Data analysis method and system

Abstract

The application discloses a data analysis method and system. The method comprises the following steps: a first network element receives a first request; the first request is used for requesting to establish an alliance among a network device, the first network element and a second network element; the first request carries first information related to the network device and authentication information; the first network element performs authentication by using the authentication information in the first request; if the authentication is passed, the first request is forwarded to the second network element; the second network element judges whether to agree to establish the alliance by using the first information in the first request; if the alliance is agreed to be established, the alliance is established among the network device, the first network element and the second network element, so that the second network element and the network device collect different types of data, and the second network element and the network device jointly train a model for inference analysis based on the collected different types of data through the alliance.

Description

Technical Field

[0001] This invention relates to the field of wireless technology, and more particularly to a data analysis method and system. Background Technology

[0002] In live network implementations, due to data privacy and security considerations, the Network Data Analytics Function (NWDAF) may not be able to obtain complete data for training the model used for inference analysis, thus preventing effective model training. For example, in the intelligent evaluation scenario of application / user / slice Quality of Experience (QoE), the inability to obtain complete data for training the model for inference analysis leads to the inability to complete effective model training, and consequently, the inability to effectively evaluate QoE using the model. Summary of the Invention

[0003] In view of this, embodiments of the present invention aim to provide a data analysis method and system.

[0004] The technical solution of this invention is implemented as follows:

[0005] At least one embodiment of the present invention provides a data analysis method, the method comprising:

[0006] The first network element receives a first request; the first request is used to request the establishment of an alliance between the network device, the first network element, and the second network element; the first request carries first information and authentication information related to the network device;

[0007] The first network element uses the authentication information in the first request to perform authentication; if the authentication is successful, the first request is forwarded to the second network element.

[0008] The second network element uses the first information in the first request to determine whether it agrees to establish an alliance; if it agrees to establish an alliance, an alliance is established between the network device, the first network element, and the second network element, so that the second network element and the network device can collect different types of data, and jointly train a model for inference analysis based on the collected different types of data through the alliance.

[0009] Furthermore, according to at least one embodiment of the present invention, after an alliance is established between the network device, the first network element, and the second network element, the method further includes:

[0010] The second network element collects the first data; the first data represents data within the scope of the operator.

[0011] The network device collects second data; the second data represents data outside the scope of the operator.

[0012] Furthermore, according to at least one embodiment of the present invention, before the second network element collects the first data and the network device collects the second data, the method further includes:

[0013] The second network element negotiates with the network device to ensure that the collected first and second data meet preset conditions.

[0014] Furthermore, according to at least one embodiment of the present invention, the method further includes:

[0015] The second network element uses the collected first data to train the first model for inference analysis, obtains the intermediate training results of the first model, and sends the intermediate training results of the first model to the first network element.

[0016] The network device uses the collected second data to train the second model for inference analysis, and obtains the intermediate training results of the second model; and sends the intermediate training results of the second model to the first network element.

[0017] Furthermore, according to at least one embodiment of the present invention, the method further includes:

[0018] The first network element receives the intermediate training results of the first model sent by the second network element; and receives the intermediate training results of the second model sent by the network device; and fuses the intermediate training results of the first model and the intermediate training results of the second model to obtain the fused model parameters;

[0019] The first network element determines whether the fused model parameters have converged. If it is determined that the fused model parameters have not converged, the fused model parameters are decomposed into a first intermediate parameter and a second intermediate parameter. The first intermediate parameter is sent to the second network element, and the second intermediate parameter is sent to the network device. The first intermediate parameter is used by the second network element to continue training the first model, and the second intermediate parameter is used by the network device to continue training the second model. This process continues until the fused model parameters converge. When the fused model parameters converge, the fused model parameters are decomposed into a first model parameter and a second model parameter. The first model parameter is sent to the second network element, and the second model parameter is sent to the network device. The first model parameter is used by the second network element to obtain the first model, and the second model parameter is used by the network device to obtain the second model.

[0020] Furthermore, according to at least one embodiment of the present invention, the method further includes:

[0021] The first network element determines the model identifier corresponding to the converged model parameters and establishes a correspondence between the model identifier and the analysis identifier; and registers the model identifier and the analysis identifier with the third network element.

[0022] Furthermore, according to at least one embodiment of the present invention, the method further includes:

[0023] The first network element obtains a second request; the second request carries a first analysis identifier; the second request is used to request joint inference through the established consortium;

[0024] The first network element determines the first model identifier corresponding to the first analysis identifier based on the first analysis identifier and the correspondence between the model identifier and the analysis identifier; and determines the corresponding second network element and network device based on the first model identifier.

[0025] The first network element forwards the second request to the determined second network element and network device respectively.

[0026] Furthermore, according to at least one embodiment of the present invention, the method includes:

[0027] The second network element receives the second request; and responds to the second request by collecting the first data to be processed, using the first model for inference analysis to infer the first data to be processed, and obtaining the first inference result; and reporting the first inference result to the first network element; the first data to be processed is data within the scope of the operator.

[0028] The network device receives the second request; and responds to the second request by collecting second data to be processed; using a second model for inference analysis, it performs inference on the second data to be processed to obtain a second inference result; and reports the second inference result to the first network element; the second data to be processed is data outside the scope of the operator.

[0029] Furthermore, according to at least one embodiment of the present invention, the method further includes:

[0030] The first network element receives the first inference result and the second inference result; it then fuses the first inference result and the second inference result to obtain the final inference result.

[0031] At least one embodiment of the present invention provides a data analysis system, comprising:

[0032] The first network element is used to receive a first request; the first request is used to request the establishment of an alliance between the network device, the first network element, and the second network element; the first request carries first information and authentication information related to the network device.

[0033] The first network element is also used to perform authentication using the authentication information in the first request; if the authentication is successful, the first request is forwarded to the second network element.

[0034] The second network element is used to determine whether to agree to establish an alliance using the first information in the first request; if the alliance is agreed to be established, an alliance is established between the network device, the first network element, and the second network element, so that the second network element and the network device collect different types of data, and the alliance performs joint training of a model for inference analysis based on the collected different types of data.

[0035] Furthermore, according to at least one embodiment of the present invention, the system further includes: a network device;

[0036] The second network element is also used to collect first data; the first data represents data within the scope of the operator.

[0037] The network device is used to collect second data; the second data represents data outside the scope of the operator.

[0038] Furthermore, according to at least one embodiment of the present invention, the second network element is also used to negotiate with the network device to ensure that the collected first data and second data meet preset conditions.

[0039] Furthermore, according to at least one embodiment of the present invention, the second network element is also used to train a first model for inference analysis using the collected first data to obtain intermediate training results of the first model; and to send the intermediate training results of the first model to the first network element.

[0040] The network device is further configured to use the collected second data to train a second model for inference analysis, obtain intermediate training results of the second model, and send the intermediate training results of the second model to the first network element.

[0041] Furthermore, according to at least one embodiment of the present invention, the first network element is further configured to receive intermediate training results of a first model sent by the second network element; and receive intermediate training results of a second model sent by the network device; fuse the intermediate training results of the first model and the intermediate training results of the second model to obtain fused model parameters; and determine whether the fused model parameters converge; if it is determined that the fused model parameters converge, then decompose the fused model parameters into first model parameters and second model parameters; and send the first model parameters to the second network element and the second model parameters to the network device; wherein the first model parameters are used for the second network element to obtain the first model, and the second model parameters are used for the network device to obtain the second model.

[0042] Furthermore, according to at least one embodiment of the present invention, the first network element is also used to determine the model identifier corresponding to the converged model parameters, and establish a correspondence between the model identifier and the analysis identifier; and register the model identifier and the analysis identifier with the third network element.

[0043] Furthermore, according to at least one embodiment of the present invention, the first network element is further configured to obtain a second request; the second request carries a first analysis identifier; the second request is configured to request joint inference through an established alliance; and based on the first analysis identifier and the correspondence between the model identifier and the analysis identifier, determine a first model identifier corresponding to the first analysis identifier; based on the first model identifier, determine the corresponding second network element and network device; and forward the second request to the determined second network element and network device respectively.

[0044] Furthermore, according to at least one embodiment of the present invention, the second network element is also configured to receive the second request; and in response to the second request, collect first data to be processed, use a first model for inference analysis to infer the first data to be processed, obtain a first inference result; and report the first inference result to the first network element; the first data to be processed is data within the scope of the operator;

[0045] The network device is further configured to receive the second request; and respond to the second request by collecting second data to be processed; using a second model for inference analysis to infer the second data to be processed and obtain a second inference result; and report the second inference result to the first network element; the second data to be processed is data outside the scope of the operator.

[0046] Furthermore, according to at least one embodiment of the present invention, the first network element is also used to receive the first inference result and the second inference result; and to fuse the first inference result and the second inference result to obtain the final inference result.

[0047] The data analysis method and system provided in this embodiment of the invention involve a first network element receiving a first request. The first request requests the establishment of an alliance between a network device, the first network element, and a second network element. The first request carries first information and authentication information related to the network device. The first network element uses the authentication information in the first request to perform authentication. If authentication is successful, the first request is forwarded to a second network element. The second network element uses the first information in the first request to determine whether to agree to establish the alliance. If the alliance is agreed to, it is established between the network device, the first network element, and the second network element, enabling the second network element and the network device to collect different types of data. The alliance then performs joint training of a model for inference analysis based on the collected different types of data. By establishing an alliance between the network device, the first network element, and the second network element, the second network element and the network device can collect different types of data. This allows for joint training of a model for inference analysis based on the collected different types of data. Since complete data for training the model can be obtained, effective model training can be guaranteed. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the implementation flow of the data analysis method according to an embodiment of the present invention. Figure 1 ;

[0049] Figure 2 This is a schematic diagram illustrating the establishment of an alliance according to an embodiment of the present invention;

[0050] Figure 3 This is a schematic diagram of the implementation flow of the data analysis method according to an embodiment of the present invention. Figure 2 ;

[0051] Figure 4 This is a schematic diagram of data acquisition according to an embodiment of the present invention;

[0052] Figure 5 This is a schematic diagram of the implementation flow of the data analysis method according to an embodiment of the present invention. Figure 3 ;

[0053] Figure 6 This is a schematic diagram of the model training process according to an embodiment of the present invention;

[0054] Figure 7 This is a schematic diagram of the implementation flow of the data analysis method according to an embodiment of the present invention. Figure 4 ;

[0055] Figure 8 This is a schematic diagram of data reasoning in an embodiment of the present invention;

[0056] Figure 9This is a schematic diagram of the composition structure of the data analysis system according to an embodiment of the present invention. Detailed Implementation

[0057] Before introducing the technical solutions of the embodiments of the present invention, the relevant technologies will be explained first.

[0058] In related technologies, within the network intelligent architecture, NWDAF serves as a big data analytics engine function within the network. It is responsible for centrally collecting data from various layers, including network elements, network management systems, and applications. It independently trains data models, obtains relevant data models, and performs inference processes based on real-time class data. Simultaneously, NWDAF adheres to the 5G (5th Generation) service-oriented architecture and service registration and discovery mechanisms. It can register its services with the Network Repository Function (NRF). Before other network functions (NFs) can invoke NWDAF services, they must obtain the analytics identifier (ID) and access address of the relevant analytics services through service discovery via the NRF.

[0059] However, in live network practice, due to data privacy and security considerations, NWDAF may not be able to obtain the relevant data for training the data model, especially information related to third-party applications, thus hindering effective data model training. For example, in the application / user / slice QoE intelligent evaluation scenario, operators possess abundant network data but lack application-related data (such as QoE scoring, ambiguity, lag, etc.), resulting in ineffective QoE evaluation. Although a small sample can be obtained through dial-up testing and software development kit (SDK) integration, the manpower and cost are very high, and the limited sample size restricts the accuracy of training.

[0060] Based on this, in this embodiment of the invention, a first network element receives a first request; the first request is used to request the establishment of an alliance between a network device, the first network element, and a second network element; the first request carries first information and authentication information related to the network device; the first network element uses the authentication information in the first request to perform authentication; if authentication is successful, the first request is forwarded to the second network element; the second network element uses the first information in the first request to determine whether to agree to establish an alliance; if it agrees to establish an alliance, an alliance is established between the network device, the first network element, and the second network element, so that the second network element and the network device collect different types of data, and jointly train a model for inference analysis based on the collected different types of data through the alliance.

[0061] Figure 1 This is a schematic diagram illustrating the implementation flow of the data analysis method according to an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes steps 101 to 103:

[0062] Step 101: The first network element receives the first request; the first request is used to request the establishment of an alliance between the network device, the first network element, and the second network element; the first request carries first information and authentication information related to the network device.

[0063] It is understood that the network equipment mentioned may refer to equipment deployed by non-carriers.

[0064] For example, the network device is a third-party artificial intelligence (AI) device; or, the network device may also refer to a unit in a device not deployed by the operator, such as a third-party AI unit.

[0065] It is understandable that the first network element and the second network element can refer to equipment deployed by the operator.

[0066] For example, the first network element may be a network data analytics function (NWDAF) coordinator device deployed by the operator, and the second network element may be an NWDAF client deployed by the operator.

[0067] By establishing an alliance between the network device, the first network element, and the second network element, the second network element and the network device can collect different types of data, and the alliance can jointly train a model for inference analysis based on the collected different types of data.

[0068] It is understood that the first information may refer to information related to the network device, such as the address of the network device; or the federation role requested by the network device; the federation role may refer to a device deployed as a non-carrier in a federation; or the AI ​​service capabilities of the network device. The first information may also include parameters related to the requested federation, such as whether the federation type is horizontal or vertical; the federation may also be referred to as a consortium. The first information may also include parameters related to the model obtained through joint training via the federation, such as the model ID.

[0069] It is understood that the authentication information may refer to the enterprise ID; or it may refer to the authentication key / ciphertext.

[0070] Step 102: The first network element performs authentication using the authentication information in the first request; if the authentication is successful, the first request is forwarded to the second network element.

[0071] It is understandable that the first network element uses the authentication information in the first request for authentication, which may refer to verifying whether the network device has the authority to establish an alliance with the first network element and the second network element.

[0072] Taking the authentication information as an enterprise ID as an example, if the first network element stores the enterprise ID, it can be determined that the network device has the authority to establish an alliance with the first network element and the second network element.

[0073] Taking the authentication information as an authentication key / ciphertext as an example, if the first network element stores the authentication key / ciphertext, it can be determined that the network device has the authority to establish an alliance with the first network element and the second network element.

[0074] Step 103: The second network element uses the first information in the first request to determine whether it agrees to establish an alliance; if it agrees to establish an alliance, an alliance is established between the network device, the first network element, and the second network element.

[0075] It is understandable that the second network element can match the first information with the information stored locally. If the first information matches the information stored locally, it agrees to establish an alliance; otherwise, it disagrees to establish an alliance.

[0076] It is understandable that if the second network element agrees to establish an alliance, an alliance can be established between the network device, the first network element, and the second network element through the first network element.

[0077] Furthermore, after establishing an alliance between the network device, the first network element, and the second network element, the second network element and the network device can collect different types of data, and the alliance can jointly train a model for inference analysis based on the collected different types of data.

[0078] Figure 2 This is a diagram illustrating the establishment of an alliance. The NWDAF coordinator corresponds to the first network element, the NWDAF client corresponds to the second network element, and the third-party AI unit corresponds to the network device. For example... Figure 2 As shown, the process of establishing an alliance includes:

[0079] First, the Application Function (AF) network element sends a first request to the Network Exposure Function (NEF) network element; the first request carries first information and authentication information related to the third-party AI unit.

[0080] The first piece of information includes: the address of the third-party AI unit, its federation role, federation type, and model ID; the federation role can refer to a device deployed as a non-carrier in the federation; the federation type can be vertical; and the model ID can refer to the QoE model ID.

[0081] The authentication information may refer to the enterprise ID or the authentication key / ciphertext.

[0082] It is understood that the authentication information is used by the NWDAF coordinator to verify whether the network device has the authority to establish an alliance with the NWDAF coordinator and the NWDAF client.

[0083] Understandably, the first piece of information is used by the NWDAF client to determine whether it agrees to establish an alliance.

[0084] It should be noted that before the alliance is established, NWDAF clients and NWDAF coordinators complete service registration through NRF. The registration information includes: whether federation is supported, the type of federation supported (e.g., horizontal or vertical), the federation role (coordinator or client), the service area (e.g., carrier area or non-carrier area), and AI service capabilities (e.g., computing power).

[0085] Second, the NEF network element forwards the first request sent by the AF network element to the NWDAF coordinator.

[0086] Third, the NWDAF coordinator makes a preliminary judgment based on the authentication information carried in the first request.

[0087] Taking the enterprise ID as an example, if the NWDAF coordinator stores the enterprise ID, it can be determined that the network device has the authority to establish an alliance with the NWDAF coordinator and the NWDAF client; otherwise, the authentication fails.

[0088] Taking the authentication information as an authentication key / ciphertext as an example, if the NWDAF coordinator stores the authentication key / ciphertext, it can be determined that the network device has the authority to establish an alliance with the NWDAF coordinator and the NWDAF client; otherwise, authentication fails.

[0089] If authentication fails, the alliance establishment fails and a failure message is returned to the NEF network element; the NEF network element then notifies the AF network element of the failure message.

[0090] If authentication is successful, the NWDAF coordinator selects an NWDAF client to establish an alliance.

[0091] Fourth, the NWDAF coordinator sends the first request to the NWDAF client through service discovery or local configuration.

[0092] Fifth, the NWDAF client determines whether to agree to establish an alliance based on the first information in the first request.

[0093] If the NWDAF client has relevant information that matches the first information stored locally, it can agree to establish an alliance; otherwise, it will not agree to establish an alliance.

[0094] If the NWDAF client does not agree to establish the alliance, it returns a message to the NWDAF coordinator refusing to establish the alliance. The NWDAF coordinator then notifies the AF network element of the alliance failure through the NEF network element.

[0095] If an alliance is agreed to be established, the NWDAF client records the address and federation role of the NWDAF coordinator, as well as the address and federation role of the third-party AI unit. The federation role of the NWDAF coordinator can refer to the coordinator device deployed as an operator within the alliance; the federation role of the third-party AI unit can refer to the device deployed as a non-operator within the alliance.

[0096] Simultaneously, the NWDAF client returns an agreement to establish a federation to the NWDAF coordinator. The NWDAF coordinator further notifies the AF network element of the NWDAF client's address through the NEF network element. The AF network element then informs the third-party AI unit so that the third-party AI unit can record the address and federation role of the NWDAF coordinator, as well as the address and federation role of the NWDAF client.

[0097] It should be noted that after the alliance is formed, the interaction between the NWDAF coordinator, the NWDAF client and the third-party AI unit can be carried out directly, or through the NWDAF coordinator / NWDAF client -> NEF -> AF to interact with the third-party AI unit.

[0098] In this embodiment of the invention, the first network element sends a first request to the second network element, which has the following advantages:

[0099] (1) When the second network element agrees to establish an alliance, an alliance can be established between the network device, the first network element and the second network element.

[0100] (2) By establishing an alliance between the network device, the first network element and the second network element, the second network element and the network device can collect different types of data, that is, obtain complete data for training the model, thus ensuring effective model training.

[0101] Figure 3This is a schematic diagram illustrating the implementation flow of the data analysis method according to an embodiment of the present invention, as shown below. Figure 3 As shown, the method includes steps 301 to 307:

[0102] Step 301: The first network element receives a first request; the first request is used to request the establishment of an alliance between the network device, the first network element, and the second network element; the first request carries first information and authentication information related to the network device.

[0103] Step 302: The first network element performs authentication using the authentication information in the first request; if the authentication is successful, the first request is forwarded to the second network element.

[0104] Step 303: The second network element uses the first information in the first request to determine whether it agrees to establish an alliance; if it agrees to establish an alliance, then proceed to step 304; otherwise, proceed to step 307.

[0105] Step 304: Establish an alliance between the network device, the first network element, and the second network element.

[0106] Step 305: The second network element collects the first data; the first data represents data within the scope of the operator.

[0107] It is understood that the first data can specifically refer to data collected from network elements, network management systems, application processors (AF), radio access networks (RAN), terminals, etc. For example, user data collected from a terminal within the scope of a mobile operator.

[0108] Step 306: The network device collects second data; the second data represents data outside the scope of the operator.

[0109] It is understood that the second data may specifically refer to data collected from a certain company, for example, the second data may be application data collected from a certain company.

[0110] It is understandable that before the second network element collects the first data and the network device collects the second data, the second network element can negotiate with the network device to ensure that the collected first data and second data meet the preset collection conditions.

[0111] Here, meeting the preset data acquisition conditions may include at least one of the following:

[0112] The first data collection period and the second data collection period are the same;

[0113] The first data contains the same number of samples as the second data.

[0114] The sample ID corresponding to the first data is the same as the sample ID corresponding to the second data.

[0115] Step 307: The second network element returns a rejection message to the first network element.

[0116] Figure 4 This is a schematic diagram of data acquisition. The NWDAF coordinator corresponds to the first network element, the NWDAF client corresponds to the second network element, and the third-party AI unit corresponds to the network device. Figure 4 As shown, the data acquisition process includes:

[0117] First, the NWDAF client negotiates with the third-party AI unit to ensure that the first data collected by the NWDAF client and the second data collected by the third-party AI unit meet the preset collection conditions.

[0118] Understandably, to ensure the validity of the data, the NWDAF client negotiates with the third-party AI unit. The negotiation may include the time period for data collection, sample size, sample ID, etc.

[0119] Taking the time period of the data collection as an example, if the time period of the first data collected by the NWDAF client is the same as the time period of the second data collected by the third-party AI unit, it indicates that the first data collected by the NWDAF client and the second data collected by the third-party AI unit meet the preset collection conditions.

[0120] Taking the negotiated content as the sample size as an example, if the sample size of the first data collected by the NWDAF client is the same as the sample size of the second data collected by the third-party AI unit, it indicates that the first data collected by the NWDAF client and the second data collected by the third-party AI unit meet the preset collection conditions.

[0121] Taking the negotiated content as the sample ID as an example, if the sample ID corresponding to the first data collected by the NWDAF client is the same as the sample ID corresponding to the second data collected by the third-party AI unit, it indicates that the first data collected by the NWDAF client and the second data collected by the third-party AI unit meet the preset collection conditions.

[0122] It should be noted that the data of the same user using a certain service, their network domain data, and application layer data must be correlated at least in terms of time.

[0123] Second, based on the model ID or the negotiated time period and sample size, the NWDAF client collects first data from network elements, network management, AF, radio access network (RAN), terminals, etc., to form a local subset dataset.

[0124] Third, a third-party AI unit collects second data to form a local subset dataset.

[0125] In this embodiment of the invention, establishing an alliance between the network device, the first network element, and the second network element has the following advantages:

[0126] (1) It can collect operator data and application data based on the NWDAF architecture.

[0127] (2) It is possible to obtain complete data for training the model, so that effective model training can be completed subsequently.

[0128] Figure 5 This is a schematic diagram illustrating the implementation flow of the data analysis method according to an embodiment of the present invention, as shown below. Figure 5 As shown, the method includes steps 501 to 509:

[0129] Step 501: The first network element receives a first request; the first request is used to request the establishment of an alliance between the network device, the first network element, and the second network element; the first request carries first information and authentication information related to the network device.

[0130] Step 502: The first network element uses the authentication information in the first request to perform authentication; if the authentication is successful, the first request is forwarded to the second network element.

[0131] Step 503: The second network element uses the first information in the first request to determine whether it agrees to establish an alliance; if it agrees to establish an alliance, then proceed to step 504; otherwise, proceed to step 510.

[0132] Step 504: Establish an alliance between the network device, the first network element, and the second network element.

[0133] Step 505: The second network element collects first data; the first data represents data within the operator's scope; using the collected first data, the first model used for inference analysis is trained to obtain intermediate training results of the first model; and the intermediate training results of the first model are sent to the first network element.

[0134] Here, before the second network element collects the first data and the network device collects the second data, the second network element can also negotiate with the network device to ensure that the collected first data and second data meet the preset collection conditions.

[0135] Here, the second network element can be trained using the usual neural network model training method to train the first model used for inference analysis.

[0136] Step 506: The network device collects second data; the second data represents data outside the operator's scope; using the collected second data, the second model used for inference analysis is trained to obtain intermediate training results of the second model; and the intermediate training results of the second model are sent to the first network element.

[0137] Here, the network device can use a conventional neural network model training method to train the second model used for inference analysis.

[0138] Step 507: The first network element receives the intermediate training results of the first model sent by the second network element; and receives the intermediate training results of the second model sent by the network device; the intermediate training results of the first model and the intermediate training results of the second model are fused to obtain the fused model parameters; it is determined whether the fused model parameters have converged; if it is determined that the fused model parameters have converged, then step 508 is executed; otherwise, step 509 is continued.

[0139] Here, the first network element can use a fusion algorithm to fuse and update the intermediate training results of the first model and the intermediate training results of the second model to form global model parameters.

[0140] The fusion algorithm includes, but is not limited to, linear weighted fusion algorithm, cross fusion algorithm, etc.

[0141] Step 508: Decompose the fused model parameters into first model parameters and second model parameters; send the first model parameters to the second network element and send the second model parameters to the network device.

[0142] The first model parameters are used to provide the second network element with a first model. The second model parameters are used to provide the network device with a second model.

[0143] It is understood that the first network element determines the model identifier corresponding to the converged model parameters and establishes a correspondence between the model identifier and the analysis identifier; and registers the model identifier and the analysis identifier with the third network element, such as the NRF network element.

[0144] Step 509: Decompose the fused model parameters into first intermediate parameters and second intermediate parameters; send the first intermediate parameters to the second network element and the second intermediate parameters to the network device; the first intermediate parameters are used for the second network element to continue training the first model, and the second intermediate parameters are used for the network device to continue training the second model; and so on, until the fused model parameters converge.

[0145] Here, the process of decomposing the model parameters can be the reverse process of fusing the model parameters.

[0146] Step 510: The second network element returns a rejection message to the first network element.

[0147] Figure 6 This is a schematic diagram of the model training process. The NWDAF coordinator corresponds to the first network element, the NWDAF client corresponds to the second network element, and the third-party AI unit corresponds to the network device. Figure 6 As shown, the model training process includes:

[0148] First, the NWDAF coordinator sends the public key to the NWDAF client and the third-party AI unit;

[0149] Understandably, the NWDAF client and the third-party AI unit use public keys and encryption algorithms to encrypt and align sample IDs, ensuring that the sample ID corresponding to the first data is the same as the sample ID corresponding to the second data.

[0150] Second, the NWDAF coordinator, NWDAF client, and third-party AI unit perform vertical federated computation, i.e., model training, which includes the following steps:

[0151] 1. The NWDAF client uses the collected first data to train the first model for inference analysis, and obtains the intermediate training results of the first model; and sends the intermediate training results of the first model to the NWDAF coordinator after encryption.

[0152] 2. The third-party AI unit uses the collected second data to train the second model used for inference analysis, and obtains the intermediate training results of the second model; and sends the encrypted intermediate training results of the second model to the NWDAF coordinator.

[0153] The algorithms for encrypting intermediate training results include, but are not limited to, their respective encrypted gradients and encrypted losses.

[0154] 3. The NWDAF coordinator receives the encrypted intermediate training results of the first model sent by the NWDAF client, decrypts the intermediate iterative structure of the encrypted first model, and obtains the decrypted intermediate training results of the first model.

[0155] 4. The NWDAF coordinator receives the encrypted intermediate training results of the second model sent by the third-party AI unit, decrypts the intermediate iterative structure of the encrypted second model, and obtains the decrypted intermediate training results of the second model.

[0156] 5. The NWDAF coordinator merges and updates the intermediate training results of the first model and the intermediate training results of the second model to form global model parameters.

[0157] A fusion algorithm can be used to merge and update the intermediate training results of the first model and the intermediate training results of the second model to form global model parameters.

[0158] The fusion algorithm includes, but is not limited to, linear weighted fusion algorithm, cross fusion algorithm, etc.

[0159] Taking the linear weighted algorithm as an example, different weights can be assigned based on the contributions of the NWDAF client and the third-party AI unit to the global model; using the assigned weights, as well as the intermediate training results of the first model and the second model, the fused model parameters can be obtained.

[0160] 6. The NWDAF coordinator determines whether the fused model parameters have converged. If it is determined that the fused model parameters have converged, the fusion algorithm is used to decompose the fused model parameters into model parameters for the NWDAF client and the third-party AI unit, respectively, and send them to the NWDAF client and the third-party AI unit.

[0161] Meanwhile, the NWDAF coordinator informs the AF network element model that training is complete via the NEF network element.

[0162] Third, the NWDAF coordinator and NWDAF client shall register the supported model training capabilities with the NRF. The registration content shall include at least: model training results (such as model ID), model training method (such as vertical federation), and (if federated) federation architecture and components (such as third-party AI unit address).

[0163] Fourth, the NWDAF coordinator determines the model identifier corresponding to the converged model parameters, and stores the model identifier (ModelID) and analysis identifier (Analytics ID) locally after associating them.

[0164] The NWDAF coordinator registers the Analytics ID and the associated Model ID with the NRF network element.

[0165] The NRF element stores information about the association between the Model ID and the Analytics ID, as well as the model training method (such as vertical federation), federation architecture and components (such as the address of a third-party AI unit) related to the Model ID.

[0166] It should be noted that the training unit and inference unit of the NWDAF client are separate; therefore, the NWDAF client refers to the training unit of the NWDAF client. Furthermore, after model training is complete, the training unit of the NWDAF client transmits the model training results to the inference unit of the NWDAF client. The same applies to third-party AI units.

[0167] In this embodiment of the invention, establishing an alliance between the network device, the first network element, and the second network element has the following advantages:

[0168] (1) The second network element uses the collected first data to iteratively optimize the first model, and the network device uses the collected second data to iteratively optimize the second model. The first network element fuses the intermediate results of the iteration of the second network element and the network device. When the parameters of the fused model converge, the models of the second network element and the network device are decomposed to obtain their respective models.

[0169] (2) It can perform intelligent joint analysis based on the NWDAF architecture, thereby enabling the modeling of operator data and application data without direct interaction and ensuring the data security and no leakage of each party. While protecting data privacy, it can intelligently learn and analyze, realize the linkage between operators and applications, and jointly complete the training of the model.

[0170] Figure 7 This is a schematic diagram illustrating the implementation flow of the data analysis method according to an embodiment of the present invention, as shown below. Figure 7 As shown, the method includes steps 701 to 706:

[0171] Step 701: The first network element obtains the second request; the second request carries a first analysis identifier; the second request is used to request joint inference through the established alliance.

[0172] Step 702: The first network element determines the first model identifier corresponding to the first analysis identifier based on the first analysis identifier and the correspondence between the model identifier and the analysis identifier; and determines the corresponding second network element and network device based on the first model identifier.

[0173] Here, we assume that the first analysis identifier is represented by Analytics ID and the first model identifier is represented by ModeID. If Analytics ID = 1, based on the correspondence between the model identifier and the analysis identifier, we determine that ModeID = 2. Then, based on ModeID = 2, we determine that the corresponding second network element can be the second network element 1 and the corresponding network device can be the network device 3.

[0174] Step 703: The first network element forwards the second request to the determined second network element and network device respectively.

[0175] Step 704: The second network element receives the second request; and responds to the second request by collecting the first data to be processed, using the first model for inference analysis to infer the first data to be processed, and obtaining the first inference result; and reporting the first inference result to the first network element; the first data to be processed is data within the scope of the operator.

[0176] Here, the second network element can use a conventional reasoning method, using a first model for reasoning analysis, to reason about the first data to be processed and obtain a first reasoning result.

[0177] Step 705: The network device receives the second request; and responds to the second request by collecting the second data to be processed; using the second model for inference analysis, it performs inference on the second data to be processed to obtain a second inference result; and reports the second inference result to the first network element; the second data to be processed is data outside the scope of the operator.

[0178] Here, the network device can use a conventional reasoning method to reason about the second data to be processed using a second model for reasoning analysis, and obtain a second reasoning result.

[0179] Step 706: The first network element receives the first inference result and the second inference result; the first inference result and the second inference result are fused to obtain the final inference result.

[0180] Here, the first network element can use a fusion algorithm to fuse the first inference result and the second inference result to obtain the final inference result.

[0181] The fusion algorithm includes, but is not limited to, linear weighted fusion algorithm, cross fusion algorithm, etc.

[0182] Understandably, the first network element will send the final inference result to the entity that sent the first request.

[0183] Figure 8 This is a schematic diagram of data inference. The NWDAF coordinator corresponds to the first network element, the NWDAF client corresponds to the second network element, and the third-party AI unit corresponds to the network device. Figure 8 As shown, the data reasoning process includes:

[0184] First, assuming the requesting entity is an AF network element, the AF network element sends a second request to the NEF network element; the second request carries the first analysis identifier.

[0185] The second request may also carry information about the analysis samples, such as the sample ID, sample collection time, or region, etc., with the aim of synchronizing data collection among alliance clients and maintaining the consistency and validity of training data. For example, in QoE joint modeling, data from the same user's use of a certain service needs to be correlated at least in terms of time.

[0186] It should be noted that there may be one or more AF network elements. In addition to requesting, the AF network element can also subscribe to the joint inference analysis results via a subscription method.

[0187] Second, the NEF network element forwards the second request sent by the AF network element to the NWDAF coordinator.

[0188] It should be noted that when the requesting entity is NF, AF, or OAM, the NF network element sends the second request to the NWDAF coordinator.

[0189] Third, the NWDAF coordinator determines the first model identifier corresponding to the first analysis identifier based on the first analysis identifier and the correspondence between model identifiers and analysis identifiers stored locally; based on the first model identifier, it determines the corresponding NWDAF client and third-party AI unit. Then, it initiates the joint inference process, sending the second request to the determined NWDAF client and third-party AI unit respectively.

[0190] Fourth, the NWDAF client collects the first set of data to be processed based on the Analytics ID or information from the analysis sample. The third-party AI unit collects the second set of data to be processed.

[0191] The first data to be processed can be data within the scope of the operator. The second data to be processed can be data outside the scope of the operator, such as application data.

[0192] Fifth, the NWDAF client uses the first set of data to be processed to perform inference, obtains the inference result μ1, and sends it to the NWDAF coordinator; the third-party AI unit uses the second set of data to be processed to perform inference, obtains the inference result μ2, and sends it to the NWDAF coordinator.

[0193] Sixth, the NWDAF coordinator will aggregate and merge the inference results μ1 from the NWDAF client and μ2 from the third-party AI unit to form the final inference analysis result. This final analysis result will then be returned to the requesting entity.

[0194] It should be noted that if the training unit and inference unit of the NWDAF client are separate, then the NWDAF client refers to the inference unit of the NWDAF client.

[0195] In this embodiment of the invention, establishing an alliance between the network device, the first network element, and the second network element has the following advantages:

[0196] (1) Joint reasoning analysis is achieved through the network device, the first network element and the second network element.

[0197] (2) The second network element performs data reasoning on the first data to be processed collected, the network device performs data reasoning on the second data collected, and the first network element fuses the reasoning results of the second network element and the network device respectively to obtain the fused reasoning result.

[0198] To implement the data analysis method of the present invention, the present invention also provides a data analysis system. Figure 9 This is a schematic diagram of the composition structure of the data analysis system according to an embodiment of the present invention, such as... Figure 9 As shown, the system includes:

[0199] The first network element 91 is used to receive a first request; the first request is used to request the establishment of an alliance between the network device, the first network element, and the second network element; the first request carries first information and authentication information related to the network device.

[0200] The first network element 91 is also used to perform authentication using the authentication information in the first request; if the authentication is successful, the first request is forwarded to the second network element.

[0201] The second network element 92 is used to determine whether to agree to establish an alliance using the first information in the first request; if the alliance is agreed to be established, an alliance is established between the network device, the first network element and the second network element, so that the second network element and the network device collect different types of data, and the alliance performs joint training of a model for inference analysis based on the collected different types of data.

[0202] In one embodiment, the system further includes: a network device;

[0203] The second network element 92 is also used to collect first data; the first data represents data within the scope of the operator.

[0204] The network device is used to collect second data; the second data represents data outside the scope of the operator.

[0205] In one embodiment, the second network element is further configured to negotiate with the network device to ensure that the collected first data and second data meet preset conditions.

[0206] In one embodiment, the second network element 92 is further configured to use the collected first data to train a first model for inference analysis, obtain intermediate training results of the first model, and send the intermediate training results of the first model to the first network element.

[0207] The network device is further configured to use the collected second data to train a second model for inference analysis, obtain intermediate training results of the second model, and send the intermediate training results of the second model to the first network element.

[0208] In one embodiment, the first network element 91 is further configured to receive intermediate training results of a first preset model sent by the second network element; and receive intermediate training results of a second model sent by the network device; fuse the intermediate training results of the first model and the intermediate training results of the second model to obtain fused model parameters; and determine whether the fused model parameters converge; if it is determined that the fused model parameters do not converge, the fused model parameters are decomposed into first intermediate parameters and second intermediate parameters; and the first intermediate parameters are sent to the second network element, and the second intermediate parameters are sent to the network device; wherein, the first intermediate parameters are used for the second network element to continue training the first model, and the second intermediate parameters are used for the network device to continue training the second model; and so on, until the fused model parameters converge; when the fused model parameters converge, the fused model parameters are decomposed into first model parameters and second model parameters; and the first model parameters are sent to the second network element, and the second model parameters are sent to the network device; wherein, the first model parameters are used for the second network element to obtain the first model, and the second model parameters are used for the network device to obtain the second model.

[0209] In one embodiment, the first network element 91 is further configured to determine the model identifier corresponding to the converged model parameters, establish a correspondence between the model identifier and the analysis identifier, and register the model identifier and the analysis identifier with the third network element.

[0210] In one embodiment, the first network element is further configured to obtain a second request; the second request carries a first analysis identifier; the second request is configured to request joint inference through an established consortium; and based on the first analysis identifier and the correspondence between the model identifier and the analysis identifier, determine a first model identifier corresponding to the first analysis identifier; based on the first model identifier, determine the corresponding second network element and network device; and forward the second request to the determined second network element and network device respectively.

[0211] In one embodiment, the second network element 92 is further configured to receive the second request; and in response to the second request, collect first data to be processed, use a first model for inference analysis to infer the first data to be processed, obtain a first inference result; and report the first inference result to the first network element; the first data to be processed is data within the scope of the operator;

[0212] The network device is further configured to receive the second request; and respond to the second request by collecting second data to be processed; using a second model for inference analysis to infer the second data to be processed and obtain a second inference result; and report the second inference result to the first network element; the second data to be processed is data outside the scope of the operator.

[0213] In one embodiment, the first network element 91 is further configured to receive the first inference result and the second inference result; and to fuse the first inference result and the second inference result to obtain the final inference result.

Claims

1. A data analysis method, characterized by, The method includes: The first network element receives a first request; the first request is used to request the establishment of an alliance between the network device, the first network element, and the second network element; the first request carries first information and authentication information related to the network device; The first network element uses the authentication information in the first request to perform authentication; if the authentication is successful, the first request is forwarded to the second network element. The second network element uses the first information in the first request to determine whether it agrees to establish an alliance; if it agrees to establish an alliance, an alliance is established between the network device, the first network element, and the second network element, so that the second network element and the network device can collect different types of data, and the alliance can jointly train a model for inference analysis based on the collected different types of data. The second network element uses the collected first data to train the first model for inference analysis, obtains the intermediate training results of the first model, and sends the intermediate training results of the first model to the first network element. The network device uses the collected second data to train the second model for inference analysis, obtains the intermediate training results of the second model, and sends the intermediate training results of the second model to the first network element. The first network element receives the intermediate training results of the first model sent by the second network element; and receives the intermediate training results of the second model sent by the network device; and fuses the intermediate training results of the first model and the intermediate training results of the second model to obtain the fused model parameters; The first network element determines whether the fused model parameters have converged. If it is determined that the fused model parameters have not converged, the fused model parameters are decomposed into a first intermediate parameter and a second intermediate parameter. The first intermediate parameter is sent to the second network element, and the second intermediate parameter is sent to the network device. The first intermediate parameter is used by the second network element to continue training the first model, and the second intermediate parameter is used by the network device to continue training the second model. This process continues until the fused model parameters converge. When the fused model parameters converge, the fused model parameters are decomposed into a first model parameter and a second model parameter. The first model parameter is sent to the second network element, and the second model parameter is sent to the network device. The first model parameter is used by the second network element to obtain the first model, and the second model parameter is used by the network device to obtain the second model. The method further includes, after establishing an alliance among the network device, the first network element, and the second network element: The second network element collects the first data; the first data represents data within the scope of the operator. The network device collects second data; the second data represents data outside the scope of the operator.

2. The method of claim 1, wherein, Before the second network element collects the first data and the network device collects the second data, the method further includes: The second network element negotiates with the network device to ensure that the collected first and second data meet the preset collection conditions.

3. The method according to claim 1 or 2, characterized in that, The method further includes: The first network element determines the model identifier corresponding to the converged model parameters and establishes a correspondence between the model identifier and the analysis identifier; and registers the model identifier and the analysis identifier with the third network element.

4. The method of claim 3, wherein, The method further includes: The first network element obtains a second request; the second request carries a first analysis identifier; the second request is used to request joint inference through the established consortium; The first network element determines the first model identifier corresponding to the first analysis identifier based on the first analysis identifier and the correspondence between the model identifier and the analysis identifier; and determines the corresponding second network element and network device based on the first model identifier. The first network element forwards the second request to the determined second network element and network device respectively.

5. The method of claim 4, wherein, The method includes: The second network element receives the second request; and responds to the second request by collecting the first data to be processed, using the first model for inference analysis to infer the first data to be processed, and obtaining the first inference result; and reporting the first inference result to the first network element; the first data to be processed is data within the scope of the operator. The network device receives the second request; and responds to the second request by collecting second data to be processed; using a second model for inference analysis, it performs inference on the second data to be processed to obtain a second inference result; and reports the second inference result to the first network element; the second data to be processed is data outside the scope of the operator.

6. The method of claim 5, wherein, The method further includes: The first network element receives the first inference result and the second inference result; it then fuses the first inference result and the second inference result to obtain the final inference result.

7. A data analysis system, characterized by include: The first network element is used to receive the first request; The first request is used to request the establishment of an alliance between the network device, the first network element, and the second network element; The first request carries first information and authentication information related to the network device; The first network element is also used to perform authentication using the authentication information in the first request; if the authentication is successful, the first request is forwarded to the second network element. The second network element is used to determine whether to agree to establish an alliance by using the first information in the first request; If an alliance is agreed to be established, an alliance is established between the network device, the first network element, and the second network element, so that the second network element and the network device can collect different types of data, and jointly train a model for inference analysis based on the collected different types of data through the alliance. The second network element is also used to train the first model for inference analysis using the collected first data to obtain the intermediate training results of the first model; And send the intermediate training results of the first model to the first network element; A network device is used to train a second model for inference analysis using the collected second data, and to obtain intermediate training results of the second model. The intermediate training results of the second model are then sent to the first network element. The first network element is also used to receive the intermediate training results of the first model sent by the second network element; It receives the intermediate training results of the second model sent by the network device; it fuses the intermediate training results of the first model and the intermediate training results of the second model to obtain the fused model parameters; And determine whether the fused model parameters converge; If it is determined that the fused model parameters do not converge, the fused model parameters are decomposed into a first intermediate parameter and a second intermediate parameter; the first intermediate parameter is sent to the second network element, and the second intermediate parameter is sent to the network device; wherein, the first intermediate parameter is used for the second network element to continue training the first model, and the second intermediate parameter is used for the network device to continue training the second model; and so on, until the fused model parameters converge; when the fused model parameters converge, the fused model parameters are decomposed into a first model parameter and a second model parameter; the first model parameter is sent to the second network element, and the second model parameter is sent to the network device; wherein, the first model parameter is used for the second network element to obtain the first model, and the second model parameter is used for the network device to obtain the second model; The second network element is also used to collect first data; the first data represents data within the scope of the operator. The network device is also used to collect second data; the second data represents data outside the scope of the operator.

8. The system according to claim 7, characterized in that, The second network element is further configured to negotiate with the network device before the second network element collects the first data and the network device collects the second data, so as to ensure that the collected first data and the second data meet the preset collection conditions.

9. The system according to claim 7 or 8, characterized in that, The first network element is also used to determine the model identifier corresponding to the converged model parameters, and establish the correspondence between the model identifier and the analysis identifier; and register the model identifier and the analysis identifier with the third network element.

10. The system according to claim 9, characterized in that, The first network element is further configured to obtain a second request; the second request carries a first analysis identifier; the second request is configured to request joint inference through the established consortium; and based on the first analysis identifier and the correspondence between the model identifier and the analysis identifier, determine a first model identifier corresponding to the first analysis identifier; based on the first model identifier, determine the corresponding second network element and network device; and forward the second request to the determined second network element and network device respectively.

11. The system according to claim 10, characterized in that, The second network element is also used to receive the second request; and respond to the second request, collect the first data to be processed, and use the first model for reasoning analysis to reason about the first data to be processed to obtain the first reasoning result; The first inference result is reported to the first network element; the first data to be processed is data within the scope of the operator. The network device is also configured to receive the second request; In response to the second request, the system collects the second data to be processed; and uses the second model for reasoning analysis to reason about the second data to be processed, thereby obtaining the second reasoning result. The second inference result is then reported to the first network element; the second data to be processed is data outside the scope of the operator.

12. The system according to claim 11, characterized in that, The first network element is also used to receive the first inference result and the second inference result; and to fuse the first inference result and the second inference result to obtain the final inference result.

Images