Model collaboration method and device, communication equipment, storage medium and computer program product

Abstract

The invention discloses a model collaboration method and device, communication equipment, a storage medium and a computer program product. The method comprises the steps that a first network element receives first index data sent by a plurality of nodes; the plurality of nodes comprise a first node and a plurality of candidate nodes, and first index data of the first node is sent through a first message used for requesting a data analysis task; determining a second node participating in the data analysis task according to the first index data of the plurality of candidate nodes; indicating the first node and the second node to collect second index data respectively, and obtaining the second index data from the first node and the second node; the first model is trained according to the second index data of the first node and the second node, a first parameter set is obtained, and values of at least part of parameters in the first parameter set are different from values of corresponding parameters after training of the first model is completed; and sending the first parameter set to the first node, so that the first node configures the second model according to the first parameter set.

Description

Technical Field

[0001] This invention relates to the field of communication technology, specifically to a model collaboration method, apparatus, communication equipment, storage medium, and computer program product. Background Technology

[0002] With the continuous development of 5G-A (5G-Advanced) mobile communication networks, each network element in the wireless network possesses a certain computing power for tasks such as data processing, analysis, and model inference. This distributed computing power enables the wireless network to better support various applications and services, such as edge operation and real-time data processing. By providing lower latency, higher real-time performance, and more flexible allocation of computing power, it meets the computing needs of mobile networks for rapid response and proximity to users, promoting the intelligence of mobile networks and the enrichment of application scenarios.

[0003] The ubiquitous computing power of wireless networks has become a major trend in the development of 5G-A mobile communication networks. However, deploying intelligent models based on ubiquitous computing power still faces many challenges. On one hand, the limited size and energy consumption of network elements in a wireless network, due to their deployment locations, may cause performance bottlenecks or insufficient computing power for some elements when processing complex models and intelligent tasks. Therefore, their capabilities lag significantly behind traditional cloud-based data centers. On the other hand, deploying intelligent models based on ubiquitous computing power requires different models and appropriate computing power allocation strategies to ensure that intelligent tasks at various locations can be efficiently distributed and executed within the wireless network. This makes intelligent model collaboration more complex. Summary of the Invention

[0004] To address the existing technical problems, embodiments of the present invention provide a model collaboration method, apparatus, communication device, storage medium, and computer program product.

[0005] To achieve the above objectives, the technical solution of this invention is implemented as follows:

[0006] In a first aspect, embodiments of the present invention provide a model collaboration method, the method being applied to a first network element, the method comprising: receiving first indicator data sent by multiple nodes; the multiple nodes including a first node and multiple candidate nodes, wherein the first indicator data of the first node is sent through a first message, the first message being used to request a data analysis task;

[0007] The second node participating in the data analysis task is determined based on the first indicator data of the multiple candidate nodes;

[0008] The first node and the second node are respectively instructed to collect the second indicator data, and to obtain the second indicator data from the first node and the second node;

[0009] The first model is trained based on the second index data of the first node and the second node to obtain a first parameter set, wherein at least some parameters in the first parameter set have values ​​different from the corresponding parameters after the first model is trained.

[0010] The first parameter set is sent to the first node so that the first node can configure the second model according to the first parameter set. The second model has the same structure as the first model.

[0011] In the above scheme, the first message includes: task type, sender name, receiver name, tag corresponding to the indicator data, acquisition duration of the indicator data, and storage address of the indicator data.

[0012] In the above scheme, before determining the second node participating in the data analysis task based on the first indicator data of the plurality of candidate nodes, the method further includes: determining whether to execute the data analysis task based on the indicator data of the first node;

[0013] Accordingly, determining the second node participating in the data analysis task based on the first indicator data of the plurality of candidate nodes includes: after determining to execute the data analysis task, determining the second node participating in the data analysis task based on the first indicator data of the plurality of candidate nodes.

[0014] In the above scheme, determining the second node participating in the data analysis task based on the first indicator data of the plurality of candidate nodes includes: constructing a first matrix and a second matrix based on the first indicator data of the plurality of candidate nodes; the first matrix is ​​a matrix related to the candidate nodes, the number of the first matrices is the number of categories of indicator data, and the second matrix is ​​a matrix related to the indicator data; using the first matrix and the second matrix to determine the score of each candidate node, and determining the second node based on the score.

[0015] In the above scheme, determining the score of each candidate node using the first matrix and the second matrix includes: summing the row vectors of the first matrix and the second matrix respectively, and standardizing the column vectors obtained by the row vector summation to obtain a first weight vector corresponding to the first matrix and a second weight vector corresponding to the second matrix respectively; multiplying the second weight vector by each first weight vector to obtain the score of each candidate node.

[0016] In the above scheme, instructing the first node and the second node to collect the second indicator data, and obtaining the second indicator data from the first node and the second node, respectively, includes: sending a second message to the first node and the second node, the second message being used to instruct the collection of indicator data; receiving a third message sent by the first node and the second node, the third message including the storage address of the indicator data; and obtaining the second indicator data collected by the first node and the second node according to the storage address of the indicator data.

[0017] In the above scheme, the second message includes: task type, sender name, receiver name, indicator data, tag corresponding to the indicator data, acquisition time of the indicator data, and storage address of the indicator data; the third message also includes: task type, sender name, receiver name, indicator data, tag corresponding to the indicator data, and acquisition time of the indicator data.

[0018] In the above scheme, the first network element and each node exchange messages through the first interface.

[0019] Secondly, embodiments of the present invention provide a model collaboration method, the method being applied to a first node, the method including sending a first message to a first network element, the first message being used to request a data analysis task, the first message including first indicator data;

[0020] The second indicator data is collected based on the instructions of the first network element, so that the first network element can obtain the second indicator data;

[0021] The system receives a first parameter set from the first network element and configures a second model according to the first parameter set. The first parameter set is obtained by the first network element after training the first model based on at least the second indicator data. The values ​​of at least some parameters in the first parameter set are different from the values ​​of the corresponding parameters after the first model is trained. The second model has the same structure as the first model.

[0022] In the above scheme, the first message includes: task type, sender name, receiver name, tag corresponding to the indicator data, acquisition duration of the indicator data, and storage address of the indicator data.

[0023] In the above scheme, the step of collecting second indicator data based on the instruction of the first network element so that the first network element can obtain the second indicator data includes: the first node receiving a second message from the first network element, the second message being used to instruct the collection of indicator data; collecting the second indicator data based on the second message, and sending a third message to the first network element, the third message including the storage address of the indicator data.

[0024] In the above scheme, the second message includes: task type, sender name, receiver name, indicator data, tag corresponding to the indicator data, acquisition time of the indicator data, and storage address of the indicator data; the third message also includes: task type, sender name, receiver name, indicator data, tag corresponding to the indicator data, and acquisition time of the indicator data.

[0025] In the above scheme, the first node and the first network element exchange messages through the first interface.

[0026] Thirdly, embodiments of the present invention also provide a model collaboration device, which is applied to a first network element and includes: a first communication unit, a first processing unit, and a training unit; wherein...

[0027] The first communication unit is used to receive first indicator data sent by multiple nodes; the multiple nodes include a first node and multiple candidate nodes, and the first indicator data of the first node is sent through a first message, which is used to request a data analysis task.

[0028] The first processing unit is configured to determine a second node participating in the data analysis task based on the first indicator data of the plurality of candidate nodes;

[0029] The first communication unit is further configured to instruct the first node and the second node to collect the second indicator data, and to obtain the second indicator data from the first node and the second node, respectively;

[0030] The training unit is used to train the first model based on the second index data of the first node and the second node to obtain a first parameter set, wherein at least some parameters in the first parameter set have values ​​different from the values ​​of the corresponding parameters after the first model has been trained.

[0031] The first communication unit is further configured to send the first parameter set to the first node so that the first node can configure the second model according to the first parameter set, wherein the second model has the same structure as the first model.

[0032] Fourthly, embodiments of the present invention also provide a model collaboration device, the device being applied to a first node, the device comprising: a second communication unit and a second processing unit; wherein,

[0033] The second communication unit is used to send a first message to the first network element. The first message is used to request a data analysis task and includes first indicator data.

[0034] The second processing unit is used to collect second indicator data based on the instructions of the first network element, so that the first network element can obtain the second indicator data;

[0035] The second communication unit is further configured to receive a first set of parameters from the first network element;

[0036] The second processing unit is further configured to configure a second model according to the first parameter set; wherein the first parameter set is obtained by the first network element after training the first model based on the second index data, and the values ​​of at least some parameters in the first parameter set are different from the values ​​of the corresponding parameters after the first model is trained, and the second model has the same structure as the first model.

[0037] Fifthly, embodiments of the present invention also provide a communication device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method applied to a first network element or a first node as described in the embodiments of the present invention.

[0038] In a sixth aspect, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method applied to a first network element or a first node as described in the embodiments of the present invention.

[0039] In a seventh aspect, embodiments of the present invention also provide a computer program product, including a computer program that, when executed by a processor, implements the method described in the embodiments of the present invention for application to a first network element or a first node.

[0040] This invention provides a model collaboration method, apparatus, communication device, storage medium, and computer program product. The method involves a first network element receiving first indicator data sent by a first node to initiate a data analysis task. A second node participating in the data analysis task is determined based on first indicator data sent by multiple candidate nodes. A first model in the first network element is trained using second indicator data collected by the first and second nodes to obtain a first parameter set, which is then sent to the first node. This allows the first node to configure a second model with the same structure as the first model in the first network element based on the first parameter set. This enables the first node to process intelligent tasks based on the second model. In other words, through intelligent model collaboration between network elements, the advantages of each element are complemented, computing power is fully utilized, and overall network gain is improved. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating the model collaboration method according to an embodiment of the present invention. Figure 1 ;

[0042] Figure 2 This is a schematic diagram of the method for determining the second node in the model collaboration method of this invention.

[0043] Figure 3 This is a schematic diagram of the hierarchical analysis model in the model collaboration method of this invention.

[0044] Figure 4 This is a schematic diagram of the process for obtaining the second indicator data in the model collaboration method of this invention.

[0045] Figure 5 This is a schematic diagram of the architecture of the model collaboration method according to an embodiment of the present invention;

[0046] Figure 6 This is a flowchart illustrating the model collaboration method according to an embodiment of the present invention. Figure 2 ;

[0047] Figure 7 This is a schematic diagram of the process for collecting the second indicator data in the model collaboration method of this invention.

[0048] Figure 8 This is a schematic diagram of the interaction process of the model collaboration method according to an embodiment of the present invention;

[0049] Figure 9 This is a schematic diagram of the composition structure of the model collaboration device according to an embodiment of the present invention. Figure 1 ;

[0050] Figure 10 This is a schematic diagram of the composition structure of the model collaboration device according to an embodiment of the present invention. Figure 2 ;

[0051] Figure 11 This is a schematic diagram of the hardware composition structure of a communication device according to an embodiment of the present invention. Detailed Implementation

[0052] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0053] The technical solutions of this invention can be applied to various communication systems, such as GSM (Global System of Mobile communication), LTE (Long Term Evolution), or 5G systems. Optionally, a 5G system or 5G network can also be referred to as a New Radio (NR) system or NR network.

[0054] For example, the communication system used in this embodiment of the invention may include network devices and terminal devices (also referred to as terminals, communication terminals, etc.); the network device may be a device that communicates with the terminal device. The network device can provide communication coverage within a certain area and can communicate with terminals located within that area. Optionally, the network device may be a base station in various communication systems, such as an evolved Node B (eNB) in an LTE system, or a gNB in ​​a 5G or NR system.

[0055] It should be understood that devices with communication functions in the network / system of this application embodiment can be referred to as communication devices. Communication devices may include network devices and terminals with communication functions. Network devices and terminal devices can be the specific devices described above, which will not be repeated here. Communication devices may also include other devices in the communication system, such as network controllers, mobility management entities, and other network entities. This embodiment of the present invention does not limit these.

[0056] It should be understood that the terms "system" and "network" are often used interchangeably in this document. The term "and / or" in this document merely describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Furthermore, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0057] The terms “first,” “second,” etc., used in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0058] Before providing a detailed description of the sensing method in the embodiments of the present invention, a brief description of the related technologies will be given first.

[0059] Network elements in mobile communication networks include, but are not limited to, User Equipment (UE), base stations, and functional entities. Functional entities, as key components of the intelligent network architecture, provide specific functions or services to meet different service requirements. For example, the Operation and Maintenance Center (OMC) is primarily responsible for the centralized management of network operation and maintenance in the wireless network, providing functions such as fault detection, performance management, and configuration management. In mobile communication networks, these network elements connect and communicate through standard interfaces to complete various tasks and functions within the network.

[0060] With the arrival of the 5G-A era, deploying computing power and carrying out intelligent transformation within traditional network elements such as OMCs and base stations has become a major development trend. For example, the main applications of OMCs include fault monitoring and management, performance management and optimization, configuration management, and resource management. After intelligent transformation of OMCs, computing hardware is widely deployed on the base station side, transforming some capabilities or applications within the base station into intelligent ones, such as anomaly detection, real-time data analysis and decision-making, network optimization, and resource allocation.

[0061] This demonstrates that each network element in a mobile wireless network possesses a certain level of computing power, which can be used for tasks such as data processing, analysis, and model inference. This distributed computing power enables mobile wireless networks to better support various applications and services, such as edge computing and real-time data processing. Its advantages lie in providing lower latency, higher real-time performance, and more flexible allocation of computing power, thereby meeting the computing needs of mobile networks for rapid response and proximity to users, promoting the intelligence of mobile networks and enriching application scenarios.

[0062] As a key architecture for the development of 5G-A mobile communication networks, ubiquitous computing power in wireless networks can provide computing, storage, and other network services at the network edge between the cloud and terminal devices. However, due to the limited size and energy consumption of the deployment location, some wireless network elements may experience performance bottlenecks or insufficient computing power when processing complex models and intelligent tasks. Therefore, their capabilities are significantly different from those of traditional cloud-based data centers.

[0063] Deploying intelligent models directly on ubiquitous computing power requires different models and appropriate computing power allocation strategies to ensure that intelligent tasks at various locations can be efficiently distributed and executed within the wireless network. This makes intelligent model collaboration more complex. Therefore, how to deploy intelligent models on traditional wireless network elements has become an urgent problem to be solved.

[0064] For example, in wireless network application scenarios, taking base stations and OMC as examples, the intelligent solution of base stations has high timeliness, but poor globality, lacks understanding and perception of surrounding base stations, and has limited computing power; on the contrary, OMC has globality, can comprehensively grasp the situation of surrounding base stations, has a lot of computing power, but its timeliness is lower than that of base stations.

[0065] Based on this, embodiments of the present invention provide a model collaboration method, which aims to achieve full utilization of computing power and improve overall network gain by intelligently collaborating between network elements to complement each other's advantages.

[0066] Figure 1 This is a flowchart illustrating the model collaboration method according to an embodiment of the present invention. Figure 1 ;like Figure 1 As shown, the method is applied to a first network element, and the method includes:

[0067] Step 101: Receive first indicator data sent by multiple nodes; the multiple nodes include a first node and multiple candidate nodes, the first indicator data of the first node is sent through a first message, the first message is used to request a data analysis task;

[0068] Step 102: Determine the second node to participate in the data analysis task based on the first indicator data of the multiple candidate nodes;

[0069] Step 103: Instruct the first node and the second node to collect the second indicator data, and obtain the second indicator data from the first node and the second node respectively;

[0070] Step 104: Train the first model based on the second index data of the first node and the second node to obtain a first parameter set, wherein at least some parameters in the first parameter set have values ​​different from the corresponding parameters after the first model has been trained.

[0071] Step 105: Send the first parameter set to the first node so that the first node can configure the second model according to the first parameter set. The second model has the same structure as the first model.

[0072] Here, the first network element can be a network element in the wireless network that provides the functional requirements for services, such as a functional entity. The functional entity used varies depending on the domain and scenario. For example, in the mobile communication field, functional entities such as the Broadcast Control Functional Entity (BCFE) handle broadcast functions to ensure the effective utilization of wireless resources; while OMC is used for centralized management of wireless network operation and maintenance, providing functions such as fault detection, performance management, and configuration management. The first node and second node can be access network devices that support model collaboration, such as base stations.

[0073] In this embodiment, the first network element receives first indicator data sent by the first node and multiple candidate nodes. The first indicator data of the first node is sent through a first message for requesting a data analysis task. That is, the first network element receives a first message from the first node for requesting a data analysis task. The first message includes first indicator data, which may be data actively collected and reported by the first node for requesting data analysis, and includes multiple related indicators (or features).

[0074] For example, in a load balancing application scenario, when the first node (i.e., the base station) detects a network overload problem during routine operation and maintenance, it needs to send a first message containing first indicator data to the first network element (i.e., the OMC) responsible for centralized network operation and maintenance. This allows the first network element to initiate a data analysis task for the network overload problem. The first indicator data includes multiple indicators (or characteristics) related to the network overload problem, such as: Reference Signal Received Power (RSRP), Signal-to-Interference plus Noise Ratio (SINR), and Physical Resource Block (PRB).

[0075] It should be noted that candidate nodes are nodes that can provide indicator data related to the first indicator data sent by the first node, so as to select a second node to participate in the data analysis task. This embodiment of the invention does not impose specific restrictions on the method by which candidate nodes upload the first indicator data. For example, the first network element can periodically or non-periodically receive the first indicator data reported by the candidate nodes by sending collection instructions to them; that is, the candidate nodes passively report the first indicator data. Alternatively, the first network element can receive the first indicator data actively reported by the candidate nodes.

[0076] In some implementations, the first message includes: task type, sender name, receiver name, tag corresponding to the indicator data, acquisition duration of the indicator data, and storage address of the indicator data.

[0077] In this embodiment, the first indicator data received by the first network element can be sent through a first message. The first message is used to request a data analysis task for the first node. In addition to including the first indicator data, the first message may also include: task type, sender name, receiver name, tag corresponding to the indicator data, acquisition duration of the indicator data, and storage address of the indicator data.

[0078] Among them, the task type indicates the purpose of the first message, which may include task category and task subclass; the sender name indicates the name of the network element that sends the first message; the receiver name indicates the name of the network element that receives the first message; the label corresponding to the indicator data indicates the label corresponding to the feature of the first indicator data; the acquisition time of the indicator data is the time for the first node to collect the first indicator data; and the storage address of the indicator data is the address where the first node stores the first indicator data.

[0079] For example, based on the above-mentioned load balancing application scenario, the configuration of the first message received by the first network element from the first node is shown in Table 1. The first message may include task category, task subclass, sender name, receiver name, parameter set, first indicator data, indicator data tag, indicator data acquisition duration and indicator data storage address. Each item has a corresponding field name, field description, field type and field value.

[0080] Table 1

[0081] Field Name Field Description Field type Field value task_type Task Categories string error report subtask_type Task subclass string data_analysis star_id Sender Name string gNB_1 obj_id Receiver name string OMC_1 para_set Parameter set list null data_during Duration of time to obtain indicator data int 600 data_x First indicator data list RSRP; SINR; PRB; OPT data_y Labels corresponding to indicator data list Thr data_adr Storage address of indicator data string xxx.xxx

[0082] In this context, the task category (task_type) is a string field, and the corresponding field value is error_report, indicating that the task category in the current scenario is reporting network overload issues; the task subcategory (subtask_type) is a string field, and the corresponding field value is data_analysis, indicating that the task subcategory in the current scenario is request data analysis; the sender name (star_id) is a string field, and the corresponding field value is gNB_1, indicating that the first base station is sending the first message in the current scenario; the receiver name (obj_id) is a string field, and the corresponding field value is OMC_1, indicating that the current receiver name is the first base station. In this scenario, the OMC receives the first message; the parameter set (para_set) is a list, and since the parameter set is not involved in the process of the first node sending the first message to the first network element, the current field value is empty; the acquisition time of the indicator data (data_during) is an integer, and the corresponding field value is 600, which can be in seconds (S); the first indicator data (data_x) is a list, and the corresponding field value can be RSRP, SINR, PRB, and a custom field (OPT). OPT can be empty and is used to meet the special indicator requirements of the node. For example, in a load balancing scenario, OPT can also be the block error rate (BLER); the label corresponding to the indicator data (data_y) is a list, and the corresponding field value is the label corresponding to the first indicator data, such as a value representing network overload in the current scenario; the storage address of the indicator data (data_adr) is a string, and the corresponding field value can be a specific storage address.

[0083] In this embodiment, after receiving the first indicator data sent by the first node and multiple candidate nodes, the first network element determines the second node to participate in the analysis of the first indicator data sent by the first node based on the first indicator data sent by the multiple candidate nodes.

[0084] In some implementations... Figure 2 This is a schematic diagram of the second node determination method in the model collaboration method of this invention; as shown. Figure 2 As shown, step 102 includes:

[0085] Step 201: Construct a first matrix and a second matrix based on the first indicator data of the plurality of candidate nodes; the first matrix is ​​a matrix related to the candidate nodes, the number of the first matrices is the number of categories of indicator data, and the second matrix is ​​a matrix related to the indicator data.

[0086] In this embodiment, in order to determine the second node from multiple candidate nodes, the first network element constructs a hierarchical analysis model based on the first indicator data of the multiple candidate nodes. The hierarchical analysis model includes multiple first matrices of the same type as the first indicator data and related to the candidate nodes, and second matrices related to the indicator data.

[0087] For example, Figure 3 This is a schematic diagram of the hierarchical analysis model in the model collaboration method of this invention; as shown below. Figure 3 As shown, the hierarchical analysis model constructed in this embodiment of the invention includes three layers: the target layer, the criterion layer, and the scheme layer.

[0088] The target layer includes the decision objective, i.e. the problem to be solved. In this embodiment, the decision objective corresponding to the target layer is to select a second node to participate in data analysis. The criterion layer includes the factors to be considered or the criteria for decision-making for the decision objective. In this embodiment, the criterion factors corresponding to the criterion layer are the types of first indicator data sent by the first node. The number of criterion factors corresponds to the number of types of first indicator data sent by the first node (for example, the types of data_x in Table 1 include RSRP, SINR, and PRB, i.e., there are 3 criterion factors in the criterion layer). The scheme layer includes alternative schemes for decision-making. In this embodiment, the alternative schemes corresponding to the scheme layer are the first indicator data sent by multiple candidate nodes. The number of alternative schemes is the number of types of first indicator data.

[0089] Furthermore, the first matrix, as a matrix related to the candidate nodes, corresponds to the scheme judgment matrix constructed in the scheme layer. In this embodiment of the invention, the first matrix is ​​an n*n matrix, where n is the number of candidate nodes, and the number of first matrices is the number m of the first indicator data types. That is, each first matrix corresponds to a first indicator data type of a first node. Each row of the first matrix corresponds to a candidate node, for example, rows of the first matrix correspond to candidate nodes 1,…n; each column of the first matrix corresponds to a candidate node, for example, columns of the first matrix correspond to candidate nodes 1,…n; each first matrix corresponds to one data type of the first indicator data sent by the first node, for example, m first matrices correspond to each type of first indicator data. Then, each element in the first matrix represents the influence weight of the candidate node x in the corresponding row on the first indicator data type of the first matrix and the candidate node y in the corresponding column, or the influence weight of the candidate node y in the corresponding column on the first indicator data type of the first matrix and the candidate node x in the corresponding row. The specific influence rules can be preset. For example, the value of the element on the diagonal of each first matrix can be 1. In some optional embodiments, the values ​​of corresponding elements in each first matrix can be determined based on the same type of first index data of two candidate nodes. The present invention does not impose specific restrictions on the calculation method of the element values ​​in each first matrix.

[0090] The second matrix, as a matrix related to the indicator data, corresponds to the criterion judgment matrix constructed in the criterion layer. In this embodiment, the second matrix is ​​an m*m matrix, where m represents the number of types of the first indicator data. The rows and columns of the second matrix correspond to the types of the first indicator data, respectively. Each element in the second matrix represents the influence weight of indicator x in the corresponding row on indicator y in the corresponding column, or the influence weight of indicator y in the corresponding column on indicator x in the corresponding row. The specific influence rules can be preset. For example, the diagonal elements of the second matrix can have a value of 1. In some optional embodiments, the value of the corresponding element can be determined by the value of the indicator in the row vector corresponding to the element and the value of the indicator in the column vector corresponding to the element. This embodiment does not impose specific restrictions on the calculation method of the element values ​​in the second matrix.

[0091] Step 202: Determine the score of each candidate node using the first matrix and the second matrix, and determine the second node based on the score.

[0092] In this embodiment, after constructing the hierarchical analysis model, the score of each candidate node is calculated using multiple first matrices constructed in the scheme layer and the second matrix constructed in the criterion layer. This score represents the score applicable to the data analysis task of the first node, that is, the higher the score of the candidate node, the more suitable it is to participate in the data analysis task of the first node.

[0093] It should be noted that, in this embodiment of the invention, there is no specific limitation on the number of candidate nodes selected as the second node. The candidate node with the highest score or one or more candidate nodes with relatively high scores can be set as the second node.

[0094] In some implementations, step 202 includes: summing the row vectors of the first matrix and the second matrix respectively, and standardizing the column vectors obtained by the row vector summation to obtain a plurality of first weight vectors corresponding to the first matrix and second weight vectors corresponding to the second matrix respectively; multiplying the second weight vectors by each first weight vector to obtain the score of each candidate node.

[0095] In this embodiment, the score of each candidate node is calculated using multiple first matrices constructed in the scheme layer and a second matrix constructed in the criterion layer. Specifically, the row vectors of the multiple first matrices and the second matrix can be summed, and the column vectors obtained by the row vector summation can be standardized to obtain multiple first weight vectors and second weight vectors. The second weight vector is multiplied by each first weight vector to obtain the final score of each candidate node.

[0096] It should be noted that although the first indicator data sent by each candidate node is of the same type as the first indicator data sent by the first node, some candidate nodes may send first indicator data with some values ​​of zero, meaning they lack indicator data of that type. In this case, the candidate node with missing indicator data will have a lower final score, meaning that a candidate node with missing indicator data will definitely not be selected as a second node to participate in the data analysis task.

[0097] In some implementations, before step 102, the method further includes: determining whether to execute the data analysis task based on the indicator data of the first node;

[0098] Accordingly, determining the second node participating in the data analysis task based on the first indicator data of the plurality of nodes includes: after determining to execute the data analysis task, determining the second node participating in the data analysis task based on the first indicator data of the plurality of nodes.

[0099] In this embodiment, before determining the second node to participate in the data analysis task based on the first indicator data of multiple candidate nodes, the first network element needs to determine whether to execute the data analysis task based on the first indicator data sent by the first node. Only when the first network element determines that the data analysis task needs to be executed can the second node to participate in the data analysis task be determined based on the first indicator data of multiple nodes.

[0100] For example, in the above load balancing scenario, if the first network element confirms that it does not need to perform a data analysis task based on the first indicator data sent by the first node, it indicates that the current first node may not have a network overload problem and does not need to perform further data analysis on the first indicator data sent by the first node; if the first network element confirms that it needs to perform a data analysis task based on the first indicator data sent by the first node, it indicates that the current first node may have a network overload problem and needs to perform data analysis on the first indicator data through model training.

[0101] This invention does not impose specific limitations on the method for determining whether a data analysis task should be performed by the first network element. For example, a threshold method can be used. For instance, a corresponding threshold can be preset for each type of first indicator data. When the values ​​of some types of first indicator data received from the first node exceed their corresponding preset thresholds, it can be determined that a data analysis task needs to be performed. To avoid individual data affecting the judgment, the number of first indicator data types exceeding the preset threshold can also be set. That is, only when the number of first indicator data types exceeding the preset threshold exceeds a preset number is it confirmed that a data analysis task needs to be performed.

[0102] In this embodiment, after the first network element determines the second node participating in the data analysis task, it instructs the first node and the second node to collect second instruction data for training the first model, so as to obtain the second instruction data from the first node and the second node.

[0103] In some implementations... Figure 4 This is a schematic diagram of the process for obtaining the second indicator data in the model collaboration method of this invention; as shown below. Figure 4 As shown, step 103 includes:

[0104] Step 301: Send a second message to the first node and the second node respectively. The second message is used to instruct the collection of indicator data.

[0105] In this embodiment, the first network element can send a second message to the first node and the second node respectively to instruct the first node and the second node to collect second indicator data. The second indicator data can be historical data of the same type as the first indicator data that the first node and the second node obtain in real time.

[0106] Step 302: Receive a third message sent by the first node and the second node, wherein the third message includes the storage address of the indicator data.

[0107] In this embodiment, after the first network element sends the second message to the first node and the second node respectively, it needs to receive the third message from the first node and the second node respectively, which includes the storage address of the indicator data, in order to obtain the second indicator data.

[0108] Step 303: Obtain the second indicator data collected by the first node and the second node according to the storage address of the indicator data.

[0109] In this embodiment, after receiving the third message from the first node and the second node respectively, which includes the storage address of the indicator data, the first network element will obtain the second indicator data collected by the first node and the second node based on the storage address of their respective indicator data, so as to use it for subsequent training of the first model.

[0110] In some implementations, the second message includes: task type, sender name, receiver name, indicator data, tag corresponding to the indicator data, acquisition time of the indicator data, and storage address of the indicator data; the third message further includes: task type, sender name, receiver name, indicator data, tag corresponding to the indicator data, and acquisition time of the indicator data.

[0111] In this embodiment, the second message sent by the first network element to the first node and the second node to indicate the collection of the second indicator data includes: task type, sender name, receiver name, indicator data, tag corresponding to the indicator data, acquisition duration of the indicator data, and storage address of the indicator data.

[0112] Among them, the task type indicates the purpose of sending the second message, which may include a task category and a task subcategory; the sender name indicates the name of the network element sending the second message; the receiver name indicates the name of the network element receiving the second message; the tag corresponding to the indicator data indicates the tag corresponding to the feature that the first network element instructs the first node or the second node to collect the second indicator data; the acquisition duration of the indicator data is the duration for the first node or the second node to collect the second indicator data; and the storage address of the indicator data is the address where the first node or the second node stores the second indicator data.

[0113] For example, the configuration of the second message sent by the first network element to the first node and the second node is shown in Table 2. The second message may include task type, sender name, receiver name, parameter set, second indicator data, label corresponding to the indicator data, acquisition time of the indicator data and storage address of the indicator data. Each item in Table 2 has a corresponding field name, field description, field type and field value.

[0114] Table 2

[0115]

[0116]

[0117] In this context, the task category (task_type) is a string field, and the corresponding field value is model_train, indicating that the task category in the current scenario is training the first model used for data analysis to address network overload issues. The task subcategory (subtask_type) is a string field, and the corresponding field value is data_collect, indicating that the task subcategory in the current scenario is instructing the first or second node to collect the second indicator data. The sender name (star_id) is a string field, and the corresponding field value is OMC_1, indicating that the first network element (OMC) is sending the second message in the current scenario. The receiver name (obj_id) is a string field, and the corresponding field value is gNB_1 or g NB_2 indicates that the first or second node (i.e., the first or second base station) is receiving the second message in the current scenario. The parameter set (para_set) is a list. Since the parameter set is not involved in the process of the first network element sending the second message to the first or second node, the current field value is empty. The acquisition time of the indicator data (data_during) is an integer, and the corresponding field value is 600, which can be in seconds (S). The second indicator data (data_x) is a list. The corresponding field value can be RSRP, SINR, PRB, and optional field (OPT). OPT can be empty and is used to meet the special indicator requirements of the node. For example, in the load balancing scenario, OPT can also be the Block Error Rate (BLER). The label corresponding to the indicator data (data_y) is a list. The corresponding field value is the label corresponding to the second indicator data. The storage address of the indicator data (data_adr) is a string. The corresponding field value can be a specific storage address.

[0118] In this embodiment, the third message received by the first network element from the first node and the second node includes, in addition to the storage address of the indicator data, the following: task type, sender name, receiver name, indicator data, the tag corresponding to the indicator data, and the acquisition time of the indicator data.

[0119] Among them, the task type indicates the purpose of sending the third message, which may include task category and task sub-category; the sender name indicates the name of the network element sending the third message; the receiver name indicates the name of the network element receiving the third message; the label corresponding to the indicator data indicates the label corresponding to the feature of the second indicator data collected by the first node or the second node received by the first network element; the acquisition time of the indicator data is the time for the first node or the second node to collect the second indicator data; and the storage address of the indicator data is the address where the first node or the second node stores the second indicator data.

[0120] For example, the configuration of the third message received by the first network element from the first node and the second node is shown in Table 3. The third message may include task type, sender name, receiver name, parameter set, second indicator data, tag corresponding to the indicator data, acquisition time of the indicator data and storage address of the indicator data. Each item in Table 3 has a corresponding field name, field description, field type and field value.

[0121] Table 3

[0122] Field Name Field Description Field type Field value task_type Task Categories string model_train subtask_type Task subclass string data_feedback star_id Sender Name string gNB_1; gNB_2 obj_id Receiver name string OMC_1 para_set Parameter set list null data_during Duration of time to obtain indicator data int 600 data_x Second indicator data list RSRP; SINR; PRB; OPT data_y The label corresponding to the second indicator data list Thr data_adr Storage address of indicator data string http: / / xxx

[0123] In this context, the task category (task_type) is a string field, and the corresponding field value is model_train, indicating that the task category in the current scenario is training the first model used for data analysis to address network overload issues. The task subcategory (subtask_type) is a string field, and the corresponding field value is data_feedback, indicating that the task subcategory in the current scenario is the second indicator data collected by the first or second node. The sender name (star_id) is a string field, and the corresponding field value is gNB_1 or gNB_2, indicating that the third message is sent by the first or second node (i.e., the first or second base station). The receiver name (obj_id) is a string field, and the corresponding field value is O. MC_1 indicates that the first network element (OMC) is currently receiving the third message. The parameter set (para_set) is a list; since the first network element does not receive the third message from the first or second node, this field is currently empty. The data acquisition duration (data_during) is an integer, with a value of 600, which can be in seconds (S), representing the duration for the first or second node to collect the second indicator data. The second indicator data (data_x) is a list; its values ​​can be RSRP, SINR, PRB, and a custom field (OPT). OPT can be empty to meet specific node indicator requirements; for example, in a load balancing scenario, OPT can also be the block error rate (Block). Error Rate (BLER); The label (data_y) field corresponding to the indicator data is of type list, and the corresponding field value is the label corresponding to the second indicator data collected by the first node or the second node; The storage address (data_adr) field is of type string, and the corresponding field value can be a specific storage address, so that the first network element can obtain the second indicator data fed back by the first node or the second node based on the storage address.

[0124] In some optional embodiments, after the first network element sends a second message to the first node and the second node, it receives a first response from the first node and the second node, wherein the first response indicates that the first node and the second node have received the second message.

[0125] In this embodiment, after the first network element sends a second message to the first and second nodes to instruct the collection of second indicator data, it needs to receive a first response from the first and second nodes based on the second message. This confirms that the node has received the second message and has begun collecting the second indicator data within the collection period specified in the second message. For example, the first response received by the first network element may be an acknowledgment character (ACK) from both the first and second nodes.

[0126] As an example, the first response received by the first network element from the first node and the second node can be as shown in Table 4. The first response may include the task type, the sender name and the receiver name. Each item in Table 4 corresponds to a field name, field description, field type and field value.

[0127] Table 4

[0128] Field Name Field Description Field type Field value task_type Task Categories string model_train subtask_type Task subclass string task_ack star_id Sender Name string gNB_1; gNB_2 obj_id Receiver name string OMC_1

[0129] The task category (task_type) is a string field, and the corresponding field value is model_train, indicating that the task category in the current scenario is training the first model used for data analysis to address network overload issues. The task subcategory (subtask_type) is a string field, and the corresponding field value is task_ack, indicating that the task subcategory in the current scenario is the first response (e.g., ACK) from the first or second node. The sender name (star_id) is a string field, and the corresponding field value is gNB_1 or gNB_2, indicating that the first response in the current scenario is from the first or second node (i.e., the first or second base station). The receiver name (obj_id) is a string field, and the corresponding field value is OMC_1, indicating that the first response in the current scenario is received by the first network element (i.e., OMC).

[0130] In this embodiment, after the first network element obtains the second indicator data of the first node and the second node according to the storage address of the indicator data in the third message, it begins to train the first model for data analysis based on the second indicator data, thereby obtaining the corresponding model parameters after the model training is completed.

[0131] It should be noted that there are no specific restrictions on how the first network element obtains the second indicator data based on the storage address of the indicator data. For example, it can obtain the second indicator data of the first node and the second node through the File Transfer Protocol (FTP). After obtaining the second indicator data of the first node and the second node, the training of the first model begins.

[0132] The embodiments of the present invention do not impose specific restrictions on the training process of the first model. The model can be trained and the parameters adjusted by existing neural network methods, so that the model parameters of the first model can be obtained after the model training is completed. Further details are not provided here.

[0133] In this embodiment, after model training is completed and the corresponding model parameters are obtained, a first parameter set is obtained through model distillation. At least some parameters in the first parameter set have values ​​different from the model parameters obtained after the first model training is completed. For example, after model distillation, some parameters in the first parameter set may have been set to zero compared to the model parameters obtained after the first model training is completed, that is, some parameter values ​​are different.

[0134] It should be noted that the model distillation processing method used in the embodiments of the present invention is an existing model compression and acceleration technology. Model distillation processing is mainly used to reduce the model size during deployment, improve inference speed, reduce computing resource consumption, and enhance the robustness and generalization ability of the model, thereby achieving efficient inference and prediction on nodes with limited resources (such as the first node). Further details will not be elaborated here.

[0135] In this embodiment, after the first network element completes the training of the first model and obtains the first parameter set, it sends the first parameter set to the first node so that the first node can configure a second model with the same structure as the first model according to the first parameter set.

[0136] For example, the first network element can send a fourth message including a first parameter set to the first node, so that the first node can extract the first parameter set and configure the second model with the same structure as the first model according to the first parameter set, thereby realizing model collaboration. The configuration of the fourth message sent by the first network element to the first node can be as shown in Table 5. The fourth message may include a task type, a sender name, a receiver name, and a first parameter set. The task type may include a task category and a task subcategory. Each item in Table 4 corresponds to a field name, field description, field type, and field value.

[0137] Table 5

[0138] Field Name Field Description Field type Field value task_type Task Categories string model_train subtask_type Task subclass string para_feedback star_id Sender Name string OMC_1 obj_id Receiver name string gNB_1 para_set First parameter set list <![CDATA[P1=xx;P2=xx;P3=xx;P4=xx]]>

[0139] Among them, the task category (task_type) is a string field, and the corresponding field value is model_train, meaning that the task category in the current scenario is training the first model used for data analysis to address the network overload problem; the task subcategory (subtask_type) is a string field, and the corresponding field value is para_feedback, meaning that the task subcategory in the current scenario is the fourth message fed back from the first network element to the first node; the sender name (star_id) is a string field, and the corresponding field value is OMC_1, meaning that the first network element (i.e., OMC) is feeding back the fourth message in the current scenario; the receiver name (obj_id) is a string field, and the corresponding field value is gNB_1, meaning that the first node (i.e., the first base station) is receiving the fourth message in the current scenario; the first parameter set (para_set) is a list field, and the corresponding field value is the parameter used to configure each node in the second model (such as P1=xx, P2=xx, P3=xx, P4=xx in Table 5).

[0140] It should be noted that, in this embodiment of the invention, the first model deployed in the first network element for data analysis and the second model deployed in the first node that sends the first indicator data can be the same intelligent algorithm model. That is, the model parameter setting method can be the same structure agreed upon in advance, only differing in the specific model parameter values.

[0141] For example, both the first and second models can use an N-layer neural network (N can be configured by parameters). Their hidden layer weight functions are the same, but the specific parameter values ​​are different (for example, both are function models of y = ax + b, but the values ​​of a and b are different). That is, the first network element only needs to send the first parameter set to the first node after completing the training of the first model and obtaining the first parameter set, so that the first node can complete the configuration of the second model according to the first parameter, thereby realizing model collaboration.

[0142] In some implementations, after the first network element sends the first parameter set to the first node, it can receive a second response from the first node, the second response indicating that the first node has received the first parameter set.

[0143] In this embodiment, after the first network element sends a fourth message including a first set of parameters to the first node, it needs to collect the second response from the first node based on the fourth message to confirm that the node has received the fourth message and has begun configuring the second model according to the first set of parameters in the fourth message. For example, the second response received by the first network element can be an ACK from the first node.

[0144] As an example, the second response received by the first network element from the first node can be as shown in Table 6. The second response may include task type, sender name and receiver name. The task type may include task category and task subcategory. Each item in Table 6 corresponds to a field name, field description, field type and field value.

[0145] Table 6

[0146] Field Name Field Description Field type Field value task_type Task Categories string model_train subtask_type Task subclass string para_ack star_id Sender Name string gNB_1 obj_id Receiver name string OMC_1

[0147] The task category (task_type) is a string field, and the corresponding field value is model_train, meaning that the task category in the current scenario is training the first model used for data analysis to address network overload issues. The task subcategory (subtask_type) is a string field, and the corresponding field value is para_ack, meaning that the task subcategory in the current scenario is the second response (e.g., ACK) fed back by the first node. The sender name (star_id) is a string field, and the corresponding field value is gNB_1, meaning that the first node (i.e., the first base station) fed back the second response in the current scenario. The receiver name (obj_id) is a string field, and the corresponding field value is OMC_1, meaning that the first network element (i.e., OMC) received the second response in the current scenario.

[0148] In some implementations, the first network element exchanges messages with each node through a first interface.

[0149] In this embodiment, the first network element interacts with both the first node and the second node through a first interface. This first interface is an interface added in this embodiment for intelligent model collaboration. Specifically, the first network element can receive a first message sent by another network element through the first interface, then send a second message to the first node and a determined second node through the first interface, then receive a first response and a third message from the first node and the second node through the first interface, and finally send a first parameter set to the first node and receive a second response from the first node through the first interface.

[0150] It is understandable that although the second node does not report the first message, including the first indicator data, to the first network element when a problem occurs, the first network element can still interact with the second node through the first interface because the second node needs to participate in the data analysis task of the first indicator data. That is, the second node can interact with the first network element through the first interface.

[0151] For example, Figure 5 This is a schematic diagram of the architecture of the model collaboration method according to an embodiment of the present invention; as shown Figure 5 As shown, the architecture includes a first network element, a first node, and a second node. The first network element is connected to both the first and second nodes via a first interface. That is, the first interface is not limited to the connection between the first network element and the first node that reported the first indicator data; it can also be used for the connection between the first network element and each node (including the second node) that supports model collaboration. It should be noted that this embodiment of the invention does not impose specific limitations on the design of the first interface; it can be designed based on the Mapi (Microsoft Application Programming Interface) interface.

[0152] In some implementations, after receiving the second response from the first node, the first network element can receive the relevant operating status and observation results of indicator data reported by the first node during the operation of the second model after the first node has configured the second model based on the first parameter set.

[0153] This invention provides a model collaboration method. Figure 6 This is a flowchart illustrating the model collaboration method according to an embodiment of the present invention. Figure 2 ;like Figure 6 As shown, the method is applied to the first node, and the method includes:

[0154] Step 401: Send a first message to the first network element. The first message is used to request a data analysis task. The first message includes first indicator data.

[0155] Step 402: Collect second indicator data based on the instructions of the first network element, so that the first network element can obtain the second indicator data;

[0156] Step 403: Receive a first parameter set from the first network element, and configure a second model according to the first parameter set; wherein, the first parameter set is obtained by the first network element after training the first model based on at least the second index data, and at least some of the parameters in the first parameter set have values ​​different from the corresponding parameters after the first model is trained, and the second model has the same structure as the first model.

[0157] Here, the first node can be an access network device that supports model collaboration, such as a base station. The first network element can be a network element in the wireless network that can provide the functional requirements of services, such as a functional entity. Different functional entities are used for different scenarios in different fields. For example, in the field of mobile communications, functional entities such as BCFE are used to handle broadcast functions to ensure the effective utilization of wireless resources, while OMC is used for centralized management of wireless network operation and maintenance, providing functions such as fault detection, performance management, and configuration management.

[0158] In this embodiment, the first node sends a first message including first indicator data to the first network element. That is, the first node can actively collect and report first indicator data composed of multiple related indicators (or features) to request the first network element to initiate a data analysis task based on the first indicator data.

[0159] For example, in a load balancing application scenario, when the first node (i.e., the base station) detects a network overload problem during routine operation and maintenance, it needs to send a first message containing first indicator data to the first network element (i.e., the OMC) responsible for centralized network operation and maintenance. This allows the first network element to initiate a data analysis task for the network overload problem. The first indicator data includes multiple indicators (or features) related to the network overload problem, such as RSRP, SINR, PRB, etc.

[0160] In some implementations, the first message includes: task type, sender name, receiver name, tag corresponding to the indicator data, acquisition duration of the indicator data, and storage address of the indicator data.

[0161] In this embodiment, the first node sends the first indicator data to the first network element through the first message, so that the first network element can initiate a data analysis task. In addition to including the first indicator data, the first message may also include: task type, sender name, receiver name, tag corresponding to the indicator data, acquisition duration of the indicator data, and storage address of the indicator data.

[0162] The task type indicates the purpose of sending the first message, which may include a task category and a task subcategory; the sender name indicates the name of the network element sending the first message; the receiver name indicates the name of the network element receiving the first message; the tag corresponding to the indicator data indicates the tag corresponding to the feature of the first indicator data; the acquisition time of the indicator data is the time for the first node to collect the first indicator data; and the storage address of the indicator data is the address where the first node stores the first indicator data. It should be noted that the description of the first message in this embodiment has been described in detail above and will not be repeated here.

[0163] In this embodiment, the first node sends a first message including first indicator data to the first network element, and after the first network element starts the data analysis task, it collects second indicator data based on the instructions of the first network element for training the first model deployed in the first network element, so that the first network element can obtain the second indicator data.

[0164] In some implementations... Figure 7 This is a schematic diagram of the process for collecting the second indicator data in the model collaboration method of this invention; as shown. Figure 7 As shown, step 402 includes:

[0165] Step 501: The first node receives a second message from the first network element, the second message being used to instruct the collection of indicator data.

[0166] In this embodiment, the first node needs to receive a second message sent by the first network element, which is used to indicate the collection of indicator data. That is, after receiving the second message sent by the first network element, the first node begins to collect the second indicator data.

[0167] Step 502: Based on the second message, collect the second indicator data and send a third message to the first network element, wherein the third message includes the storage address of the indicator data.

[0168] In this embodiment, after receiving the second message sent by the first network element, the first node completes the collection of the second indicator data according to the second message. After completing the collection of the second indicator data on the network, the first node sends a third message including the storage address of the indicator data to the first network element, so that the first network element can obtain the second indicator data according to the storage address of the indicator data in the third message, thereby completing the training of the first model.

[0169] In some implementations, the second message includes: task type, sender name, receiver name, indicator data, tag corresponding to the indicator data, acquisition duration of the indicator data, and storage address of the indicator data;

[0170] The third message also includes: task type, sender name, receiver name, indicator data, the tag corresponding to the indicator data, and the acquisition time of the indicator data.

[0171] In this embodiment, the second message sent by the first network element to the first node for instructing the collection of second indicator data includes: task type, sender name, receiver name, indicator data, tag corresponding to the indicator data, acquisition duration of the indicator data, and storage address of the indicator data.

[0172] Among them, the task type indicates the purpose of sending the second message, which may include a task category and a task subcategory; the sender name indicates the name of the network element sending the second message; the receiver name indicates the name of the network element receiving the second message; the tag corresponding to the indicator data indicates the tag corresponding to the feature that the first network element instructs the first node or the second node to collect the second indicator data; the acquisition duration of the indicator data is the duration for the first node or the second node to collect the second indicator data; and the storage address of the indicator data is the address where the first node or the second node stores the second indicator data.

[0173] It should be noted that the description of the second message in the embodiments of the present invention has been described in detail above, and will not be repeated here.

[0174] In this embodiment, the third message sent by the first node to the first network element includes, in addition to the storage address of the indicator data, the following: task type, sender name, receiver name, indicator data, the tag corresponding to the indicator data, and the acquisition time of the indicator data.

[0175] Among them, the task type indicates the purpose of sending the third message, which may include task category and task sub-category; the sender name indicates the name of the network element sending the third message; the receiver name indicates the name of the network element receiving the third message; the label corresponding to the indicator data indicates the label corresponding to the feature of the second indicator data collected by the first node or the second node received by the first network element; the acquisition time of the indicator data is the time for the first node or the second node to collect the second indicator data; and the storage address of the indicator data is the address where the first node or the second node stores the second indicator data.

[0176] It should be noted that the description of the third message in the embodiments of the present invention has been described in detail above, and will not be repeated here.

[0177] In some optional embodiments, after receiving the second message sent by the first network element, the first node sends a first response back to the first network element. The first response is used to indicate that the first node has confirmed that it has received the second message.

[0178] In this embodiment, after receiving the second message instructing the collection of second indicator data, the first node can send a first response to the first network element based on the second message, informing the first network element that the node has received the second message and will begin collecting the second indicator data within the collection period specified in the second message. For example, the first response sent by the first node to the first network element can be an ACK (acknowledgment) from the first node.

[0179] It should be noted that the description of the first response in the embodiments of the present invention has been described in detail above, and will not be repeated here.

[0180] In this embodiment, after the first node sends a third message to the first network element, the first network element completes the training of the first model based on at least the second indicator data, receives the first parameter set obtained by the first network element from the training of the first model, and configures the second model with the same structure as the first model according to the first parameter set. The values ​​of at least some parameters in the first parameter set are different from the values ​​of the corresponding parameters after the first model is trained.

[0181] It should be noted that the embodiments of the present invention do not impose specific restrictions on the training process of the first model. Existing neural network methods can be used for model training and parameter adjustment to obtain the parameter set of the first model after the model training is completed. This will not be elaborated further here.

[0182] After the first network element completes model training and obtains the corresponding model parameters, a first parameter set is obtained through model distillation. At least some parameters in the first parameter set have values ​​different from the model parameters obtained after the first model is trained. For example, after model distillation, some parameters in the first parameter set may have been set to zero compared to the model parameters obtained after the first model is trained, meaning that some parameter values ​​are different.

[0183] It should also be noted that the model distillation processing method used in the embodiments of the present invention is an existing model compression and acceleration technology. Model distillation is mainly used to reduce the model size during deployment, improve inference speed, reduce computing resource consumption, and enhance the robustness and generalization ability of the model, thereby achieving efficient inference and prediction on nodes with limited resources (such as the first node). Further details will not be elaborated here.

[0184] In this embodiment, the first node can receive a fourth message from the first network element, which includes a first set of parameters. Based on the fourth message, the first node obtains the first set of parameters and configures a second model with the same structure as the first model according to the first set of parameters. This enables model collaboration, allowing the second model to solve different problems in different scenarios. For example, in the load balancing scenario described above, after configuring the second model based on the first set of parameters, when a network overload problem occurs, the first node uses the second model to identify the network element experiencing the overload problem and performs corresponding processing. The description of the fourth message in this embodiment has been detailed above and will not be repeated here.

[0185] It should be noted that the second model deployed in the first node that sends the first indicator data can be the same intelligent algorithm model as the first model deployed in the first network element for data analysis. That is, the model parameter settings can be the same structure agreed upon in advance, with only differences in the specific model parameter values.

[0186] For example, both the first and second models can use an N-layer neural network (N can be configured by parameters). Their hidden layer weight functions are the same, but the specific parameter values ​​are different (for example, both are function models of y = ax + b, but the values ​​of a and b are different). That is, the first node only needs to complete the configuration of the second model according to the first parameter set obtained by the first network element training the first model after receiving the first parameter set, thereby realizing model collaboration.

[0187] In some implementations, after the first network element sends the first parameter set to the first node, it can receive a second response from the first node, the second response indicating that the first node has received the first parameter set.

[0188] In this embodiment, after receiving the fourth message including the first parameter set sent by the first network element, the first node can send a second response to the first network element based on the fourth message to inform the first network element that the node has received the fourth message and begins configuring the second model according to the first parameter set in the fourth message. For example, the second response received by the first network element can be a response from the first node in the form of an ACK. The description of the second response in this embodiment has been detailed above and will not be repeated here.

[0189] In some implementations, the first node and the first network element exchange messages through a first interface.

[0190] In this embodiment, the first node and the first network element can interact via a first interface, which is an interface added in this embodiment for intelligent model collaboration. Specifically, the first node can send a first message to the first network element through the first interface, then receive a second message from the first network element through the first interface, then send a first response and a third message back to the first network element through the first interface, and finally receive a first set of parameters from the first network element through the first interface, and then send a second response back to the first network element through the first interface.

[0191] It should be noted that the embodiments of the present invention do not impose specific restrictions on the design of the first interface, and the first interface can be designed based on the Mapi interface.

[0192] In some implementations, after receiving the first parameter set sent by the first network element and responding with the second response, the first node will configure the second model based on the first parameter set and run the second model configured with the first parameter set, and report the observation results of relevant operating status and indicator data to the first network element in real time or periodically.

[0193] As an example, Figure 8 This is a schematic diagram of the interaction flow of the model collaboration method according to an embodiment of the present invention; as shown below. Figure 8 As shown, in a load balancing scenario, the specific process for achieving model collaboration between the first network element, the first node, and the second node through the first interface includes:

[0194] Step 601: Based on the detected network overload problem, the first node reports a first message to the first network element, including the first indicator data related to the problem;

[0195] Step 602: The first network element determines whether a data analysis task needs to be initiated based on the first indicator data. After confirming that a data analysis task needs to be initiated, it determines the second node to participate in the data analysis task based on the first indicator data sent by multiple candidate nodes.

[0196] Step 603: The first network element sends a second message to the first node and the second node respectively to indicate the collection of indicator data;

[0197] Step 604: After receiving the second message, the first node and the second node respectively send a first response to the first network element to inform the first network element that the node has received the second message and to start collecting the second indicator data;

[0198] Step 605: After completing the collection of the second indicator data, the first node and the second node respectively send a third message to the first network element, including the storage address of the indicator data;

[0199] Step 606: The first network element obtains the second indicator data according to the storage address of the indicator data in the third message, and trains the first model based on the second indicator data to obtain model parameters, and obtains the first parameter set through model distillation;

[0200] Step 607: The first network element sends the first parameter set to the first node;

[0201] Step 608: The first node configures the second model according to the first parameter set and runs the second model configured with the first parameter set;

[0202] Step 609: The first node reports the relevant operating status and indicator observation results of the second model to the first network element.

[0203] It is understandable that through the process of steps 601 to 609 above, it is possible to achieve customized reporting of network load anomalies, distributed collection of indicator data of each node, unified training of data analysis models, and fully utilize the computing power and globality of the first network element to achieve the effect of distributed deployment.

[0204] In this embodiment, a data analysis task is initiated by receiving first indicator data sent by a first node, and a second node participating in the data analysis task is determined by first indicator data sent by multiple candidate nodes. A first model in the first network element is trained based on the second indicator data collected by the first and second nodes, and a first parameter set is sent to the first node. This allows the first node to configure a second model with the same structure as the first model based on the first parameter set. As a result, the first node can process intelligent tasks based on the second model configured with the first parameter set. That is, through intelligent model collaboration between network elements, the advantages of network elements are complemented, the computing power is fully utilized, and the overall network gain is improved.

[0205] Based on the above embodiments, this invention also provides a model collaboration device, which is applied to a first network element. Figure 9 This is a schematic diagram of the composition structure of the model collaboration device according to an embodiment of the present invention. Figure 1;like Figure 9 As shown, the device includes: a first communication unit 71, a first processing unit 72, and a training unit 73; wherein,

[0206] The first communication unit 71 is used to receive first indicator data sent by multiple nodes; the multiple nodes include a first node and multiple candidate nodes, and the first indicator data of the first node is sent through a first message, which is used to request a data analysis task.

[0207] The first processing unit 72 is used to determine the second node participating in the data analysis task based on the first indicator data of the plurality of candidate nodes;

[0208] The first communication unit 71 is also configured to instruct the first node and the second node to collect the second indicator data, and to obtain the second indicator data from the first node and the second node, respectively.

[0209] The training unit 73 is used to train the first model based on the second index data of the first node and the second node to obtain a first parameter set, wherein at least some parameters in the first parameter set have values ​​different from the values ​​of the corresponding parameters after the first model has been trained.

[0210] The first communication unit 71 is further configured to send the first parameter set to the first node so that the first node can configure the second model according to the first parameter set, wherein the second model has the same structure as the first model.

[0211] In an optional embodiment of the present invention, the first message includes: task type, sender name, receiver name, tag corresponding to the indicator data, acquisition duration of the indicator data, and storage address of the indicator data.

[0212] In an optional embodiment of the present invention, the first processing unit 72 is further configured to determine whether to execute the data analysis task based on the indicator data of the first node before determining the second node participating in the data analysis task based on the first indicator data of the plurality of candidate nodes; and after determining to execute the data analysis task, determine the second node participating in the data analysis task based on the first indicator data of the plurality of nodes.

[0213] In an optional embodiment of the present invention, the first processing unit 72 is configured to construct a first matrix and a second matrix based on the first indicator data of the plurality of candidate nodes; the first matrix is ​​a matrix related to the candidate nodes, the number of the first matrices is the number of categories of the indicator data, and the second matrix is ​​a matrix related to the indicator data; the score of each candidate node is determined using the first matrix and the second matrix, and a second node is determined based on the score.

[0214] In an optional embodiment of the present invention, the first processing unit 72 is configured to perform row vector summation on the first matrix and the second matrix respectively, and to perform standardization on the column vectors obtained by the row vector summation to obtain a plurality of first weight vectors corresponding to the first matrix and second weight vectors corresponding to the second matrix respectively; and to multiply the second weight vectors by each first weight vector to obtain the score of each candidate node.

[0215] In an optional embodiment of the present invention, the first communication unit 71 is configured to send a second message to the first node and the second node respectively, the second message being used to instruct the collection of indicator data; and to receive a third message sent by the first node and the second node, the third message including the storage address of the indicator data;

[0216] The first processing unit 72 is further configured to obtain the second indicator data collected by the first node and the second node through the first communication unit 71 according to the storage address of the indicator data.

[0217] In an optional embodiment of the present invention, the second message includes: task type, sender name, receiver name, indicator data, tag corresponding to the indicator data, acquisition duration of the indicator data, and storage address of the indicator data;

[0218] The third message also includes: task type, sender name, receiver name, indicator data, the tag corresponding to the indicator data, and the acquisition time of the indicator data.

[0219] In an optional embodiment of the present invention, the first communication unit 71 exchanges messages with each node through a first interface.

[0220] In this invention, the first processing unit 72 and the training unit 73 in the device can both be implemented by a central processing unit (CPU), digital signal processor (DSP), microcontroller unit (MCU), or field-programmable gate array (FPGA) in the first network element in practical applications; the first communication unit 71 in the device can be implemented by a communication module (including: basic communication kit, operating system, communication module, standardized interface and protocol, etc.) and transceiver antenna in practical applications.

[0221] This invention also provides a model collaboration device, which is applied to a first node. Figure 10This is a schematic diagram of the composition structure of the model collaboration device according to an embodiment of the present invention. Figure 2 ;like Figure 10 As shown, the device includes: a second communication unit 81 and a second processing unit 82; wherein,

[0222] The second communication unit 81 is used to send a first message to the first network element. The first message is used to request a data analysis task and includes first indicator data.

[0223] The second processing unit 82 is used to collect second indicator data based on the instruction of the first network element, so that the first network element can obtain the second indicator data;

[0224] The second communication unit 81 is also configured to receive a first set of parameters from the first network element;

[0225] The second processing unit 82 is further configured to configure a second model according to the first parameter set; wherein the first parameter set is obtained by the first network element after training the first model based on the second index data, and the values ​​of at least some parameters in the first parameter set are different from the values ​​of the corresponding parameters after the first model is trained, and the second model has the same structure as the first model.

[0226] In an optional embodiment of the present invention, the first message includes: task type, sender name, receiver name, tag corresponding to the indicator data, acquisition duration of the indicator data, and storage address of the indicator data.

[0227] In an optional embodiment of the present invention, the second processing unit 82 is further configured to receive a second message from the first network element through the second communication unit 81, the second message being used to instruct the collection of indicator data; collect second indicator data based on the second message; and send a third message to the first network element through the second communication unit 81, the third message including the storage address of the indicator data.

[0228] In an optional embodiment of the present invention, the second message includes: task type, sender name, receiver name, indicator data, tag corresponding to the indicator data, acquisition duration of the indicator data, and storage address of the indicator data;

[0229] The third message also includes: task type, sender name, receiver name, indicator data, the tag corresponding to the indicator data, and the acquisition time of the indicator data.

[0230] In an optional embodiment of the present invention, the second communication unit 81 and the first network element exchange messages through a first interface.

[0231] In the implementation of this invention, the second processing unit 82 in the device can be implemented by the CPU, DSP, MCU or FPGA in the first node in practical applications; the second communication unit 81 in the device can be implemented by a communication module (including: basic communication kit, operating system, communication module, standardized interface and protocol, etc.) and transceiver antenna in practical applications.

[0232] It should be noted that the above-described model collaboration device is only illustrated by the division of the program modules described above. In practical applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program modules to complete all or part of the processing described above. In addition, the model collaboration device and the model collaboration method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0233] This invention also provides a communication device. Figure 11 This is a schematic diagram of the hardware composition structure of the communication device according to an embodiment of the present invention, such as... Figure 11 As shown, the communication device includes a memory 92, a processor 91, and a computer program stored in the memory 92 and executable on the processor 91.

[0234] Optionally, the communication device may specifically be a first network element or a first node in the embodiments of the present invention; when the processor 91 executes the program, it implements the steps of the model collaboration method applied to the first network element or the first node in the embodiments of the present invention.

[0235] Optionally, the communication device further includes at least one communication component 94. The various components in the communication device can be coupled together via a bus system 93. It is understood that the bus system 93 is used to implement communication between these components. In addition to a data bus, the bus system 93 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 11 The general labeled all buses as Bus System 93.

[0236] It is understood that memory 92 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memory 92 described in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0237] The methods disclosed in the above embodiments of the present invention can be applied to or implemented by processor 91. Processor 91 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 91 or by instructions in software form. The processor 91 may be a general-purpose processor, DSP, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 91 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of the present invention can be directly manifested as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in memory 92. Processor 91 reads the information in memory 92 and completes the steps of the aforementioned method in conjunction with its hardware.

[0238] In an exemplary embodiment, the communication device may be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), FPGAs, general-purpose processors, controllers, MCUs, microprocessors, or other electronic components to perform the aforementioned method.

[0239] This invention also provides a computer-readable storage medium having a computer program stored thereon.

[0240] Optionally, the computer-readable storage medium can be applied to the model collaboration device of the present invention embodiment; then when the program is executed by the processor, it implements the steps of the model collaboration method of the present invention applied to the first network element or the first node.

[0241] This invention also provides a computer program product, including a computer program that can be executed by a processor 91 of a communication device to complete the steps of the model collaboration method described in this invention.

[0242] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

[0243] The features disclosed in the several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.

[0244] The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method or device embodiments.

[0245] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0246] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0247] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0248] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0249] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0250] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A model collaboration method, characterized in that, The method is applied to a first network element, and the method includes: Receive first indicator data sent by multiple nodes; the multiple nodes include a first node and multiple candidate nodes, and the first indicator data of the first node is sent through a first message, which is used to request a data analysis task. The second node participating in the data analysis task is determined based on the first indicator data of the multiple candidate nodes; The first node and the second node are respectively instructed to collect the second indicator data, and to obtain the second indicator data from the first node and the second node; The first model is trained based on the second index data of the first node and the second node to obtain a first parameter set, wherein at least some parameters in the first parameter set have values ​​different from the corresponding parameters after the first model is trained. The first parameter set is sent to the first node so that the first node can configure the second model according to the first parameter set. The second model has the same structure as the first model.

2. The method according to claim 1, characterized in that, The first message includes: task type, sender name, receiver name, tag corresponding to the indicator data, acquisition time of the indicator data, and storage address of the indicator data.

3. The method according to claim 1, characterized in that, Before determining the second node participating in the data analysis task based on the first indicator data of the plurality of candidate nodes, the method further includes: Determine whether to execute the data analysis task based on the indicator data of the first node; Accordingly, determining the second node participating in the data analysis task based on the first indicator data of the plurality of candidate nodes includes: After determining to execute the data analysis task, a second node is determined to participate in the data analysis task based on the first indicator data of the multiple candidate nodes.

4. The method according to claim 1, characterized in that, The step of determining the second node to participate in the data analysis task based on the first indicator data of the plurality of candidate nodes includes: A first matrix and a second matrix are constructed based on the first indicator data of the plurality of candidate nodes; the first matrix is ​​a matrix related to the candidate nodes, and the number of the first matrices is the number of categories of indicator data; the second matrix is ​​a matrix related to the indicator data. The score of each candidate node is determined using the first matrix and the second matrix, and the second node is determined based on the score.

5. The method according to claim 4, characterized in that, The step of determining the score of each candidate node using the first matrix and the second matrix includes: The row vectors of the first matrix and the second matrix are summed respectively, and the column vectors obtained by the row vector summation are standardized to obtain multiple first weight vectors corresponding to the first matrix and second weight vectors corresponding to the second matrix. The second weight vector is multiplied by each of the first weight vectors to obtain the score of each candidate node.

6. The method according to claim 1, characterized in that, The step of respectively instructing the first node and the second node to collect the second indicator data, and obtaining the second indicator data from the first node and the second node, includes: Send a second message to the first node and the second node respectively. The second message is used to instruct the collection of indicator data. Receive a third message sent by the first node and the second node, wherein the third message includes the storage address of the indicator data; The second indicator data collected by the first node and the second node is obtained according to the storage address of the indicator data.

7. The method according to claim 6, characterized in that, The second message includes: task type, sender name, receiver name, indicator data, tag corresponding to the indicator data, acquisition time of the indicator data, and storage address of the indicator data; The third message also includes: task type, sender name, receiver name, indicator data, the tag corresponding to the indicator data, and the acquisition time of the indicator data.

8. The method according to any one of claims 1 to 7, characterized in that, The first network element exchanges messages with each node through the first interface.

9. A model collaboration method, characterized in that, The method is applied to the first node, and the method includes: Send a first message to the first network element. The first message is used to request a data analysis task. The first message includes first indicator data. The second indicator data is collected based on the instructions of the first network element, so that the first network element can obtain the second indicator data; The system receives a first parameter set from the first network element and configures a second model according to the first parameter set. The first parameter set is obtained by the first network element after training the first model based on at least the second indicator data. The values ​​of at least some parameters in the first parameter set are different from the values ​​of the corresponding parameters after the first model is trained. The second model has the same structure as the first model.

10. The method according to claim 9, characterized in that, The first message includes: task type, sender name, receiver name, tag corresponding to the indicator data, acquisition time of the indicator data, and storage address of the indicator data.

11. The method according to claim 9, characterized in that, The step of collecting second indicator data based on the instruction of the first network element, so that the first network element can obtain the second indicator data, includes: The first node receives a second message from the first network element, the second message being used to instruct the collection of indicator data; Based on the second message, second indicator data is collected, and a third message is sent to the first network element, the third message including the storage address of the indicator data.

12. The method according to claim 11, characterized in that, The second message includes: task type, sender name, receiver name, indicator data, tag corresponding to the indicator data, acquisition time of the indicator data, and storage address of the indicator data; The third message also includes: task type, sender name, receiver name, indicator data, the tag corresponding to the indicator data, and the acquisition time of the indicator data.

13. The method according to any one of claims 9 to 12, characterized in that, The first node and the first network element exchange messages through the first interface.

14. A model collaboration device, characterized in that, The device is applied to a first network element, and the device includes: a first communication unit, a first processing unit, and a training unit; wherein... The first communication unit is used to receive first indicator data sent by multiple nodes; the multiple nodes include a first node and multiple candidate nodes, and the first indicator data of the first node is sent through a first message, which is used to request a data analysis task. The first processing unit is configured to determine a second node participating in the data analysis task based on the first indicator data of the plurality of candidate nodes; The first communication unit is further configured to instruct the first node and the second node to collect the second indicator data, and to obtain the second indicator data from the first node and the second node, respectively; The training unit is used to train the first model based on the second index data of the first node and the second node to obtain a first parameter set, wherein at least some parameters in the first parameter set have values ​​different from the values ​​of the corresponding parameters after the first model has been trained. The first communication unit is further configured to send the first parameter set to the first node so that the first node can configure the second model according to the first parameter set, wherein the second model has the same structure as the first model.

15. A model collaboration device, characterized in that, The device is applied to a first node, and the device includes: a second communication unit and a second processing unit; wherein... The second communication unit is used to send a first message to the first network element. The first message is used to request a data analysis task and includes first indicator data. The second processing unit is used to collect second indicator data based on the instructions of the first network element, so that the first network element can obtain the second indicator data; The second communication unit is further configured to receive a first set of parameters from the first network element; The second processing unit is further configured to configure a second model according to the first parameter set; wherein the first parameter set is obtained by the first network element after training the first model based on the second index data, and the values ​​of at least some parameters in the first parameter set are different from the values ​​of the corresponding parameters after the first model is trained, and the second model has the same structure as the first model.

16. A communication device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 8, or; When the processor executes the program, it implements the steps of the method according to any one of claims 9 to 13.

17. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program performs the steps of the method according to any one of claims 1 to 8, or; When executed by a processor, the program implements the steps of the method according to any one of claims 9 to 13.

18. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the method according to any one of claims 1 to 8, or; When the computer program is executed by a processor, it implements the method according to any one of claims 9 to 13.

Images