collection time table are statistically estimated from data collection times corresponding to the time lengths of past analysis request periods.

According to another embodiment of the method for analyzing a core system of the present invention,
The analysis control function server further includes a model acquisition table that stores, for each analysis function server, a model acquisition time required to acquire the learning model held by the analysis control function server, and
In the second step, if the analysis control function server determines using the model acquisition table that the first analysis function server does not hold a learning model based on the analysis item, it is also preferable to search for a second analysis function server among multiple other analysis function servers that hold the learning model, whose sum of the inference time, data collection time, and model acquisition time required to acquire the learning model is less than the desired waiting time, and to send a model acquisition request to the second analysis function server, requesting that the learning model be sent to the first analysis function server.

According to another embodiment of the method for analyzing a core system of the present invention,
It is also preferable that the model acquisition time in the model acquisition table is statistically estimated from the model acquisition time corresponding to the past learning model.

According to another embodiment of the method for analyzing a core system of the present invention,
Regarding the second step,
The analysis control function server sends a network data request to the data source requesting that network data for an analysis request period be sent to a second analysis function server;
The analysis control function server sends a learning request to the second analysis function server to update its own learning model;
It is also preferred that the second analysis function server receives network data for an analysis request period from the data source and executes training to update its learning model.

According to another embodiment of the method for analyzing a core system of the present invention,
The analysis control function server acquires the current computational resource utilization rate for each analysis function server,
It is also preferable that the analysis control function server executes the second step so that a plurality of analysis function servers that satisfy a predetermined condition are selected as search candidates for the first analysis function server in descending order of current computational resource utilization rate.

According to another embodiment of the method for analyzing a core system of the present invention,
The second step is also preferably performed so that the analysis function server with the highest computational resource utilization rate is selected as a search candidate for the first analysis function server.

According to another embodiment of the method for analyzing a core system of the present invention,
The analysis control function server schedules and manages the inference time slots for the analysis items to be executed for each analysis function server.
Regarding the second step, it is also preferable that the analysis control function server executes the second step in such a way that the time slots for the inference time to be executed by each analysis function server can be divided and scheduled so that the waiting time for each analysis item is less than the desired waiting time.

According to another embodiment of the method for analyzing a core system of the present invention,
The analysis function server is NWDAF (Network Data Analytics Function),
The learning model of the analysis function server is trained by MTLF (Model Training Logical Function),
The inference process of the analytical function server is executed by AnLF (Analytics Logical Function),
It is also preferable that the consumer server inquires of the NRF (Network Repository Function) about the address of an analysis control function server that can control the analysis item in question.

According to the present invention, there is provided an analysis control function server that communicates with a plurality of analysis function servers that analyze predetermined analysis items using a learning model from network data, and functions as a producer server in a core system,
an inference time table that stores, for each analysis function server, an inference time corresponding to the length of the analysis request period;
analysis request receiving means for receiving an analysis request including an analysis item, an analysis request period, and a desired waiting time from a consumer server;
an analysis function server search means for searching for a first analysis function server whose inference time corresponding to the length of the analysis request period is equal to or shorter than the desired waiting time, using an inference time table;
an analysis request transfer means for transferring the analysis request to the searched analysis function server;
The first analysis function server is caused to execute inference processing for the analysis item using a learning model from network data, and transmits the analysis results to the consumer server.

According to the present invention, a program that communicates with a plurality of analysis function servers that analyze predetermined analysis items using a learning model from network data and causes a computer to function as a producer server in a core system,
an inference time table that stores, for each analysis function server, an inference time corresponding to the length of the analysis request period;
analysis request receiving means for receiving an analysis request including an analysis item, an analysis request period, and a desired waiting time from a consumer server;
a first analysis function server search means for searching for a first analysis function server whose inference time corresponding to the length of the analysis request period is equal to or shorter than a desired waiting time, using an inference time table;
causing the computer to function as analysis request transfer means for transferring the analysis request to the first analysis function server;
The first analysis function server is caused to execute inference processing for the analysis item using a learning model from network data, and transmits the analysis results to the consumer server.

According to the present invention, there is provided a core system having a plurality of analysis function servers that analyzes predetermined analysis items using a learning model from network data,
an analysis control function server as a producer server having an inference time table that stores inference times corresponding to the length of an analysis request period for each analysis function server;
The analysis control function server
an inference time table that stores, for each analysis function server, an inference time corresponding to the length of the analysis request period;
analysis request receiving means for receiving an analysis request including an analysis item, an analysis request period, and a desired waiting time from a consumer server;
a first analysis function server search means for searching for a first analysis function server whose inference time corresponding to the length of the analysis request period is equal to or shorter than a desired waiting time using an inference time table;
an analysis request transfer means for transferring the analysis request to the first analysis function server;
The first analysis function server is characterized by executing inference processing for the analysis item using a learning model from the network data, and transmitting the analysis results to the consumer server.

The analysis method, analysis control function server, program, and core system using the analysis control server of the present invention reduce the processing load on the repository function server and maximize the desired waiting time for analysis requests from network function consumers.

FIG. 1 is a diagram illustrating a system configuration according to a conventional technique. FIG. 1 is a sequence diagram according to the prior art. 1 is a system configuration diagram according to the present invention. 3 is an explanatory diagram of a table held by the analysis control function server of the present invention. FIG. FIG. 2 is a sequence diagram according to the present invention. FIG. 2 is a functional configuration diagram of an analysis control function server according to the present invention. FIG. 10 is an explanatory diagram of the scheduling of each analysis control server in the analysis control function server of the present invention.

The following describes in detail an embodiment of the present invention using the drawings.

Figure 3 is a system configuration diagram for the present invention.

The core system of Fig. 3 has multiple NWDAFs (analysis function servers) 12 that analyze predetermined analysis items using learning models from network data, similar to Fig. 1. The NWDAFs 12 are existing systems and have the following two functions:
AnLF(Analytics Logical Function)
MTLF (Model Training Logical Function)
The learning model of the NWDAF 12 is trained by the MTLF, which retrains the learning model using log data acquired from a data source and allows the AnLF to use the retrained learning model.
The inference process of the NWDAF 12 is executed by the AnLF, which receives an analysis request and returns the analysis result.

Furthermore, compared to Figure 1, the core system of Figure 3 further includes an analysis control function server (NWDAF controller) 13.

[Analysis control function server (NWDAF controller) 13]
The NWDAF controller 13 functions as a producer server for the NF consumer 10 .

Figure 4 is an explanatory diagram of the tables held by the analysis control function server of the present invention.

According to FIG. 4, the NWDAF controller 13 has the following tables:
<Inference time table>
<Collection time table>
<Model acquisition table>
It should be noted that although the inference time table is required, the collection time table and the model acquisition table may be optional depending on the embodiment.

<Inference time table>
The inference time table stores inference times corresponding to the length of the analysis request period for each NWDAF 12. As shown in Fig. 4, this may be represented by a graph (an approximate straight line or an approximate curve) with the length of the analysis request period on the horizontal axis and the inference time on the vertical axis.
The inference times in the inference time table are statistically estimated from inference times corresponding to the length of past analysis request periods. Specifically, they may be trained by a supervised machine learning engine using training data in which the length of past analysis request periods is used as an explanatory variable and the past inference time at that time is used as a target variable. The inference time can be predicted by inputting the length of the analysis request period at the time of estimation.

<Collection time table>
The collection time table stores data collection times for acquiring network data from data sources according to the length of the analysis request period. As shown in Fig. 4, the collection time table may be represented by a graph (an approximate straight line or an approximate curve) with the length of the analysis request period on the horizontal axis and the data collection time on the vertical axis.
The data collection time in the collection time table is statistically estimated from the data collection time corresponding to the length of the past analysis request period. Specifically, it may be trained by a supervised machine learning engine using training data in which the length of the past analysis request period is used as the explanatory variable and the past data collection time at that time is used as the objective variable. The data collection time can be predicted by entering the length of the analysis request period at the time of estimation.

<Model acquisition table>
The model acquisition table stores, for each NWDAF, the model acquisition time required to acquire the learning model held by that NWDAF. As shown in Figure 4, the model acquisition table records, for each NWDAF, the "model ID" held by that NWDAF, the "model acquisition time" required to acquire that model, and the "update date and time" when that model was last trained using network data. The update date and time are recorded to determine the need for updating and training the learning model in relation to the analysis request period.
The model acquisition time in the model acquisition table is a statistical estimate based on the model acquisition times corresponding to past learning models. Specifically, it may be the average or median of the past model acquisition times.

Figure 5 is a sequence diagram for the present invention.

(S0) The NRF 11 discovers various NF servers required by the NF consumer 10, as in the prior art.
First, the NF consumer 10 sends a discovery request to discover the NWDAF 12 based on the analysis item α to the NRF 11. In response, the NRF 11 returns the address (URL (Uniform Resource Locator)) of the NWDAF controller 13 to the NF consumer 10.
That is, the NF consumer 10 inquires of the NRF 11 about the address of the NWDAF controller 13 that can control the analysis item α.
Of course, if the NF consumer 10 already knows the address of the NWDAF controller 13 to which the analysis request is to be sent, there is no need to make an inquiry to the NRF 11 .

In addition, multiple analysis function control servers 13 may be deployed for each type of analysis item. In such a case, the NRF 11 selects an analysis function control server 13 according to the analysis item in the discovery request received from the NF consumer 10, and returns the address of that analysis function control server 13 to the NF consumer 10.

According to the present invention, NRF11 no longer needs to discover NWDAF12 according to the analysis items in the analysis request, nor does it need to manage the Supported Analytics Delay for each NWDAF12. NRF11 can reduce the processing load in managing NWDAF12 and can instead focus on the processing load of discovering various other NF servers.

(S1) The NF consumer 10 sends an "analysis request" including the following information to the NWDAF controller 13. This is the same as the prior art described above.
・Analysis items ・Requested analysis period ・Desired waiting time

(S21) The NWDAF controller 13 manages the current computational resource utilization rate for each NWDAF 12. The NWDAF controller 13 may periodically receive computational resource utilization rates from multiple NWDAFs 12 under its control.
The NWDAF controller 13 selects, as search candidates, multiple NWDAFs 12 that satisfy a predetermined condition, in descending order of current computational resource utilization rate (in descending order of computational resource efficiency). Here, the "predetermined condition" refers to a predetermined rate or less. In other words, if there are only NWDAFs 12 with computational resource utilization rates higher than the predetermined rate, no search candidates are selected. On the other hand, the NWDAF with the highest computational resource utilization rate and a rate less than the predetermined rate may be selected as a search candidate.

(S22) Next, the NWDAF controller 13 selects, from among the NWDAFs 12 that are search candidates, an NWDAF 12 that satisfies the expected waiting time of the analysis request using the inference time table, collection time table, and model acquisition table described above in Figure 4.

The NWDAF controller 13 uses the inference time table to search for an NWDAF 12 whose inference time corresponding to the length of the analysis request period (Analytics target period) is equal to or shorter than the desired waiting time.
Inference time≦desired waiting time If the NWDAF 12 cannot find the desired waiting time, it returns an error response to the NF consumer 10. At this time, the NWDAF controller 13 sets the inference time as the revised waiting time in the error response. This allows the NF consumer 10 to send the analysis request again, taking the revised waiting time into consideration.

(S23) Next, the NWDAF controller 13 selects NWDAFs 12 as search candidates whose sum of the inference time and the data collection time according to the length of the analysis request period (Analytics target period) using the collection time table is less than the desired waiting time.
Inference time + data collection time ≦ desired waiting time If the learning model held by the NWDAF 12 has not been trained with network data based on the analysis request period, the network data needs to be acquired from the data source.

(S24) Next, the NWDAF controller 13 uses the model acquisition table to determine whether the NWDAF 12 that is a search candidate holds a learning model (learning model) based on the analysis item. For example, according to the above-mentioned FIG. 4, it is assumed that the NWDAF 121 in the model acquisition table does not hold a learning model C that can analyze the analysis item α.
If it is determined that the learning model C is not held in this manner, the NWDAF controller 13 searches for an NWDAF 12 among multiple other NWDAFs 12 that hold the learning model C, the NWDAF whose sum of the inference time, data collection time, and model acquisition time required to acquire the learning model is less than the desired waiting time.
Inference time + data collection time + model acquisition time ≦ desired waiting time Here, it is assumed that NWDAF123 (second analysis function server) holds learning model C. Note that if NWDAF121 already holds learning model C, the model acquisition time will be 0.
If the learning model held by the NWDAF 12 has not been trained with network data based on the analysis request period, the network data must be acquired.

If a plurality of NWDAFs 12 are found through S22 to S24, the NWDAF 12 with the lowest computational resource usage rate (worst computational resource efficiency) in S21 is selected.
On the other hand, if there is no NWDAF 12 found in S22 to S24, the processing time (= inference time + data collection time + model acquisition time) required by the NWDAF 12 with the lowest computational resource utilization rate in S21 is set as the revised waiting time in the error response.The NWDAF controller 13 then returns the error response to the NF consumer 10.

(S25) The NWDAF controller 13 schedules and manages the time slots for inference of analysis items to be executed for each NWDAF 12. That is, it schedules the timing for processing multiple analysis requests in the same NWDAF 12.
The NWDAF controller 13 may divide and schedule the time slots of the inference time to be executed by each NWDAF 12 so that the waiting time for each analysis item is equal to or less than the desired waiting time. Specific scheduling will be described later with reference to FIG. 7.

(S26) The NWDAF controller 13 sends a "model acquisition request" to NWDAF123 (second analysis function server) requesting that learning model C be sent to NWDAF121 (first analysis function server). In response, NWDAF123 sends learning model C that it holds to NWDAF121.

(S27) The NWDAF controller 13 also transmits a "network data request" to the data source, requesting that network data for the analysis request period be transmitted to the NWDAF 121 (first analysis function server). In response, the data source transmits network data corresponding to the analysis request period to the NWDAF 121.
At this time, the NWDAF controller 13 may transmit a network data request to the data source requesting that network data for the analysis request period be transmitted to the NWDAF 123 (second analysis function server). In response to this, the data source transmits network data corresponding to the analysis request period to the NWDAF 123.

(S3) The NWDAF controller 13 transfers the "analysis request" to the NWDAF 121 (first analysis function server). When the NWDAF controller 13 sends multiple analysis requests to the same NWDAF 12, the timing of the transmission is the time scheduled in S25.
The NWDAF 121 inputs network data into a learning model to perform inference processing, and the learning model may be retrained using the network data.

Regarding S26, S27, and S3, the NWDAF controller 13 may transmit the following three request messages simultaneously in parallel.
Model acquisition request to other NWDAF12 Network data request to data source Analysis request to NWDAF12

(S4) Upon receiving the analysis request, the NWDAF 121 (first analysis function server) executes inference processing for the analysis item using a learning model from the network data, and transmits the analysis results to the consumer.
It is also preferable that the NWDAF 121 transmits an analysis completion notification to the NWDAF controller 13. The analysis completion notification includes processing times consisting of inference time, data acquisition time, and model acquisition time. In response to this, the NWDAF controller 13 updates each table.

(S5) The NWDAF controller 13 may also send a "learning request" to the NWDAF 123 (second analysis function server) to update its own learning model.
In response to this, the NWDAF 123 receives network data for the analysis request period from the data source and trains to update its own learning model. The NWDAF 123 then sends a training completion notification to the NWDAF controller 13. The analysis completion notification includes the inference time for training as processing time. This causes the NWDAF controller 13 to update the inference time table.

Figure 6 is a functional configuration diagram of the analysis control function server in the present invention.

As shown in FIG. 6, the NWDAF controller 13 has an analysis request receiving unit 131, a computational resource usage rate determining unit 132, an analysis function server searching unit 133, a scheduling unit 134, a model acquisition request unit 135, a network data request unit 136, and an analysis request forwarding unit 137. These functional components are realized by executing a program that causes the computer installed in the NWDAF controller to function. The processing flow of these functional components can also be understood as an analysis control method.

The analysis request receiving unit 131 functions in the same manner as in S1 described above.
The computational resource utilization rate determination unit 132 functions in the same manner as in S21 described above.
The analysis function server search unit 133 functions in the same manner as in S22 to S24 described above.
The scheduling unit 134 functions in the same manner as in S25 described above.
The model acquisition request unit 135 functions in the same manner as in S26 described above.
The network data request unit 136 functions in the same manner as in S3 described above.

Figure 7 is an explanatory diagram of the scheduling of each analysis control server in the analysis control function server of the present invention.

(0 seconds) It is assumed that the NWDAF 12 receives the following two analysis requests.
Analysis request 1: Analysis item 1
Analysis request period
Desired waiting time: 50 seconds
Processing time: 20 seconds Analysis request 2: Analysis item 2
Analysis request period
Desired waiting time: 70 seconds
Processing time: 30 seconds. Note that "processing time" refers to the inference time + data collection time + model acquisition time.
At time 0 seconds, the analysis process for analysis item 1 is scheduled to start first, and then the analysis process for analysis item 2 is scheduled to start.

Note that according to the 3GPP standard, the NWDAF processes analysis requests received from NF consumers on a first-in, first-out (FIFO) basis. Therefore, if the NWDAF 12 receives analysis requests for analysis items 1 and 2 at approximately the same time, it will process them according to FIFO.

(20 seconds) Analysis processing for analysis item 1 ends. Then, analysis processing for analysis item 2 begins.

(30 seconds) Assume that the NWDAF 12 further receives the following analysis request.
Analysis request 3: Analysis item 3
Analysis request period
Desired waiting time: 30 seconds
Processing time: 20 seconds. In this case, if analysis processing for analysis item 3 is started after analysis processing for analysis item 2 (at 50 seconds), analysis processing for analysis item 3 will end at 80 seconds, and the desired waiting time of 30 seconds for analysis item 3 will not be met.
Therefore, analysis item 2 is scheduled to be processed in parts to satisfy the desired waiting time of 30 seconds for analysis item 3. Specifically, the analysis process for analysis item 2 is scheduled to be interrupted once at the 40-second mark, and analysis process for analysis item 3 is scheduled to start.

(40 seconds) Analysis processing for analysis item 2 is temporarily suspended. Then, analysis processing for analysis item 3 begins.

(60 seconds) Analysis processing for analysis item 3 ends. Then, analysis processing resumes for analysis item 2.

(70 seconds) Analysis processing for analysis item 2 is completed.

By dividing and scheduling the analysis process as shown in Figure 7, the expected waiting time for as many analysis requests as possible can be met. Of course, scheduling can also be performed according to a predetermined policy. The policy may be controlled by the mobile core operator.

In addition, by having the NWDAF controller 13 perform scheduling for each NWDAF 12, there is no need to update the Supported Analytics Delay between the NRF 11 and the NWDAF 12, which further reduces the processing load on the NRF 11 and avoids network congestion.

As explained in detail above, the analysis method, analysis control function server, program, and core system using the analysis control server of the present invention reduce the processing load on the repository function server and can meet the desired waiting time for analysis requests from network function consumers as much as possible.

Furthermore, this will enable, for example, "reducing the processing load on core systems of mobile communications networks," which will contribute to Goal 9 of the United Nations-led Sustainable Development Goals (SDGs), which is to "build resilient infrastructure, promote sustainable industrialization and foster innovation."

With regard to the various embodiments of the present invention described above, those skilled in the art will be able to easily make various changes, modifications, and omissions that fall within the scope of the technical spirit and principles of the present invention. The foregoing description is merely illustrative and is not intended to be limiting in any way. The present invention is limited only by the claims and their equivalents.

REFERENCE SIGNS LIST 10 NF consumer, application function server 11 NRF, repository function server 12 NWDAF, analysis function server 13 NWDAF controller, analysis control function server 131 Analysis request receiving unit 132 Computational resource usage rate determining unit 133 Analysis function server search unit 134 Scheduling unit 135 Model acquisition request unit 136 Network data request unit 137 Analysis request forwarding unit

Claims (14)

An analysis method for a core system having a plurality of analysis function servers that analyzes predetermined analysis items using a learning model from network data, comprising:
an analysis control function server as a producer server having an inference time table that stores inference times corresponding to the length of an analysis request period for each analysis function server;
A first step in which an analysis control function server receives an analysis request from a consumer server, the analysis request including an analysis item, an analysis request period, and a desired waiting time;
a second step in which the analysis control function server uses an inference time table to search for a first analysis function server whose inference time corresponding to the length of the analysis request period is equal to or shorter than the desired waiting time;
a third step in which the analysis control function server transfers the analysis request to the first analysis function server;
A method for analyzing a core system, characterized by having a fourth step in which the first analysis function server performs inference processing for the analysis item using a learning model from network data and transmits the analysis results to the consumer server. 2. A method for analyzing a core system according to claim 1, wherein the inference time in the inference time table is statistically estimated from the inference time corresponding to the length of a past analysis request period. the analysis control function server further includes a collection time table that stores, for each analysis function server, a data collection time for acquiring network data from a data source according to the length of the analysis request period;
In the second step, the analysis control function server selects as a search candidate a first analysis function server whose sum of the inference time and the data collection time according to the length of the analysis request period using the collection time table is equal to or less than the desired waiting time,
In the third step, the analysis control function server sends a network data request to the data source, requesting that the data source send network data for an analysis request period to the first analysis function server;
2. The method for analyzing a core system according to claim 1, wherein in the fourth step, the first analysis function server receives network data from a data source. 4. The method for analyzing a core system according to claim 3, wherein the data collection time in the collection time table is statistically estimated from data collection times corresponding to the length of a past analysis request period. The analysis control function server further includes a model acquisition table that stores, for each analysis function server, a model acquisition time required to acquire the learning model held by the analysis control function server, and
In the second step, if the analysis control function server determines using the model acquisition table that the first analysis function server does not hold a learning model based on the analysis item, it searches for a second analysis function server among multiple other analysis function servers that hold the learning model, whose sum of the inference time, data collection time, and model acquisition time required to acquire the learning model is less than the desired waiting time, and sends a model acquisition request to the second analysis function server, requesting that the learning model be sent to the first analysis function server.This is a core system analysis method described in claim 1. 6. The method for analyzing a core system according to claim 5, wherein the model acquisition time in the model acquisition table is statistically estimated from model acquisition times corresponding to past learning models. Regarding the second step,
The analysis control function server sends a network data request to the data source requesting that network data for an analysis request period be sent to a second analysis function server;
The analysis control function server sends a learning request to the second analysis function server to update its own learning model;
The core system analysis method according to claim 5, wherein the second analysis function server receives network data for an analysis request period from a data source and executes training to update its own learning model. The analysis control function server acquires the current computational resource utilization rate for each analysis function server,
A core system analysis method described in any one of claims 1 to 7, characterized in that in the second step, the analysis control function server executes the method to select multiple analysis function servers that meet specified conditions in order of highest current computing resource usage rate as search candidates for the first analysis function server. 9. The method for analyzing a core system according to claim 8, wherein the second step is executed so that the analysis function server with the highest utilization rate of computational resources is selected as a search candidate for the first analysis function server. The analysis control function server schedules and manages the inference time slots for the analysis items to be executed for each analysis function server.
A method for analyzing a core system according to any one of claims 1 to 7, characterized in that in the second step, the analysis control function server executes the inference time to be executed by each analysis function server in a manner that enables scheduling by dividing the time period so that the time period is equal to or less than the desired waiting time for each analysis item. The analysis function server is NWDAF (Network Data Analytics Function),
The learning model of the analysis function server is trained by MTLF (Model Training Logical Function),
The inference process of the analytical function server is executed by AnLF (Analytics Logical Function),
4. The method for analyzing a core system according to claim 1, wherein the consumer server executes the method by inquiring of an NRF (Network Repository Function) about the address of an analysis control function server that can control the analysis item. An analysis control function server that communicates with a plurality of analysis function servers that analyze predetermined analysis items using a learning model from network data and functions as a producer server in a core system,
an inference time table that stores, for each analysis function server, an inference time corresponding to the length of the analysis request period;
analysis request receiving means for receiving an analysis request including an analysis item, an analysis request period, and a desired waiting time from a consumer server;
an analysis function server search means for searching for a first analysis function server whose inference time corresponding to the length of the analysis request period is equal to or shorter than the desired waiting time, using an inference time table;
an analysis request transfer means for transferring the analysis request to the searched analysis function server;
An analysis control function server that causes a first analysis function server to perform inference processing for the analysis item using a learning model from network data and transmit the analysis results to a consumer server. A program that communicates with a plurality of analysis function servers that analyze network data using a learning model for a predetermined analysis item, and causes a computer to function as a producer server in a core system,
an inference time table that stores, for each analysis function server, an inference time corresponding to the length of the analysis request period;
analysis request receiving means for receiving an analysis request including an analysis item, an analysis request period, and a desired waiting time from a consumer server;
a first analysis function server search means for searching for a first analysis function server whose inference time corresponding to the length of the analysis request period is equal to or shorter than a desired waiting time, using an inference time table;
causing the computer to function as analysis request transfer means for transferring the analysis request to the first analysis function server;
A program that causes a first analysis function server to execute inference processing for a relevant analysis item using a learning model from network data, and transmits the analysis results to a consumer server. A core system having a plurality of analysis function servers that analyzes predetermined analysis items using a learning model from network data,
an analysis control function server as a producer server having an inference time table that stores inference times corresponding to the length of an analysis request period for each analysis function server;
The analysis control function server
an inference time table that stores, for each analysis function server, an inference time corresponding to the length of the analysis request period;
analysis request receiving means for receiving an analysis request including an analysis item, an analysis request period, and a desired waiting time from a consumer server;
a first analysis function server search means for searching for a first analysis function server whose inference time corresponding to the length of the analysis request period is equal to or shorter than a desired waiting time using an inference time table;
an analysis request transfer means for transferring the analysis request to the first analysis function server;
A core system characterized in that the first analysis function server executes inference processing for the analysis item using a learning model from network data and transmits the analysis results to the consumer server.