Network data analysis model scheduling method and apparatus, and computer device

Abstract

The present application relates to a network data analysis model scheduling method and apparatus, a computer device, a storage medium, and a computer program product. The method comprises: in response to a task request initiated by a service requester for network data analysis tasks, determining the request type of the task request (S210), wherein the task request is used for requesting a network data analysis service network element to allocate models to process the network data analysis tasks; performing matching from a pre-updated prediction model matching library to obtain a target prediction model corresponding to the request type (S220), wherein the target prediction model is used for outputting an operation parameter prediction result for the network data analysis tasks, and the operation parameter prediction result represents resource occupation information required for completing the network data analysis tasks; and scheduling the network data analysis tasks according to the operation parameter prediction result (S230).

Description

Network data analysis model scheduling method, device and computer equipment

[0001] Related applications

[0002] This application claims priority to Chinese patent application number 2023108713171, filed on July 17, 2023, entitled “Network Data Analysis Model Scheduling Method, Apparatus and Computer Equipment,” the entire text of which is hereby incorporated by reference. Technical Field

[0003] The present application relates to the fields of network technology and security technology, and in particular to a network data analysis model scheduling method, apparatus, computer equipment, storage medium, and computer program product. Background Art

[0004] In the 5G era, due to the diversification of network services and the richness of application scenarios, the network scale is growing and the network environment is becoming more complex. To enhance operators' current analysis and processing capabilities, 3GPP plans to introduce artificial intelligence technology into 5G networks and add a new NWDAF (Network Data Analytics Function) network element. With the help of AI (Artificial Intelligence) technology, it automatically perceives and analyzes the network and participates in the entire life cycle of network planning, construction, operation and maintenance, network optimization, and operation to better support network on-demand services, improve network resource utilization, and enhance customer experience and satisfaction.

[0005] However, the training and inference tasks of massive heterogeneous models pose huge challenges to NWDAF network elements in terms of efficiency and resources. The current standards for NWDAF do not involve sections related to model scheduling. Traditional model scheduling methods include: (1) executing tasks in the order they are received; (2) completing scheduling by manually configuring the empirical values ​​of model operation parameters. The above model scheduling methods generally have the problem of low resource utilization.

[0006] Summary of the Invention

[0007] In a first aspect, the present application provides a network data analysis model scheduling method. The method comprises:

[0008] In response to a task request initiated by a service requester for a network data analysis task, determining a request type of the task request; the task request is used to request a network data analysis service network element allocation model to process the network data analysis task;

[0009] A target prediction model corresponding to the request type is obtained by matching from a pre-updated prediction model matching library; the target prediction model is used to output an operating parameter prediction result for the network data analysis task; the operating parameter prediction result represents resource occupancy information required to complete the network data analysis task;

[0010] The network data analysis tasks are scheduled according to the operation parameter prediction results.

[0011] In one embodiment, the method further comprises:

[0012] Obtaining actual measurement results of operating parameters of the network data analysis task; the actual measurement results of operating parameters represent the resource occupancy information of the network data analysis task during execution detected in real time;

[0013] When there is a preset deviation between the actual measured results of the operating parameters and the predicted results of the operating parameters, spare resources are used to expand the resource capacity of the target node; the target node includes a working node in the network data analysis service network element where the network data analysis task is located.

[0014] In one embodiment, the method further comprises:

[0015] When there is no preset deviation between the actual measurement result of the operating parameter and the predicted result of the operating parameter, and the network data analysis task is executed normally, the backup resource is allocated to each of the working nodes.

[0016] In one embodiment, scheduling each of the network data analysis tasks according to the operating parameter prediction results includes:

[0017] Obtaining a target weight value corresponding to each operating parameter in the operating parameter prediction result;

[0018] The network data analysis tasks are scheduled according to the priority order reflected in the target weight values.

[0019] In one embodiment, obtaining the target weight value corresponding to each operating parameter in the operating parameter prediction result includes:

[0020] According to the business requirements of the network data analysis task, set initial weight values ​​for each operating parameter;

[0021] According to the occupancy ratio of each resource in the network data analysis service network element, the initial weight value is adjusted to obtain the target weight value.

[0022] In one embodiment, the method further comprises:

[0023] Obtain at least two prediction models that meet the request type corresponding to the network data analysis task;

[0024] Training the prediction model based on the characteristic data of the network data analysis task;

[0025] The prediction model matching library is updated according to the prediction model with the highest prediction accuracy among the trained prediction models.

[0026] In a second aspect, the present application also provides a network data analysis model scheduling device. The device includes:

[0027] A request type determination module, configured to determine a request type of a task request initiated by a service requester for a network data analysis task; the task request is used to request a network data analysis service network element allocation model to process the network data analysis task;

[0028] A prediction model matching module is used to match a target prediction model corresponding to the request type from a pre-updated prediction model matching library; the target prediction model is used to output an operating parameter prediction result for the network data analysis task; the operating parameter prediction result represents resource occupancy information required to complete the network data analysis task;

[0029] The task scheduling module is used to schedule each of the network data analysis tasks according to the prediction results of the operating parameters.

[0030] In some embodiments, the device is also used to: obtain the actual measured results of the operating parameters of the network data analysis task; the actual measured results of the operating parameters represent the resource occupancy information of the network data analysis task during execution detected in real time; and when there is a preset deviation between the actual measured results of the operating parameters and the predicted results of the operating parameters, use spare resources to expand the resource capacity of the target node; the target node includes the working node in the network data analysis service network element where the network data analysis task is located.

[0031] In some embodiments, the device is further used to: allocate the backup resources to each of the working nodes when there is no preset deviation between the actual measurement results of the operating parameters and the predicted results of the operating parameters, and when the network data analysis task is executed normally.

[0032] In some embodiments, the task scheduling module is also used to: obtain the target weight value corresponding to each operating parameter in the operating parameter prediction result; and schedule each of the network data analysis tasks according to the priority satisfaction order reflected in the target weight value.

[0033] In some embodiments, the task scheduling module is also used to: set an initial weight value for each operating parameter according to the business requirements of the network data analysis task; and adjust the initial weight value according to the occupancy ratio of each resource in the network data analysis service network element to obtain the target weight value.

[0034] In some embodiments, the device is also used to: obtain at least two prediction models that meet the request type corresponding to the network data analysis task; train the prediction models based on the characteristic data of the network data analysis task; and update the prediction model matching library based on the prediction model with the highest prediction accuracy among the trained prediction models.

[0035] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are performed:

[0036] In response to a task request initiated by a service requester for a network data analysis task, determining a request type of the task request; the task request is used to request a network data analysis service network element allocation model to process the network data analysis task;

[0037] A target prediction model corresponding to the request type is obtained by matching from a pre-updated prediction model matching library; the target prediction model is used to output an operating parameter prediction result for the network data analysis task; the operating parameter prediction result represents resource occupancy information required to complete the network data analysis task;

[0038] The network data analysis tasks are scheduled according to the operation parameter prediction results.

[0039] In some embodiments, when the processor executes the computer program, it also implements the following steps: obtaining the actual measured results of the operating parameters of the network data analysis task; the actual measured results of the operating parameters represent the resource occupancy information of the network data analysis task during execution detected in real time; and when there is a preset deviation between the actual measured results of the operating parameters and the predicted results of the operating parameters, using spare resources to expand the resource capacity of the target node; the target node includes the working node in the network data analysis service network element where the network data analysis task is located.

[0040] In some embodiments, when the processor executes the computer program, it also implements the following steps: when there is no preset deviation between the actual measured results of the operating parameters and the predicted results of the operating parameters, and when the network data analysis task is executed normally, the backup resources are allocated to each of the working nodes.

[0041] In some embodiments, scheduling each of the network data analysis tasks according to the operating parameter prediction results includes the following steps: obtaining the target weight value corresponding to each operating parameter in the operating parameter prediction results; and scheduling each of the network data analysis tasks according to the priority satisfaction order reflected in the target weight value.

[0042] In some embodiments, obtaining the target weight value corresponding to each operating parameter in the operating parameter prediction result includes the following steps: setting an initial weight value for each operating parameter according to the business requirements of the network data analysis task; and adjusting the initial weight value according to the occupancy ratio of each resource in the network data analysis service network element to obtain the target weight value.

[0043] In some embodiments, when the processor executes the computer program, it also implements the following steps: obtaining at least two prediction models that meet the request type corresponding to the network data analysis task; training the prediction models based on the characteristic data of the network data analysis task; and updating the prediction model matching library based on the prediction model with the highest prediction accuracy among the trained prediction models.

[0044] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the following steps:

[0045] In response to a task request initiated by a service requester for a network data analysis task, determining a request type of the task request; the task request is used to request a network data analysis service network element allocation model to process the network data analysis task;

[0046] A target prediction model corresponding to the request type is obtained by matching from a pre-updated prediction model matching library; the target prediction model is used to output an operating parameter prediction result for the network data analysis task; the operating parameter prediction result represents resource occupancy information required to complete the network data analysis task;

[0047] The network data analysis tasks are scheduled according to the operation parameter prediction results.

[0048] In some embodiments, when the computer program is executed by the processor, the following steps are also implemented: obtaining the actual measured results of the operating parameters of the network data analysis task; the actual measured results of the operating parameters represent the resource occupancy information of the network data analysis task during execution detected in real time; and when there is a preset deviation between the actual measured results of the operating parameters and the predicted results of the operating parameters, using spare resources to expand the resource capacity of the target node; the target node includes the working node in the network data analysis service network element where the network data analysis task is located.

[0049] In some embodiments, when the computer program is executed by the processor, the following steps are also implemented: when there is no preset deviation between the actual measured results of the operating parameters and the predicted results of the operating parameters, and when the network data analysis task is executed normally, the backup resources are allocated to each of the working nodes.

[0050] In some embodiments, scheduling each of the network data analysis tasks according to the operating parameter prediction results includes the following steps: obtaining the target weight value corresponding to each operating parameter in the operating parameter prediction results; and scheduling each of the network data analysis tasks according to the priority satisfaction order reflected in the target weight value.

[0051] In some embodiments, obtaining the target weight value corresponding to each operating parameter in the operating parameter prediction result includes the following steps: setting an initial weight value for each operating parameter according to the business requirements of the network data analysis task; and adjusting the initial weight value according to the occupancy ratio of each resource in the network data analysis service network element to obtain the target weight value.

[0052] In some embodiments, when the computer program is executed by the processor, the following steps are also implemented: obtaining at least two prediction models that meet the request type corresponding to the network data analysis task; training the prediction models based on the characteristic data of the network data analysis task; and updating the prediction model matching library based on the prediction model with the highest prediction accuracy among the trained prediction models.

[0053] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the following steps:

[0054] In response to a task request initiated by a service requester for a network data analysis task, determining a request type of the task request; the task request is used to request a network data analysis service network element allocation model to process the network data analysis task;

[0055] A target prediction model corresponding to the request type is obtained by matching from a pre-updated prediction model matching library; the target prediction model is used to output an operating parameter prediction result for the network data analysis task; the operating parameter prediction result represents resource occupancy information required to complete the network data analysis task;

[0056] The network data analysis tasks are scheduled according to the operation parameter prediction results.

[0057] In some embodiments, when the computer program is executed by the processor, the following steps are also implemented: obtaining the actual measured results of the operating parameters of the network data analysis task; the actual measured results of the operating parameters represent the resource occupancy information of the network data analysis task during execution detected in real time; and when there is a preset deviation between the actual measured results of the operating parameters and the predicted results of the operating parameters, using spare resources to expand the resource capacity of the target node; the target node includes the working node in the network data analysis service network element where the network data analysis task is located.

[0058] In some embodiments, when the computer program is executed by the processor, the following steps are also implemented: when there is no preset deviation between the actual measured results of the operating parameters and the predicted results of the operating parameters, and when the network data analysis task is executed normally, the backup resources are allocated to each of the working nodes.

[0059] In some embodiments, scheduling each of the network data analysis tasks according to the operating parameter prediction results includes the following steps: obtaining the target weight value corresponding to each operating parameter in the operating parameter prediction results; and scheduling each of the network data analysis tasks according to the priority satisfaction order reflected in the target weight value.

[0060] In some embodiments, obtaining the target weight value corresponding to each operating parameter in the operating parameter prediction result includes the following steps: setting an initial weight value for each operating parameter according to the business requirements of the network data analysis task; and adjusting the initial weight value according to the occupancy ratio of each resource in the network data analysis service network element to obtain the target weight value.

[0061] In some embodiments, when the computer program is executed by the processor, the following steps are also implemented: obtaining at least two prediction models that meet the request type corresponding to the network data analysis task; training the prediction models based on the characteristic data of the network data analysis task; and updating the prediction model matching library based on the prediction model with the highest prediction accuracy among the trained prediction models.

[0062] The details of one or more embodiments of the present application are set forth in the accompanying drawings and the description below. Other features, objects, and advantages of the present application will become apparent from the description, drawings, and claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. The drawings described below are merely some embodiments of the present application and do not constitute a limitation on the disclosure content and protection scope of the present application.

[0064] FIG1 is a diagram illustrating an application environment of a network data analysis model scheduling method according to an embodiment;

[0065] FIG2 is a flow chart of a method for scheduling a network data analysis model according to an embodiment;

[0066] FIG3 is a flow chart of a method for scheduling a network data analysis model according to another embodiment;

[0067] FIG4 is a timing diagram of a network data analysis model scheduling method according to one embodiment;

[0068] FIG5 is a timing diagram of a network data analysis model scheduling method according to another embodiment;

[0069] FIG6 is a schematic diagram showing the principle of a network data analysis model scheduling method according to an embodiment;

[0070] FIG7 is a structural block diagram of a network data analysis model scheduling device according to one embodiment;

[0071] FIG8 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0072] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0073] The network data analysis model scheduling method provided in the embodiment of the present application can be applied in the application environment shown in Figure 1. In this case, the service requester 110 communicates with the network data analysis service network element 120 through the network.

[0074] The service requester 110 can be a terminal or a server. Terminals include, but are not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart car devices. Portable wearable devices can include smart watches, smart bracelets, and head-mounted devices. The server can be implemented as a standalone server or a server cluster consisting of multiple servers.

[0075] Among them, the network data analysis service network element is a newly added NWDAF network element in the core network, which is used to receive and process requests from service requesters, complete corresponding data analysis tasks, and return data analysis results to the service requesters.

[0076] In one embodiment, as shown in FIG2 , a network data analysis model scheduling method is provided. The method is described by applying the method to the network data analysis service network element 120 in FIG1 as an example, including the following steps:

[0077] S210, in response to a task request initiated by a service requester for a network data analysis task, determining a request type of the task request;

[0078] The request type includes the requested model type and task type. The model type refers to the type of model used to process network data analysis tasks, such as statistical analysis, logical calculation, tree-structured, clustering, and deep network. The task type refers to the type of network data analysis task, such as inference and training.

[0079] The task request is used to request the network data analysis service network element allocation model to process the network data analysis task.

[0080] S220, matching and obtaining a target prediction model corresponding to the request type from a pre-updated prediction model matching library;

[0081] The prediction model matching library stores the mapping relationship between the request type and the prediction model, and the prediction model matching library needs to be pre-updated before matching.

[0082] Among them, the target prediction model is used to output the operation parameter prediction results for the network data analysis task; the operation parameter prediction results represent the resource occupancy information required to complete the network data analysis task.

[0083] Specifically, the operating parameters include but are not limited to completion time, storage capacity, computing power, network resources, memory, number of servers, etc. The operating parameter prediction result can be "the estimated completion time of the task corresponding to this request type is predicted to be 1 hour, it requires 10GB of storage space, and the minimum computing power requirement is 100MIPS."

[0084] For example, when the model type is statistical analysis and the task type is reasoning, the LSTM prediction model with the most recent update time and the highest prediction accuracy is matched.

[0085] S230: Schedule each network data analysis task according to the prediction result of the operating parameters.

[0086] For example, before using the scheduling algorithm to perform task scheduling, it is necessary to set constraints for it. The constraints include the predicted values ​​of each operating parameter and the priority of each operating parameter. Therefore, after generating the predicted values ​​of the operating parameters of each task, it is necessary to configure the parameter weights of each task. Then the scheduling center dynamically adjusts the parameter weights based on the real-time network element conditions and network environment, and then performs scheduling among multiple tasks and allocation of resources to multiple nodes, so that the loads on each working node in terms of time, storage, computing power, etc. are as close as possible.

[0087] In the above-mentioned network data analysis model scheduling method, after receiving the task request initiated by the service requester, the network data analysis service network element will determine the request type of the task request, and then match the target prediction model corresponding to the request type from the pre-updated prediction model matching library to output the operating parameter prediction results for the network data analysis task to reflect the resource occupancy information required to complete the network data analysis task. Finally, each network data analysis task is scheduled according to the operating parameter prediction results. The operating parameter prediction results based on heterogeneous models provide a more accurate basis for the resource scheduling and reasonable task scheduling of the network data analysis service network element, ensuring the request requirements for latency and improving the utilization efficiency of multiple types of resources such as network, computing power, and storage.

[0088] In one embodiment, the method further includes: obtaining measured results of operating parameters of the network data analysis task; and using backup resources to expand the resource capacity of the target node when there is a preset deviation between the measured results of the operating parameters and the predicted results of the operating parameters.

[0089] The measured results of operating parameters represent the resource usage information of the network data analysis task during its execution, which is detected in real time. It should be noted that the measured results of operating parameters are used in the following aspects: (1) as a real-time monitoring of various aspects of system resources; (2) as accumulated historical training data to train the prediction model and improve the accuracy of the prediction; (3) based on the prediction results, the deviation from the actual data is determined to identify whether the actual situation deviates from the original preset schedule, and then the resources are adjusted according to the deviation.

[0090] Among them, the target node includes the working node in the network data analysis service network element where the network data analysis task is located. Specifically, if there is a deviation in the prediction result of the operating parameters corresponding to a task, the working node where the task is located when it is executed is the target node.

[0091] For example, when the core network adopts Kubernetes for resource management, the working node is the node used to process specific tasks.

[0092] In this embodiment, the actual measured results of the operating parameters of the network data analysis task are first obtained, and then when there is a preset deviation between the actual measured results of the operating parameters and the predicted results of the operating parameters, spare resources are used to expand the resource capacity of the target node so that the corresponding task can be completed within the expected time, thereby improving the response speed of the network data analysis task.

[0093] In one embodiment, the method further includes: allocating backup resources to each working node when there is no preset deviation between the actual measurement results of the operating parameters and the predicted results of the operating parameters and the network data analysis task is executed normally.

[0094] The absence of preset deviations means that the predicted situation is consistent with the measured situation, and the normal execution of the network data analysis task means that the task has not been interrupted or cancelled.

[0095] In this embodiment, when the actual measured results of the operating parameters are consistent with the predicted results of the operating parameters, and the network data analysis task is executed normally, the backup resources are released and evenly distributed to each node to accelerate the operation while ensuring that each request model operates normally, thereby improving the utilization of resources and ensuring the normal operation of the network element.

[0096] In the above embodiment, by introducing a guarantee mechanism, existing network failures such as untimely request responses caused by prediction deviations and changes in the network environment are avoided. Taking into account the impact of the actual business of the network element, backup resources are used to ensure that tasks are completed on time and the real-time requirements of the existing network are guaranteed. This has good application prospects and economic benefits in the actual operation of network elements in the future; in addition, the network data analysis model scheduling method provided in this application not only fully and reasonably utilizes multiple types of resources, but also effectively guarantees the delay of request services, and is expected to further serve as a content supplement for the NWDAF part of 3GPP.

[0097] In one embodiment, each network data analysis task is scheduled according to the operating parameter prediction results, including: obtaining the target weight value corresponding to each operating parameter in the operating parameter prediction results; and scheduling each network data analysis task according to the priority satisfaction order reflected in the target weight value.

[0098] The target weight value is the weight of the operating parameter adjusted according to the real-time environment.

[0099] For example, if the weights of the operating parameters and their size relationship are: completion time (weight: 10) > storage (weight: 5) > computing power (weight: 4) > network resources (weight: 3), it means that when performing task scheduling, the conditions are judged in the order of completion time, storage capacity, computing power, and network resources, and task scheduling is carried out with the goal of meeting the conditions.

[0100] In this embodiment, by obtaining the target weight value corresponding to each operating parameter in the operating parameter prediction result, and scheduling each network data analysis task according to the priority satisfaction order reflected in the target weight value, the satisfaction order of the operating parameters during scheduling is clarified, the dependence of the scheduling results on important indicators is improved, and the quality of scheduling is improved.

[0101] In one embodiment, the target weight value corresponding to each operating parameter in the operating parameter prediction result is obtained, including: setting an initial weight value for each operating parameter according to the business requirements of the network data analysis task; adjusting the initial weight value according to the occupancy ratio of each resource in the network data analysis service network element to obtain the target weight value.

[0102] The initial weight value is the initial value of the weight pre-configured for each parameter according to business requirements before scheduling.

[0103] For example, in the business requirements of network data analysis tasks, the completion time of the model is the primary concern, so the initial weight value of the completion time can be set to the maximum value; in the actual scheduling process, actual adjustments can be made based on factors such as storage and computing power. For example, if the current network element storage is relatively abundant and computing power resources are relatively scarce, it can be adjusted to computing power > storage.

[0104] In this embodiment, first, an initial weight value is set for each operating parameter based on the business requirements of the network data analysis task, and then the initial weight value is adjusted according to the occupancy ratio of each resource in the network data analysis service network element to obtain the target weight value. At the same time, the impact of the business requirements before scheduling and the actual measured environment during scheduling on the weight are taken into account, thereby improving the accuracy of the target weight value.

[0105] In one embodiment, the method also includes: obtaining at least two prediction models that meet the request type corresponding to the network data analysis task; training the prediction model based on the characteristic data of the network data analysis task; and updating the prediction model matching library based on the prediction model with the highest prediction accuracy among the trained prediction models.

[0106] Among them, the characteristic data of the network data analysis task is the data used to train the prediction model, such as the real-time network environment, data volume, model parameter quantity and other characteristics.

[0107] In this embodiment, according to the model type required for the requested service, the corresponding operating parameter prediction model is constructed by classifying the characteristic data, and the best one is selected from them, effectively avoiding the isolated modeling and repeated calculations of one request and one model, which cause low computing efficiency, and the evaluation results are more accurate and reliable.

[0108] In addition, different prediction models are used to classify and estimate the operating parameters of massive heterogeneous models. On the one hand, this can more accurately evaluate the operating requirements of multiple types of heterogeneous models, so that request response and resource utilization can be most efficient; on the other hand, classification processing can save the duplication of computing and transmission resources and improve computing efficiency, thereby realizing real-time and accurate processing of large-scale data requests.

[0109] In another embodiment, as shown in FIG3 , a network data analysis model scheduling method is provided. The method is described by taking the network data analysis service network element 120 in FIG1 as an example, including the following steps:

[0110] S301, obtaining at least two prediction models that meet the request type corresponding to the network data analysis task;

[0111] S302, training a prediction model based on the characteristic data of the network data analysis task;

[0112] S303, updating the prediction model matching library according to the prediction model with the highest prediction accuracy among the trained prediction models;

[0113] S304, matching and obtaining a target prediction model corresponding to the request type from a pre-updated prediction model matching library;

[0114] S305, setting initial weight values ​​for each operating parameter according to the business requirements of the network data analysis task;

[0115] S306, adjusting the initial weight value based on the occupancy ratio of each resource in the service network element according to the network data analysis to obtain a target weight value;

[0116] S307, scheduling each network data analysis task according to the priority order reflected in the target weight value;

[0117] S308, obtaining the measured results of the operating parameters of the network data analysis task;

[0118] S309, when there is a preset deviation between the actual measured operating parameter result and the predicted operating parameter result, using backup resources to expand the resource capacity of the target node;

[0119] S310 , when there is no preset deviation between the actual measurement result of the operating parameters and the predicted result of the operating parameters, and the network data analysis task is executed normally, allocating backup resources to each working node.

[0120] It should be noted that the specific limitations of the above steps can be found in the specific limitations of a network data analysis model scheduling method mentioned above, which will not be repeated here.

[0121] In one embodiment, the model scheduling method executes tasks in the order in which they are received. In the scenario of multi-model reasoning and training tasks, executing tasks in sequence often results in the model having to finish running before resources can be released, which easily causes computing resources to be occupied by large-volume models for a long time, while small-volume models with high priority need to wait for a long time, making it impossible to complete the reasoning task in time; in another embodiment, the model scheduling method completes scheduling by manually configuring the empirical values ​​of the model running parameters. By manually configuring the model running parameters, the configured parameters cannot be modified in real time, and the flexibility is poor, and it cannot be applied to the real-time changing network environment.

[0122] In addition, the model scheduling method in the above two embodiments fails to consider the differences in factors such as the structure and complexity of the task model, which often leads to uneven resource utilization and distribution of network elements. In addition, it does not take into account the changes in the actual network environment and is unable to respond to and take safeguard measures for sudden situations, resulting in the inability to efficiently schedule tasks, thereby making resource utilization efficiency low.

[0123] Based on this, this application proposes a network data analysis model scheduling method. In response to the pressure on resources, bandwidth and latency caused by the training and inference of massive models due to large-scale analysis requests faced by NWDAF network elements, the importance of parameters is adjusted in combination with the network environment, network element load conditions and business conditions. In response to the differences in the estimated requirements of heterogeneous models for network, computing power, memory and other resources, the application scientifically, reasonably, accurately and fully allocates and schedules multiple types of resources within the NWDAF network element, combines and orchestrates multiple heterogeneous model inference and training tasks, and performs multi-dimensional trade-offs in parallel scheduling to ensure that the task loads on different nodes are as close as possible and that parallel tasks can be completed at the same time, effectively improving resource utilization and meeting real-time responses to requests. Through the heterogeneous model scheduling module and deviation guarantee expectation module within NWDAF, efficient responses to network analysis requests and full utilization of multiple types of resources such as network, computing power, and storage are achieved.

[0124] To facilitate understanding by those skilled in the art, FIG4 exemplifies a timing diagram of a network data analysis model scheduling method; as shown in FIG4 , when an NWDAF user sends an inference request for communication traffic prediction to the NWDAF scheduling center, the NWDAF scheduling center identifies the model type of the analysis request and obtains that the model of the request is xgboost (eXtreme Gradient Boosting), which is an inference task of a tree model. The NWDAF unit network element then calls the operating parameter prediction model of the tree model to predict the inference task, and sends the parameter prediction value of the requested task and the parameter importance ranking to the NWDAF scheduling center. The NWDAF scheduling center adjusts the parameter importance based on the real-time network element status and network environment, and performs scheduling among multiple tasks and allocation of multiple node resources based on the task's operating parameter prediction value.

[0125] During scheduling, a detection guarantee request is sent to the NWDAF guarantee center to start task detection. When a deviation between the actual operation and the expected situation is detected, an exception notification is sent to the scheduling center, which then activates backup resources to the corresponding nodes for resource expansion and dynamic adjustment to ensure the timely completion of the task. Finally, the analysis results are returned to the NWDAF user.

[0126] To facilitate understanding by those skilled in the art, FIG5 exemplifies a timing diagram of another network data analysis model scheduling method; as shown in FIG5 , when an NWDAF user sends a training request for network element load analysis to the NWDAF scheduling center, the NWDAF scheduling center identifies the model type of the analysis request and obtains that the model of the request is LSTM (Long Short-Term Memory), which is a deep learning training task. The NWDAF unit network element then calls the operating parameter prediction model of the deep learning model to predict the training task, and sends the parameter prediction value of the requested task and the parameter importance ranking to the NWDAF scheduling center. The NWDAF scheduling center adjusts the parameter importance based on the real-time network element status and network environment, and performs scheduling between multiple tasks and allocation of multiple node resources based on the task's operating parameter prediction value.

[0127] During scheduling, a detection guarantee request is sent to the NWDAF guarantee center to start task detection. When the detection and actual operation are in line with expectations, a normal notification is sent to the scheduling center. The NWDAF scheduling center will evenly distribute the spare resources to each working node to accelerate the completion of the task and finally return the analysis results to the NWDAF user.

[0128] The network data analysis model scheduling method is described in detail below using a specific embodiment with reference to Figure 6. It is worth noting that the following description is merely an example and does not limit the application.

[0129] As shown in FIG6 , the network data analysis model scheduling method provided by this application includes steps 1 to 6, specifically:

[0130] □Step 1: The NWDAF network element receives the analysis service request sent by the consumer;

[0131] Step 2: The NWDAF network element identifies whether the service request requires model support; if so, proceed to step 3; otherwise, proceed to step 6;

[0132] Step 3: Enter the scheduling and allocation of heterogeneous models within the NWDAF network element;

[0133] (1) Enter the model pool and identify and classify the model according to its structure and request type;

[0134] (2) Match the prediction model corresponding to the model type and task type, and combine it with the real-time network and network element environment to perform real-time prediction of operating parameters;

[0135] Step 4: Based on the real-time network environment and network element conditions, reserve spare resources and adjust parameter importance. Based on the real-time predicted values ​​of operating parameters, combine and orchestrate the inference and training tasks of multiple heterogeneous models and schedule them in parallel to ensure balanced task loads across nodes and that parallel tasks are completed at the same time.

[0136] Step 5: Before running, perform deviation detection on the model task to identify whether it is consistent with the expected situation.

[0137] (1) If there is a deviation from the expected value, the backup resource is started to dynamically expand and adjust the corresponding nodes;

[0138] (2) If everything is running normally, spare resources will be released to each node to speed up the operation;

[0139] □Step 6: Send the analysis results to consumers on time.

[0140] During the model and task classification and identification process in step 3, model classification can be based on the model's structure. For example, models can be categorized into statistical analysis, logical calculation, tree-structured, clustering, and deep network types. Specifically, statistical analysis models include 3sigma and T-test; tree-structured models include decision trees and xgboost; clustering models include kmeans, SMO, and density clustering; and deep network models include LSTM and CNN. Tasks can be categorized into inference and training types based on the request type. Specifically, in Figure 4, An_M1 represents inference model 1, and MT_M2 represents training model 2.

[0141] Among them, in the matching process of the prediction model in step 3, the "model type and task type" and the "prediction model" are a one-to-one fixed mapping relationship. This mapping relationship can be updated after a certain period of time or after the model is retrained.

[0142] In the prediction and analysis process of the operating parameters in step 3, different prediction models are established for different types of model task requests in step 3 to predict and evaluate the operating parameters. As shown in Figure 4, X represents the input item, which may include but is not limited to the data scale, model complexity, number of parameters, model type, etc.; Y represents the predicted value of the operating parameter, which may include but is not limited to the model's computational overhead, storage, computing power, network resources, and required time, etc.; the prediction model may include but is not limited to regression-type numerical prediction such as LSTM and xgboost.

[0143] Specifically, taking the model type as statistical analysis and the task type as reasoning as an example, by collecting data from multiple tasks and training multiple prediction models based on the feature data of multiple tasks, the optimal model is selected as the final prediction model for this type of task based on the prediction accuracy, and the multiple trained prediction models are sorted according to accuracy, and the model with the highest accuracy is selected as the final prediction model for this type of model task request.

[0144] During the multi-model parallel scheduling process in step 4, tasks are divided into multiple types based on the differences between models. Different operating parameters need to be collected and tested for different task types. These parameters are prioritized based on their importance to the request. Scheduling algorithms, such as SJF and SRTF, are then used to ensure that the time, storage, and computing power loads on different work nodes are as close as possible. Under this design, the number of tasks on each work node may vary, and the model tasks executed may also vary, but the total required time will be closer, so parallel tasks can be completed at the same time, meeting the time requirements of the request.

[0145] Specifically, the types of tasks include but are not limited to deep learning model training tasks, machine learning training tasks, traditional statistical computing tasks, deep learning model reasoning tasks, machine learning reasoning tasks, statistical reasoning tasks, and statistical analysis tasks.

[0146] Specifically, different task types require the collection and testing of different operating parameters. For example, for training-type deep learning tasks, considering that tasks are affected by factors such as general computing power and intelligent computing power, we need to obtain data on the memory size, GPU capacity overhead, GPU server computing power, server energy consumption level, network power consumption conditions, storage capacity, and other data for collection and predictive analysis. For inference-type machine learning tasks, considering that tasks are affected by factors such as network and computing power effects, we can obtain operating parameters and data such as network performance level, storage system performance level, network basic power consumption, CPU utilization, storage capacity, and geographical distance for collection and predictive analysis.

[0147] Specifically, the weights of the operating parameters can be expressed as: completion time (weight: 10) > storage (weight: 5) > computing power (weight: 4) > network resources (weight: 3) > = memory (weight: 2) > = hardware server (weight: 2) > = GPU capability (weight: 2).

[0148] Specifically, the weights of operational parameters can be configured and adjusted based on business needs, the time urgency of the task, the real-time network element status, and the network environment. Generally speaking, the primary consideration for NWDAF network element service requests is the model completion time. Other factors, such as storage and computing power, can be adjusted in importance based on the actual usage and idle status of the current network element resources. For example, if the current network element storage is relatively abundant but computing power resources are relatively scarce, the weight can be adjusted to have computing power > storage.

[0149] Among them, in the deviation detection, identification and guarantee process of step 5, due to the real-time changes in the network element's own resources and the network environment, it is easy to cause the predicted results of the operating parameters to deviate from the actual operating conditions, which will have a serious impact on the provision of NWDAF network element services and normal operation. Therefore, it is necessary to introduce a deviation detection, identification and guarantee mechanism. The specific mechanism content can be: (1) If a deviation is detected between the actual and predicted values, the backup resources will be started immediately, and the resources of the corresponding nodes will be dynamically expanded and adjusted; (2) If it is consistent with expectations and the task runs normally, the backup resources will be released to each node to accelerate the operation.

[0150] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0151] Based on the same inventive concept, the embodiment of the present application also provides a network data analysis model scheduling device for implementing the network data analysis model scheduling method involved above. The implementation solution provided by the device is similar to the implementation solution described in the above method. Therefore, the specific limitations in one or more embodiments of the network data analysis model scheduling device provided below can be found in the above limitations on the network data analysis model scheduling method, and will not be repeated here.

[0152] In one embodiment, as shown in FIG7 , a network data analysis model scheduling device is provided, including: a request type determination module 701 , a prediction model matching module 702 , and a task scheduling module 703 , wherein:

[0153] The request type determination module 701 is used to determine the request type of the task request in response to the task request initiated by the service requester for the network data analysis task; the task request is used to request the network data analysis service network element allocation model to process the network data analysis task;

[0154] The prediction model matching module 702 is used to match a target prediction model corresponding to the request type from a pre-updated prediction model matching library; the target prediction model is used to output an operating parameter prediction result for the network data analysis task; the operating parameter prediction result represents the resource usage information required to complete the network data analysis task;

[0155] The task scheduling module 703 is used to schedule each network data analysis task according to the operation parameter prediction results.

[0156] In one embodiment, the device is also used to: obtain the actual measured results of the operating parameters of the network data analysis task; the actual measured results of the operating parameters represent the resource occupancy information of the network data analysis task detected in real time during the execution process; when there is a preset deviation between the actual measured results of the operating parameters and the predicted results of the operating parameters, use spare resources to expand the resource capacity of the target node; the target node includes the working node in the network data analysis service network element where the network data analysis task is located.

[0157] In one embodiment, the device is further used to: allocate backup resources to each working node when there is no preset deviation between the actual measurement results of the operating parameters and the predicted results of the operating parameters, and the network data analysis task is executed normally.

[0158] In one embodiment, the task scheduling module is further used to: obtain the target weight value corresponding to each operating parameter in the operating parameter prediction result; and schedule each network data analysis task according to the priority order reflected in the target weight value.

[0159] In one embodiment, the task scheduling module is also used to: set initial weight values ​​for each operating parameter according to the business requirements of the network data analysis task; adjust the initial weight values ​​according to the occupancy ratio of each resource in the network data analysis service network element to obtain the target weight value.

[0160] In one embodiment, the device is also used to: obtain at least two prediction models that meet the request type corresponding to the network data analysis task; train the prediction model based on the characteristic data of the network data analysis task; and update the prediction model matching library based on the prediction model with the highest prediction accuracy among the trained prediction models.

[0161] Each module in the above-mentioned network data analysis model scheduling device can be implemented in whole or in part through software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of each of the above modules.

[0162] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be shown in Figure 8. The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, memory, and I / O interface are connected via a system bus, and the communication interface is connected to the system bus via the I / O interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store models and related data. The I / O interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a network data analysis model scheduling method.

[0163] Those skilled in the art will understand that the structure shown in FIG8 is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0164] In one embodiment, a computer device includes a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0165] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0166] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0167] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions.

[0168] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided herein may be, but are not limited to, a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, and the like.

[0169] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0170] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A network data analysis model scheduling method, comprising: In response to a task request initiated by a service requester for a network data analysis task, determining a request type of the task request; The task request is used to request the network data analysis service network element allocation model to process the network data analysis task; A target prediction model corresponding to the request type is matched from a pre-updated prediction model matching library; the target prediction model is used to output an operation parameter prediction result for the network data analysis task; the operation parameter prediction result represents resource occupancy information required to complete the network data analysis task; as well as The network data analysis tasks are scheduled according to the operation parameter prediction results.

2. The method according to claim 1, further comprising: Obtaining the measured results of the operating parameters of the network data analysis task; The measured results of the operating parameters represent the resource occupancy information of the network data analysis task detected in real time during the execution process; as well as In the case where there is a preset deviation between the measured result of the operating parameter and the predicted result of the operating parameter, using spare resources to expand the resource capacity of the target node; The target node includes a working node in the network data analysis service network element where the network data analysis task is located.

3. The method according to claim 2, further comprising: When there is no preset deviation between the actual measurement result of the operating parameter and the predicted result of the operating parameter, and the network data analysis task is executed normally, the backup resource is allocated to each of the working nodes.

4. The method according to claim 1, wherein: The scheduling of each of the network data analysis tasks according to the operation parameter prediction result includes: Obtaining a target weight value corresponding to each operating parameter in the operating parameter prediction result; and Each of the network data analysis tasks is scheduled according to the priority order reflected in the target weight value.

5. The method according to claim 4, wherein: The obtaining of the target weight value corresponding to each operating parameter in the operating parameter prediction result includes: According to the business requirements of the network data analysis task, setting initial weight values ​​for each operating parameter; and According to the occupancy ratio of various resources in the network data analysis service network element, the initial weight value is adjusted to obtain the target weight value.

6. The method according to any one of claims 1 to 5, further comprising: Obtain at least two prediction models that meet the request type corresponding to the network data analysis task; Training the prediction model according to the characteristic data of the network data analysis task; as well as The prediction model matching library is updated according to the prediction model with the highest prediction accuracy among the trained prediction models.

7. A network data analysis model device, comprising: A request type determination module, configured to determine a request type of the task request in response to a task request initiated by a service requester for a network data analysis task; The task request is used to request the network data analysis service network element allocation model to process the network data analysis task; A prediction model matching module, used to match a target prediction model corresponding to the request type from a pre-updated prediction model matching library; the target prediction model is used to output an operation parameter prediction result for the network data analysis task; the operation parameter prediction result represents resource occupancy information required to complete the network data analysis task; as well as The task scheduling module is used to schedule each of the network data analysis tasks according to the prediction results of the operating parameters.

8. The device according to claim 7, wherein the device is further used for: Obtaining the measured results of the operation parameters of the network data analysis task; the measured results of the operation parameters represent the resource occupancy information of the network data analysis task during the execution process detected in real time; as well as In the case where there is a preset deviation between the measured result of the operating parameter and the predicted result of the operating parameter, using spare resources to expand the resource capacity of the target node; The target node includes a working node in the network data analysis service network element where the network data analysis task is located.

9. The device according to claim 8, wherein the device is further used for: When there is no preset deviation between the actual measurement result of the operating parameter and the predicted result of the operating parameter, and the network data analysis task is executed normally, the backup resource is allocated to each of the working nodes.

10. The device according to claim 7, wherein: The task scheduling module is also used for: Obtaining a target weight value corresponding to each operating parameter in the operating parameter prediction result; and Each of the network data analysis tasks is scheduled according to the priority order reflected in the target weight value.

11. The device according to claim 10, wherein: The task scheduling module is also used for: According to the business requirements of the network data analysis task, setting initial weight values ​​for each operating parameter; and According to the occupancy ratio of various resources in the network data analysis service network element, the initial weight value is adjusted to obtain the target weight value.

12. The device according to any one of claims 7 to 11, wherein the device is further used for: Obtain at least two prediction models that meet the request type corresponding to the network data analysis task; Training the prediction model according to the characteristic data of the network data analysis task; and The prediction model matching library is updated according to the prediction model with the highest prediction accuracy among the trained prediction models.

13. A computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method according to any one of claims 1 to 6 when executing the computer program.

14. A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

15. A computer program product, comprising a computer program, which implements the steps of the method according to any one of claims 1 to 6 when executed by a processor.

Images