Network performance prediction model training method and device, equipment and storage medium

Abstract

This application discloses a method, apparatus, device, and storage medium for training a network performance prediction model, relating to the field of communication technology. The method includes: acquiring a training dataset comprising multiple network indicator data, wherein the multiple network indicator data includes: network indicator data corresponding to a first unit time period for each of several adjacent historical dates, and network indicator data corresponding to each unit time period for each of several unit time periods within a preset duration preceding the target unit time period; inputting the multiple network indicator data into a network performance prediction model to determine the predicted network parameters corresponding to a second unit time period adjacent to the target unit time period; and adjusting the model parameters of the network performance prediction model based on the difference between the predicted network parameters corresponding to the second unit time period and the actual network parameters corresponding to the second unit time period, thereby obtaining the trained network performance prediction model.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a method, apparatus, device and storage medium for training a network performance prediction model. Background Technology

[0002] Currently, with the rapid development of communication technology, the types of network application services are constantly increasing, and the number of network users and the usage of network resources are also growing rapidly. To ensure network performance, communication service operators typically need to analyze and study the network's operational status. In communication service-related technologies, the characteristics of network traffic and network quality indicators can be analyzed and predicted to evaluate network performance, enabling timely detection of network anomalies and faults, and ensuring a positive user experience.

[0003] However, current network status predictions are typically based on historical network traffic data for all users. But because different networks, such as 5G and 4G private networks, and networks operating under different usage scenarios (e.g., different users, different usage times), can have significantly different characteristics, current network performance prediction models have relatively low accuracy. Summary of the Invention

[0004] This application provides a method, apparatus, device, and storage medium for training a network performance prediction model, which is used to improve the prediction accuracy of the network performance prediction model.

[0005] To achieve the above objectives, this application adopts the following technical solution:

[0006] Firstly, a method for training a network performance prediction model is provided. This method includes: acquiring a training dataset comprising multiple network indicator data, wherein the multiple network indicator data includes: network indicator data corresponding to a first unit time period for each historical date among multiple adjacent historical dates; network indicator data corresponding to each unit time period within a preset duration preceding the target unit time period; the first unit time period being the same unit time period as the target unit time period for each historical date; the network indicator data including: network parameters, private network characteristic parameters, and time features, wherein the network parameters include at least one of the following: user traffic, packet loss rate, service latency, and user rate; the private network characteristic parameters are used to indicate the network type; and the time features are used to indicate whether the time corresponding to the network indicator data belongs to a preset date; inputting the multiple network indicator data into the network performance prediction model respectively, and determining the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period; adjusting the model parameters of the network performance prediction model based on the difference between the predicted network parameters corresponding to the second unit time period and the actual network parameters corresponding to the second unit time period, to obtain the trained network performance prediction model.

[0007] In one possible implementation, the network performance prediction model includes a private network feature learning module; the method further includes: inputting the private network feature corresponding to each network indicator data in multiple network indicator data into the private network feature learning module in the network performance prediction model, and determining the private network feature parameters corresponding to each network indicator data. The private network feature includes at least one of the following: industry type, private network type and geographical location.

[0008] In one possible implementation, the network performance prediction model further includes at least one of the following: a long-term learning module and a short-term learning module; inputting multiple network indicator data into the network performance prediction model to determine the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period, including: inputting the network indicator data corresponding to the first unit time period included in each of the multiple adjacent historical dates into the long-term learning module of the network performance prediction model to determine the first parameter corresponding to the network indicator data corresponding to the target unit time period; and / or, inputting the network indicator data corresponding to each unit time period included in the multiple unit time periods included within the preset duration adjacent to the target unit time period into the short-term learning module of the network performance prediction model to determine the second parameter corresponding to the network indicator data corresponding to the target unit time period; and determining the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period based on the first parameter and / or the second parameter corresponding to the target unit time period.

[0009] In one possible implementation, the network performance prediction model includes: a long-term learning module, a short-term learning module, and a feature fusion module; based on the first parameter and / or the second parameter corresponding to the target unit time period, the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period are determined, including: inputting the first parameter and the second parameter corresponding to the target unit time period into the feature fusion module in the network performance prediction model to determine the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period.

[0010] In one possible implementation, the method further includes: obtaining target data corresponding to a third unit time period, where the third unit time period is the time period after the current time, and the target data is network indicator data corresponding to multiple unit time periods that are the same as the third unit time period included in each of multiple historical dates before the current time; inputting the target data into the trained network performance prediction model to obtain the predicted network parameters corresponding to the third unit time period.

[0011] Secondly, a network performance prediction model training device is provided, comprising: an acquisition unit, a determination unit, and a processing unit; the acquisition unit is used to acquire a training dataset including multiple network indicator data, the multiple network indicator data including: network indicator data corresponding to a first unit time period included in each of multiple adjacent historical dates, network indicator data corresponding to each unit time period included in multiple unit time periods included in a preset duration adjacent to the target unit time period, the first unit time period being a unit time period in each historical date that is the same as the target unit time period; the network indicator data includes: network parameters, private network characteristic parameters, and time features, wherein the network parameters include at least one of the following: user traffic, packet loss rate, service latency, user rate, private network characteristic parameters are used to indicate network type, and time features are used to indicate whether the time corresponding to the network indicator data belongs to a preset date; the determination unit is used to input the multiple network indicator data into the network performance prediction model respectively, and determine the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period; the processing unit is used to adjust the model parameters of the network performance prediction model based on the difference between the predicted network parameters corresponding to the second unit time period and the actual network parameters corresponding to the second unit time period, to obtain the trained network performance prediction model.

[0012] In one possible implementation, the network performance prediction model includes a private network feature learning module and a determination unit, which inputs the private network feature corresponding to each network indicator data from multiple network indicator data into the private network feature learning module in the network performance prediction model to determine the private network feature parameters corresponding to each network indicator data. The private network feature includes at least one of the following: industry type, private network type, and geographical location.

[0013] In one possible implementation, the network performance prediction model further includes at least one of the following: a long-term learning module and a short-term learning module; a determining unit, configured to input network indicator data corresponding to a first unit time period included in each of a plurality of adjacent historical dates into the long-term learning module of the network performance prediction model, respectively, to determine a first parameter corresponding to the network indicator data corresponding to the target unit time period; and / or, a determining unit, configured to input network indicator data corresponding to each of a plurality of unit time periods included in a preset duration adjacent to the target unit time period, respectively, into the short-term learning module of the network performance prediction model, respectively, to determine a second parameter corresponding to the network indicator data corresponding to the target unit time period; and a determining unit, configured to determine the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period based on the first parameter and / or the second parameter corresponding to the target unit time period.

[0014] In one possible implementation, the network performance prediction model includes: a long-term learning module, a short-term learning module, and a feature fusion module; and a determination unit, which inputs the first and second parameters corresponding to the target unit time period into the feature fusion module in the network performance prediction model to determine the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period.

[0015] In one possible implementation, an acquisition unit is used to acquire target data corresponding to a third unit time period, where the third unit time period is the time period after the current time, and the target data is network indicator data corresponding to multiple unit time periods that are the same as the third unit time period, included in each of multiple historical dates before the current time; a determination unit is used to input the target data into the trained network performance prediction model to obtain the predicted network parameters corresponding to the third unit time period.

[0016] Thirdly, an electronic device is provided, comprising: a processor and a memory; wherein the memory is used to store one or more programs, the one or more programs including computer-executable instructions, and when the electronic device is running, the processor executes the computer-executable instructions stored in the memory to cause the electronic device to perform a network performance prediction model training method as described in the first aspect.

[0017] Fourthly, a computer-readable storage medium is provided for storing one or more programs, the one or more programs including instructions that, when executed by a computer, cause the computer to perform a network performance prediction model training method as described in the first aspect.

[0018] This application provides a method, apparatus, device, and storage medium for training a network performance prediction model, applied in scenarios where a network performance prediction model is trained to improve its prediction accuracy. When training the network performance prediction model, network indicator data corresponding to the same time period as the target time period are obtained for each historical date from multiple adjacent historical dates, as well as network indicator data corresponding to each time period from multiple time periods within a preset duration preceding the target time period, resulting in a training dataset containing multiple network indicator data. These multiple network indicator data are then input into the network performance prediction model to determine the corresponding predicted network parameters for the adjacent second time period after the target time period. Based on the difference between the predicted network parameters for the second time period and the actual network parameters for the second time period, the model parameters of the network performance prediction model are adjusted to obtain the trained network performance prediction model. The above method enables the training of a network performance prediction model based on network indicator data, including network parameters, private network characteristic parameters, and time characteristics. This allows the prediction results (predicted network parameters) of the network performance prediction model to fully consider the impact of differences in network parameters, private network characteristics, and time characteristics, thereby improving the prediction accuracy of the network performance prediction model. Attached Figure Description

[0019] Figure 1 A schematic diagram of a network performance prediction model training system provided for an embodiment of this application;

[0020] Figure 2 A flowchart illustrating a network performance prediction model training method provided in this application. Figure 1 ;

[0021] Figure 3 A flowchart illustrating a network performance prediction model training method provided in this application. Figure 2 ;

[0022] Figure 4 A flowchart illustrating a network performance prediction model training method provided in this application. Figure 3 ;

[0023] Figure 5 A flowchart illustrating a network performance prediction model training method provided in this application. Figure 4 ;

[0024] Figure 6 A flowchart illustrating a network performance prediction model training method provided in this application. Figure 5 ;

[0025] Figure 7 A flowchart illustrating a network performance prediction model training method provided in this application. Figure 6 ;

[0026] Figure 8 A flowchart illustrating a network performance prediction model training method provided in this application. Figure 7 ;

[0027] Figure 9 A flowchart illustrating a network performance prediction model training method provided in this application. Figure 8 ;

[0028] Figure 10 A schematic diagram of a network performance prediction model training device provided for an embodiment of this application;

[0029] Figure 11 This is a schematic diagram of an electronic device structure provided for an embodiment of this application. Detailed Implementation

[0030] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0031] In the description of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. "And / or" in this document is merely a description of 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, and B alone. Furthermore, "at least one" and "multiple" refer to two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0032] In network-related technologies, 5G private networks, as a dedicated service, create a dedicated network connection system for service users and enterprises, thereby providing better guarantees in terms of network stability, serviceability, and security, filling gaps in areas that public network communication cannot cover. With the popularization of 5G private networks, the types and number of network application services are also constantly increasing. In network maintenance technologies, the operational status of private networks can be analyzed to predict network traffic and ensure network performance. However, due to the inherent characteristics of 5G, such as high bandwidth and massive connectivity, the factors influencing traffic are gradually increasing, and the complexity of traffic prediction is constantly rising.

[0033] Existing traffic prediction technologies primarily rely on general training datasets, such as historical user traffic data and historical network traffic data, using various network feature parameters to predict user browsing trends across time or geographic dimensions. Regarding specific network prediction metrics, current prediction models mainly focus on improving quality of experience (QoE). Therefore, they primarily rely on network signaling data from the user's terminal to determine its network type. This involves acquiring network perception data for the user's terminal, inputting it into a pre-defined classification model, and then using the model's output to determine the user's perception status, thus achieving network perception prediction (e.g., user traffic).

[0034] It should be noted that when predicting the network performance of a private network, that is, predicting the network usage of the entire private network (such as different 5G private networks), the users in this application embodiment mainly refer to private network users. Private network users can be understood as the entire group of users of the private network. For example, predicting the network performance of private network users in an education park can be understood as predicting the network performance of the entire education park.

[0035] The network performance prediction model training method provided in this application embodiment can be applied to a network performance prediction model training system. Figure 1 A schematic diagram of one structure of the training system for this network performance prediction model is shown. Figure 1 As shown, the network performance prediction model training system 10 includes: electronic device 11, server 12 and base station 13.

[0036] The network performance prediction model training system 10 can be used in the Internet of Things. The network performance prediction model training system 10 (such as electronic device 11, server 12 and base station 13) may include hardware such as multiple central processing units (CPUs), multiple memories, and storage devices storing multiple operating systems.

[0037] Electronic device 11 can be used in the Internet of Things to process data. For example, electronic device 11 can interact with server 12 to obtain training datasets from server 12 and obtain a trained network performance prediction model.

[0038] Optionally, the electronic device 11 can also obtain network indicator data of the user to be predicted from the server 12, and then use the network indicator data of the user to be predicted to predict the network performance of the user (predict network parameters).

[0039] Server 12 is used for data storage. For example, server 12 interacts with base station 13 to obtain and store network indicator data generated by base station 13, thus obtaining a training dataset. It also interacts with base station 13 to obtain network indicator data of the user to be detected.

[0040] Optionally, server 12 can be a database that can store data, such as a MySQL database.

[0041] Base station 13 is used to record data. For example, base station 13 can record network indicator data of each access user so that server 12 can obtain the network indicator data.

[0042] Optional, such as Figure 1 As shown, there can be multiple base stations 13 to ensure that there is enough training data and data features in the training dataset, thereby ensuring the comprehensiveness of training the network performance prediction model.

[0043] The following description, in conjunction with the accompanying drawings, describes a network performance prediction model training method provided in an embodiment of this application.

[0044] like Figure 2 As shown in the embodiment of this application, a network performance prediction model training method includes S201-S203:

[0045] S201. Obtain a training dataset that includes multiple network metric data.

[0046] The network indicator data includes: network indicator data corresponding to the first unit time period of each historical date in multiple adjacent historical dates; network indicator data corresponding to each unit time period in multiple unit time periods within a preset duration preceding the target unit time period; the first unit time period is the same unit time period as the target unit time period in each historical date; the network indicator data includes: network parameters, private network characteristic parameters, and time characteristics; the network parameters include at least one of the following: user traffic, packet loss rate, service latency, and user rate; the private network characteristic parameters are used to indicate the network type; and the time characteristics are used to indicate whether the time corresponding to the network indicator data belongs to the preset date.

[0047] Optionally, historical usage data of each private network user's real private network traffic data and quality indicator data can be collected through electronic devices, and these data can be stored in a preset database (such as a MySQL database) to obtain a training dataset.

[0048] Optionally, the collected data can be normalized to eliminate differences in the units of measurement and minimize the adverse effects caused by outlier sample data.

[0049] It should be noted that normalizing the data can also reduce the gradient of the subsequent training model.

[0050] Optionally, the first unit time period included in each of the multiple adjacent historical dates can be understood as the unit time period that is the same as the target unit time period among the multiple unit time periods included in each of the multiple adjacent historical dates before the target unit time period.

[0051] For example, if the target time period is 9:00-10:00 on March 2, then the first time period can be 9:00-10:00 every day from February 2 to March 2.

[0052] Optionally, the size of multiple historical dates and unit time periods can be determined in conjunction with specific model training requirements. For example, in the above example, multiple historical dates are each day of the previous month, and the unit time period is each hour.

[0053] Optionally, each of the multiple unit time periods included within the preset duration adjacent to the target unit time period can be understood as multiple adjacent unit time periods before the target unit time period.

[0054] For example, each of the multiple unit time periods included in the preset duration adjacent to the target unit time period can be each hour (unit time period) within the twenty-four hours (preset duration) before the target unit time period.

[0055] Specifically, the target time period can be from 9:00 to 10:00 on March 2nd. Then, each of the multiple time periods included in the preset duration adjacent to the target time period can be any hour between 9:00 on March 1st and 9:00 on March 2nd.

[0056] Optionally, the preset duration and the size of the unit time period can be determined based on the specific model training requirements.

[0057] It should be noted that the network metric data corresponding to a unit time period can be understood as the average network metric data within that unit time period, such as the average network parameters within an hour.

[0058] In one possible implementation, the average network metric data within a unit time period can also be the network metric data at the start time, the end time, or the intermediate time within that time period.

[0059] It should be noted that the unit time period can be understood as the time granularity, that is, the granularity of network performance prediction, such as per hour, per half hour, per minute, etc. The specific size of the unit time period can be determined in combination with the specific network performance prediction requirements, such as the unit time period of per hour in the example above.

[0060] Optionally, in addition to user traffic, packet loss rate, private network characteristic parameters, service latency and user rate in the embodiments of this application, other network parameters can be selected in combination with specific business requirements, as well as relevant industry standards, communication protocols, etc.

[0061] In this application, when the embodiments of this application are applied to network performance prediction of private network users, user traffic mainly refers to private network user traffic.

[0062] It should be noted that existing private network performance prediction technologies primarily rely on network traffic (user traffic) for prediction, resulting in a somewhat one-sided assessment. In this embodiment, however, the aforementioned network parameters (user traffic, packet loss rate, private network characteristic parameters, service latency, and user speed) are selected to comprehensively consider multiple indicators, enabling the analysis and prediction of user network usage status and providing a more comprehensive analysis of user network quality and usage.

[0063] Optional, private network characteristic parameters can be private network characteristic parameters determined by a preset algorithm based on the network type.

[0064] Optionally, network type can be understood as a description of the source and nature of network indicator data, such as the network deployment type of network indicator data, the industry type of network indicator data, etc.

[0065] It should be noted that different industries and users have significantly different network resource usage habits and different requirements for various network indicators. For example, the network resource usage of e-commerce industries (such as online marketers) is significantly greater than that of the education industry (such as students). Therefore, this application embodiment introduces private network characteristic parameters to differentiate network indicator data for different industries and users.

[0066] Furthermore, considering that the network usage of the same user on holidays may differ significantly from that on weekdays, such as the significant impact of shopping festivals on the internet industry, especially the e-commerce industry, and the potential surge or plunge in the number of users on some private networks during holidays (such as the education industry), this application embodiment introduces time features in addition to the private network feature parameters to characterize whether the date of the network indicator data falls within a particularly special period for the user of that private network, thereby improving the accuracy of the network performance prediction model.

[0067] Optionally, whether the time corresponding to the network indicator data belongs to a preset date can be understood as whether the time corresponding to the network indicator data belongs to holidays, peak periods, etc.

[0068] For example, the preset date can be set based on specific business needs and expert experience. It can be a statutory holiday date or a fixed time period each day, such as 20:00-22:00.

[0069] It should be noted that since the usage scenarios of different private networks may differ, the preset dates for different private networks may also be different. The preset dates for each private network can be set based on the experience of the maintenance personnel and users of that private network.

[0070] S202. Input multiple network indicator data into the network performance prediction model respectively, and determine the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period.

[0071] It should be noted that during network operation, a user's network usage usually has time characteristics, that is, the value of a user's network metric data is different at different times, and the value of each network metric data (or the difference between users' network metric data) is usually related to the change over time.

[0072] Therefore, in the embodiments of this application, a network performance prediction model can be constructed based on a temporal neural network, such as a recurrent neural network (RNN), a long short-term memory network (LSTM), a gated recurrent unit (GRU), or a Transformer.

[0073] Optionally, the basic process of a temporal neural network is as follows: Figure 3 As shown, the data (values) at the previous time step t0 are used to predict the data (values) at the next time step t1. This process is repeated until the result value (predicted value) t, which has learned the temporal features, is obtained. n .

[0074] Optionally, the second unit time period adjacent to the target unit time period refers to the second unit time period that is adjacent to the target unit time period and has the same time length (unit time period) as the target unit time period.

[0075] For example, if the target time period is 9:00-10:00 on March 2, then the second time period is 10:00-11:00 on March 2.

[0076] S203. Based on the difference between the predicted network parameters corresponding to the second unit time period and the actual network parameters corresponding to the second unit time period, adjust the model parameters of the network performance prediction model to obtain the trained network performance prediction model.

[0077] Optionally, if the difference between the predicted network parameters and the actual network parameters for the second time unit exceeds a preset difference threshold, the model parameters of the network performance prediction model can be adjusted to obtain the trained network performance prediction model. The size of the preset difference threshold can be determined by combining business experience and expert experience, and adjusted and retrained based on the accuracy of the prediction results.

[0078] For example, when the difference between the predicted network parameters and the actual network parameters is greater than 50%, the model parameters of the network performance prediction model can be adjusted to obtain the trained network performance prediction model.

[0079] In one possible implementation, the training process for the network prediction model based on a temporal neural network is as follows: Figure 4 As shown, a training dataset is first obtained, and after preprocessing, it is fed into a pre-designed temporal neural network structure (network performance prediction model) for deep learning training, ultimately obtaining the learned t0-t... n The data over a time period shows patterns, and the network performance prediction model is trained accordingly.

[0080] Optionally, during network training, an error backpropagation mechanism can be used to adaptively adjust network parameters to obtain the prediction results (predicted network parameters) that best match the real results (real network parameters).

[0081] It should be noted that the error backpropagation mechanism can be understood as follows: the error between the predicted result and the actual result is backpropagated and distributed to all units (model parameters) to obtain the error signal of each parameter. This error signal is used as the basis for adjusting the parameter weights. Through repeated learning, the parameter weights are continuously adjusted, that is, the network is continuously trained. When the final output error is less than a preset threshold or the preset number of learning iterations is reached, the training ends, and the trained model is obtained.

[0082] It should be noted that the embodiments of this application are mainly used to predict the network performance of private network users (such as 5G private network users), but they can also be used in scenarios where the network performance of individual users, such as ordinary network (public network) users, is predicted.

[0083] In this embodiment, when training the network performance prediction model, network indicator data corresponding to the same unit time period as the target unit time period are obtained for each historical date among multiple adjacent historical dates, and network indicator data corresponding to each unit time period among multiple unit time periods within a preset duration preceding the target unit time period, resulting in a training dataset including multiple network indicator data. A training dataset including multiple network indicator data corresponding to the target time period, comprising network parameters, private network characteristic parameters, and time characteristics, is obtained, and these multiple network indicator data are respectively input into the network performance prediction model to determine the corresponding predicted network parameters for the adjacent second time period after the target time period. Then, based on the difference between the predicted network parameters corresponding to the second time period and the actual network parameters corresponding to the second time period, the model parameters of the network performance prediction model are adjusted to obtain the trained network performance prediction model. The above method enables the training of a network performance prediction model based on network indicator data, including network parameters, private network characteristic parameters, and time characteristics. This allows the prediction results (predicted network parameters) of the network performance prediction model to fully consider the impact of differences in network parameters, private network characteristics, and time characteristics in the network indicator data. By combining multiple features of historical network indicator data, the predicted network parameters can be obtained, thereby improving the prediction accuracy of the network performance prediction model.

[0084] In one possible implementation, the network performance prediction model includes a private network feature learning module, such as... Figure 5 As shown in the embodiment of this application, a network performance prediction model training method further includes step S301:

[0085] S301. Input the private network features corresponding to each network indicator data in the multiple network indicator data into the private network feature learning module in the network performance prediction model to determine the private network feature parameters corresponding to each network indicator data.

[0086] The characteristics of a private network include at least one of the following: industry type, private network type, and geographical location.

[0087] In the training process of network performance prediction models, if excessive attention is paid to the differences between users, such as training a separate prediction model for each user to predict their network performance, it will consume a lot of resources (such as training time and computing resources), and this method is difficult to implement when the user group is large. In addition, training using a single user's network can lead to data sparsity problems, resulting in poor training effects. Therefore, in related network prediction techniques, prediction models are usually trained based on the network parameter data of all users, resulting in a more general and universal prediction result, but with lower accuracy.

[0088] In this embodiment, by introducing the concept of private network characteristics, and utilizing the various descriptions and properties of different private networks, a neural network is used to learn and obtain fused characteristics (private network characteristic parameters) that represent each type of private network. This allows the network performance prediction model to have relative robustness while focusing on the uniqueness of private network users, thus enabling the model to provide prediction functions for unknown private network users.

[0089] Optional, industry type can be understood as the application scenario of the private network user, such as the education private network in the education park, the industrial private network in the industrial park, the medical private network in the medical unit, the railway private network in the railway, etc.; private network type can be understood as the network deployment type of the private network user, such as independent private network, hybrid private network, virtual private network, etc.; geographical location refers to the location of the private network user, such as specific province, city and other administrative region information, geographical latitude and longitude information, etc.

[0090] It should be noted that when this application is applied to private network users, the industry type, private network type, and geographical location of the same private network user are usually fixed.

[0091] Optionally, the industry type, network type, and geographical location of each private network user are determined when the private network is deployed. Therefore, when acquiring network indicator data, the industry type, network type, and geographical location of the network indicator data under the private network can be marked by data tagging or other methods.

[0092] In one possible implementation, the industry type, private network type, and geographical location corresponding to each network indicator data can be input into the private network feature learning module to obtain private network feature parameters (that is, the above-mentioned preset algorithm is integrated into the network performance prediction model), as shown in Formula 1 below:

[0093] P = f(p1, p2, p3) (Formula 1)

[0094] Wherein, P is the private network characteristic parameter, p1 is the industry type, p2 is the private network type, p3 is the geographical location, and f() is the function relationship, that is, the correspondence between the private network characteristic parameter and the private network characteristic in the embodiment of this application.

[0095] Specifically, during the training of the network performance prediction model, the industry type p1, private network type p2, and geographical location p3 corresponding to the network indicator data are input into the private network feature learning module to obtain the private network feature parameters P in the network indicator data. Then, the network performance prediction model is trained based on the network parameters, private network feature parameters, and time features in the network indicator data.

[0096] In one possible implementation, when user traffic, packet loss rate, service latency, and user speed are simultaneously selected as network parameters, and network parameters, private network characteristic parameters, and time characteristics are used as network indicator data to train the network performance prediction model, the input data can be represented by the feature vector T as shown in Formula 2 below:

[0097] Formula 2: T = [X, P, F]

[0098] Where x is the network parameter, X = [x1, x2, x3, x4], x1 is the user traffic, x2 is the packet loss rate, x3 is the service latency, x4 is the user rate, P is the private network characteristic parameter, and F is the time characteristic.

[0099] It should be noted that in Formulas 1 and 2 above, network parameter x can be represented numerically, such as user traffic x1, packet loss rate x2, service latency x3, and user speed x4, each with corresponding specific values. However, industry type p1, private network type p2, geographical location p3, and time characteristic F all need to be encoded.

[0100] Specifically, industry type p1 and private network type p2 can be encoded using one-hot encoding, geographical location p3 can use existing provincial and municipal codes, and time feature F can be represented by a binary representation of 0 and 1.

[0101] It should be noted that the binary representation of 0-1 can be understood as follows: "0" indicates that the time corresponding to the network indicator data does not belong to the preset date, and "1" indicates that the time corresponding to the network indicator data belongs to the preset date.

[0102] Optionally, the network architecture of the private network feature learning module can be determined based on specific usage needs and business experience. For example, in one implementation of this application, the private network feature learning module can be implemented using a deep neural network (DNN) architecture.

[0103] Optional, such as Figure 6 As shown, the private network feature learning module in this embodiment is mainly used to determine the private network feature parameters of the network indicator data based on the industry type, private network type, and geographical location corresponding to the network indicator data, and then determine the predicted network parameters corresponding to the second unit time period based on the network parameters, private network feature parameters, and time features included in the target indicator data.

[0104] Optionally, in one possible implementation, when the difference between the predicted network parameters corresponding to the second unit time period and the actual network parameters corresponding to the second unit time period is greater than a preset first threshold, the relevant parameters in the private network feature learning module can be adjusted. The specific adjustment method can be referred to the relevant description in section S203 above, and will not be repeated here.

[0105] Optionally, the first threshold of the private network feature learning module can be determined and adjusted by combining business experience and information such as actual prediction results.

[0106] Optionally, in one possible implementation, when the private network feature parameters of the network indicator data are determined through the private network feature learning module, the network indicator data in the training dataset may include user traffic, private network features, and time features. The network indicator data is then input into the network performance prediction model, and the private network feature parameters of the private network features are determined through the private network feature learning module in the network performance prediction model. In turn, the network performance prediction model is trained based on user traffic, private network feature parameters, and time features.

[0107] In this embodiment, the private network feature learning module learns the impact of different private network features, such as different industry types, different private network types, and different geographical locations, on network parameters. The degree of this impact is then quantified using private network feature parameters, and these parameters are incorporated into the feature generation results. This allows for full consideration of the characteristics of different private network features when predicting network performance, thereby improving the prediction accuracy of the network performance prediction model.

[0108] In one possible implementation, the network performance prediction model further includes at least one of the following: a long-term learning module and a short-term learning module, such as... Figure 7 As shown in the embodiment of this application, in a network performance prediction model training method, the above-mentioned S202 includes S401-S403:

[0109] S401. Input the network indicator data corresponding to the first unit time period of each of the multiple adjacent historical dates into the long-term learning module of the network performance prediction model to determine the first parameter corresponding to the network indicator data of the target unit time period.

[0110] S402. Input the network indicator data corresponding to each of the multiple unit time periods included in the preset duration adjacent to the target unit time period into the short-term learning module in the network performance prediction model to determine the second parameter corresponding to the network indicator data of the target unit time period.

[0111] Optionally, the first parameter can be understood as: the first predicted value of the network parameter corresponding to the second time period, derived from learning the long-term patterns of the data corresponding to the target unit time period (the network indicator data corresponding to the first unit time period included in each of the multiple adjacent historical dates); the second parameter can be understood as: the second predicted value of the network parameter corresponding to the second time period, derived from learning the short-term patterns of the data corresponding to the target unit time period (the network indicator data corresponding to each of the multiple unit time periods included in the preset duration adjacent to the target unit time period).

[0112] It should be noted that, due to the time-series nature of indicator values ​​(described in section S202 above), network indicator data typically exhibits a long-term, universal sequence, as well as instances where network indicator data (such as network parameters) shows unique characteristics at a specific time. In other words, user network usage generally exhibits temporal regularity in the long term, while short-term data better reflects the characteristics of current network usage. Therefore, in this embodiment, a long-term learning module learns the long-term universal characteristics of the data; a short-term learning module makes the prediction results closer to the current network usage; and by combining the long-term and short-term learning modules, the prediction results become more stable and accurate.

[0113] Optionally, in order to ensure the accuracy of the data and to ensure that the learning of long-term data features and short-term data features are not interfered with during the training process, in this embodiment of the application, two temporal networks N1 and N2 can be used in the network architecture. The long-term features of the data are learned through the temporal network N1 (i.e., the long-term learning module), and the short-term features of the data are learned through the temporal network N2 (i.e., the short-term learning module), so as to obtain more accurate prediction results based on the long-term features and the short-term features.

[0114] It should be noted that the specific architecture of time series network N1 and time series network N2 can be determined based on specific usage requirements.

[0115] For example, both time series networks N1 and N2 can use Transformer as the network infrastructure. Time series network N1 learns the long-term characteristics of the data, such as the network indicator data of the current hour of each day in the previous month (the target unit time period), so that time series network N1 learns the characteristic patterns of the current hour of a certain day (i.e., the target unit time period) in the long-term data. Time series network N2 learns the short-term characteristics of the data, such as the network indicator data of each hour in the previous 24 hours, so that time series network N2 learns the characteristic patterns of each hour of the current day in the short-term data.

[0116] It should be noted that the Transformer architecture mainly consists of an encoder and a decoder, unlike traditional convolutional and recurrent units. It primarily relies on a self-attention mechanism, thus breaking the limitations of sequential networks and enabling parallel computation. Furthermore, since the input feature information in this embodiment is in numerical or encoded form, rather than the non-numerical information such as text input in traditional Transformer applications, there is no need to perform traditional Transformer word embedding operations in this embodiment. Correspondingly, the six selected features (user traffic, packet loss rate, service latency, user rate, private network feature parameters, and time features) can be used as word embedding dimension data, thereby combining the features into the "word" representation input in this embodiment. The final network input dimension is shown in Formula 3 below:

[0117] Formula 3: I = I(b*m*c)

[0118] Where b refers to the size of the batch data, i.e., how many data records are in a batch; m is the time series dimension size, representing the length of the time series data. In the example above of this application embodiment, in time series network N1, it is the data length of the current hour of the previous month, and m takes the default length of 30. In time series network N2, it is the data length of the previous 24 hours, and m takes the length of 24; e is the output feature dimension (in the traditional Transformer, it refers to the word embedding dimension). In this application embodiment, when the above six features are used at the same time, e takes the length of 6.

[0119] It should be noted that when training a network performance prediction model using a training dataset, multiple training data points are usually randomly selected in batches from the training dataset for training. Each training data point includes network metric data for the first unit time period of each historical period in multiple historical periods, as well as network metric data for each unit time period in multiple unit time periods within a preset duration adjacent to the target unit time period.

[0120] It should be noted that randomly selecting multiple training data points from the training dataset in batches can be understood as randomly selecting multiple training data points from the training dataset each time, until all training data in the training dataset is removed.

[0121] Optionally, the training dataset may also include a training result set, which includes the true values ​​(true network parameters) corresponding to each training data point.

[0122] When multiple training data points are randomly selected in batches from the training dataset for training, a complete training session is considered to have occurred when all the training data in the training dataset is removed.

[0123] In one possible implementation, the training of the network performance prediction model is considered complete when the difference between the prediction result and the actual result meets the requirements in each complete training process, thus obtaining the trained network performance prediction model.

[0124] It should be noted that, since the Transformer's self-attention mechanism cannot obtain the order of data input, in this embodiment, the input time vector t = {t1, t2, ..., t...} n-1 Furthermore, positional encoding can be used to add relative positional information to the time-series vector, enabling the network performance prediction model to better capture the temporal characteristics of the data.

[0125] For example, positional coding can be implemented using the classic sine and cosine positional coding model, with the specific coding calculation formulas shown in Formulas 4 and 5 below:

[0126]

[0127]

[0128] Where d represents the dimension of the position vector; 2i represents an even-dimensional sensor, encoded using the sin function; and 2i+1 represents an odd-dimensional sensor, encoded using the cosine function cos.

[0129] Furthermore, the input parameters, after entering the network, undergo processing through multiple self-attention layers and feedforward network layers. The attention mechanism extracts the correlations within the sequence, and the calculation of the attention weights at each time step requires the participation of all data within the sequence.

[0130] It's important to note that the attention mechanism can be understood as mapping the query Q, key K, and value V to the output, and then using a softmax layer to map the output to a range. Given a dimension d... k Q and K and d v In the case of dimension V, the calculation process of dot product attention can be shown in the following formula six:

[0131]

[0132] Furthermore, in network architecture design, residual network layer normalization can be performed on the data after each layer's computation to eliminate the information loss (network degradation) caused by increasing the number of layers.

[0133] In the network output layer of the network architecture, since the output result is numerical user network indicator data, in the implementation process of this application embodiment, the original softmax classification layer can be skipped, and the feedforward network can be used to directly calculate the result.

[0134] S403. Based on the first parameter and / or the second parameter corresponding to the target unit time period, determine the prediction network parameters corresponding to the second unit time period adjacent to the target unit time period.

[0135] Optional, such as Figure 6 As shown, the long-term learning module in this embodiment is mainly used to learn network metrics based on long-term data (i.e., network metrics data corresponding to the first unit time period included in each of multiple adjacent historical dates), such as network metrics data of the current hour (target unit time period) of each day in the previous month. Figure 6 The system uses network parameters, private network characteristic parameters, and time characteristics to learn the long-term characteristics of network indicator data and determine the first parameter corresponding to the network indicator data for the target unit time period.

[0136] The short-term learning module in this application embodiment is mainly used to learn based on short-term data (i.e., network indicator data corresponding to each unit time period in multiple unit time periods included within a preset duration adjacent to the target unit time period), such as the network indicator data of each hour in the previous 24 hours. Figure 6 The system uses network parameters, private network characteristic parameters, and time characteristics to learn the short-term characteristics of network indicator data and determine the second parameter corresponding to the network indicator data for the target unit time period.

[0137] Optionally, in one possible implementation, when the difference between the predicted network parameters corresponding to the second unit time period and the actual network parameters corresponding to the second unit time period is greater than a preset second threshold, the relevant parameters in the long-term learning module and the short-term learning module can be adjusted. The specific adjustment method can be combined with the relevant description in section S203 above, and will not be repeated here.

[0138] It should be noted that, in one possible implementation, the second threshold for the long-term learning module and the second threshold for the short-term pattern learning module can be the same size or different sizes. Specifically, the second threshold for the long-term learning module and the second threshold for the short-term pattern learning module can be determined and adjusted based on business experience and actual prediction results.

[0139] In this embodiment, a long-term learning module learns the long-term characteristics of historical data, and a short-term learning module learns the short-term characteristics of recent historical data, thereby achieving bidirectional prediction in both the long-term and short-term categories and further improving the prediction accuracy of the network performance prediction model.

[0140] In one possible implementation, the network performance prediction model includes: a long-term learning module, a short-term learning module, and a feature fusion module, such as... Figure 8 As shown, in the network performance prediction model training method provided in this application embodiment, the above-mentioned S403 includes S501:

[0141] S501. Input the first and second parameters corresponding to the target unit time period into the feature fusion module in the network performance prediction model to determine the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period.

[0142] It should be noted that, as described in sections S401 and S402 above, the first predicted value of the network parameters for the second time period can be obtained through the long-term learning module, based on the long-term patterns of the data corresponding to the target unit time period; similarly, the second predicted value of the network parameters for the second time period can be obtained through the short-term learning module, based on the short-term patterns of the data corresponding to the target unit time period. Therefore, to further improve the accuracy of the prediction results for the network parameters corresponding to the second time period, the predicted network parameters that are closest to the true values ​​can be further determined using the first and second predicted values.

[0143] Optionally, the network structure of the feature fusion module can be determined based on specific usage needs and business experience. For example, in one implementation of this application, the feature fusion module can be implemented using a deep neural network (DNN) architecture.

[0144] Specifically, such as Figure 6 As shown, the feature fusion module in this embodiment can take the inputs (first parameter and second parameter) of the above neural networks N1 and N2 as inputs, and fuse the long-term features learned in the first parameter with the short-term features learned in the second parameter to obtain a more accurate prediction result.

[0145] Optionally, when the difference between the predicted network parameters corresponding to the second unit time period and the actual network parameters corresponding to the second unit time period is greater than the preset third threshold, the relevant parameters in the feature fusion module can be adjusted. The specific adjustment method can be referred to the relevant description in section S203 above, and will not be repeated here.

[0146] It should be noted that, in one possible implementation, the first threshold, the second threshold, and the third threshold in the embodiments of this application can be the same size threshold or different sizes threshold. Specifically, the first threshold, the second threshold, and the third threshold in the embodiments of this application can be determined and adjusted based on business experience and information such as actual prediction results.

[0147] In this embodiment, the first parameter obtained by the long-term learning module and the second parameter obtained by the short-term learning module are fused through the feature fusion module. That is, the long-term features and short-term features of the data are fused to achieve a comprehensive prediction of network parameters based on the long-term features and short-term features of the data, and the prediction results are more accurate.

[0148] In one possible implementation, such as Figure 9 As shown in the embodiment of this application, the network performance prediction model training method further includes steps S601-S602:

[0149] S601. Obtain the target data corresponding to the third unit time period.

[0150] The third unit time period is the time period after the current time, and the target data is the network indicator data corresponding to multiple unit time periods that are the same as the third unit time period, which are included in each of the multiple historical dates before the current time.

[0151] Optionally, the target data for the third time period can be the target data for the user to be predicted (private network user).

[0152] Optionally, after the network performance prediction model is trained, it can also be used to predict the network performance of individual users, such as predicting the network performance of a user under a private network.

[0153] S602. Input the target data into the trained network performance prediction model to obtain the predicted network parameters corresponding to the third time unit.

[0154] Optional, such as Figure 4 As shown, after obtaining the trained network performance prediction model, the target data corresponding to the third time period can be input into the trained network performance prediction model to obtain the predicted network parameters corresponding to the third time period.

[0155] It should be noted that the specific principle of obtaining the predicted network parameters corresponding to the third unit time period based on the target data through the network performance prediction model can be found in the prediction principle described above during the training process of the network performance prediction model. This principle has already been described in the relevant sections above (such as those described above regarding...). Figure 6 The relevant descriptions are omitted here.

[0156] Optionally, after obtaining the predicted network parameters corresponding to the third unit time period, the predicted network parameters can be used as a reference for the allocation of resources in the private network, so as to allocate network resources more rationally and ensure the stability and service of the private network.

[0157] In this application, the solutions provided by the embodiments of this application are mainly described from a methodological perspective. To achieve the above functions, it includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments disclosed herein, the embodiments of this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed by hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0158] This application embodiment can divide a network performance prediction model training device into functional modules based on the above method example. For example, each function can be divided into its own functional modules, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware or as a software functional module. Optionally, the module division in this application embodiment is illustrative and only represents a logical functional division; other division methods may be used in actual implementation.

[0159] Figure 10 This is a schematic diagram of a network performance prediction model training device provided in an embodiment of this application. Figure 10 As shown, the network performance prediction model training device 100 is used to improve the prediction accuracy of the network performance prediction model, for example, for performing... Figure 2 A method for training a network performance prediction model is shown. The network performance prediction model training device 100 includes: an acquisition unit 1001, a determination unit 1002, and a processing unit 1003;

[0160] The acquisition unit 1001 is used to acquire a training dataset that includes multiple network metric data.

[0161] The network indicator data includes: network indicator data corresponding to the first unit time period of each historical date in multiple adjacent historical dates; network indicator data corresponding to each unit time period in multiple unit time periods within a preset duration preceding the target unit time period; the first unit time period is the same unit time period as the target unit time period in each historical date; the network indicator data includes: network parameters, private network characteristic parameters, and time characteristics; the network parameters include at least one of the following: user traffic, packet loss rate, service latency, and user rate; the private network characteristic parameters are used to indicate the network type; and the time characteristics are used to indicate whether the time corresponding to the network indicator data belongs to the preset date.

[0162] The determination unit 1002 is used to input multiple network indicator data into the network performance prediction model to determine the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period.

[0163] The processing unit 1003 is used to adjust the model parameters of the network performance prediction model based on the difference between the predicted network parameters corresponding to the second unit time period and the actual network parameters corresponding to the second unit time period, so as to obtain the trained network performance prediction model.

[0164] In one possible implementation, the network performance prediction model includes a private network feature learning module.

[0165] The determination unit 1002 is used to input the private network features corresponding to each network indicator data in the multiple network indicator data into the private network feature learning module in the network performance prediction model, and determine the private network feature parameters corresponding to each network indicator data. The private network features include at least one of the following: industry type, private network type and geographical location.

[0166] The determining unit 1002 is used to determine the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period based on the network parameters, private network characteristic parameters and time characteristics included in each network indicator data.

[0167] In one possible implementation, the network performance prediction model may also include at least one of the following: a long-term learning module and a short-term learning module.

[0168] The determining unit 1002 is used to input the network indicator data corresponding to the first unit time period included in each of the multiple adjacent historical dates into the long-term learning module in the network performance prediction model, and determine the first parameter corresponding to the network indicator data corresponding to the target unit time period.

[0169] And / or, determining unit 1002 is used to input the network indicator data corresponding to each of the multiple unit time periods included in the preset duration adjacent to the target unit time period into the short-term learning module in the network performance prediction model, and determine the second parameter corresponding to the network indicator data corresponding to the target unit time period.

[0170] The determining unit 1002 is used to determine the prediction network parameters corresponding to the second unit time period adjacent to the target unit time period based on the first parameter and / or the second parameter corresponding to the target unit time period.

[0171] In one possible implementation, the network performance prediction model includes: a long-term learning module, a short-term learning module, and a feature fusion module.

[0172] The determining unit 1002 is used to input the first parameter and the second parameter corresponding to the target unit time period into the feature fusion module in the network performance prediction model to determine the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period.

[0173] In one possible implementation, the acquisition unit 1001 is used to acquire the target data corresponding to the third unit time period, where the third unit time period is the time period after the current time, and the target data is the network indicator data corresponding to the multiple unit time periods that are the same as the third unit time period included in each of the multiple historical dates before the current time.

[0174] The determination unit 1002 is used to input the target data into the trained network performance prediction model to obtain the prediction network parameters corresponding to the third unit time period.

[0175] In the case where the functions of the integrated modules described above are implemented in hardware, this application provides a possible structural schematic diagram of the electronic device involved in the above embodiments. For example... Figure 11 As shown, an electronic device 110 is used to improve the prediction accuracy of a network performance prediction model, for example, for performing... Figure 2 This illustrates a method for training a network performance prediction model. The electronic device 110 includes a processor 1101, a memory 1102, and a bus 1103. The processor 1101 and the memory 1102 can be connected via the bus 1103.

[0176] Processor 1101 is the control center of the communication device. It can be a single processor or a collective term for multiple processing elements. For example, processor 1101 can be a general-purpose central processing unit (CPU) or other general-purpose processors. Among them, the general-purpose processor can be a microprocessor or any conventional processor.

[0177] As one embodiment, processor 1101 may include one or more CPUs, for example Figure 11 CPU 0 and CPU 1 are shown in the diagram.

[0178] The memory 1102 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), disk storage medium or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0179] As one possible implementation, the memory 1102 can exist independently of the processor 1101. The memory 1102 can be connected to the processor 1101 via the bus 1103 and is used to store instructions or program code. When the processor 1101 calls and executes the instructions or program code stored in the memory 1102, it can implement the network performance prediction model training method provided in this application embodiment.

[0180] In another possible implementation, the memory 1102 can also be integrated with the processor 1101.

[0181] Bus 1103 can be an industry standard architecture (ISA) bus, a peripheral component interconnect (PCI) bus, or an extended industry standard architecture (EISA) bus. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 11 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0182] It should be pointed out that, Figure 11 The structure shown does not constitute a limitation on the electronic device 110. Except... Figure 11 In addition to the components shown, the electronic device 110 may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0183] As an example, combined Figure 10 The functions implemented by the acquisition unit 1001, determination unit 1002, and processing unit 1003 in the network performance prediction model training device 100 are the same as those implemented in other devices. Figure 11 The processor 1101 in it has the same function.

[0184] Optional, such as Figure 11 As shown, the electronic device 110 provided in this application embodiment may also include a communication interface 1104.

[0185] Communication interface 1104 is used to connect with other devices via a communication network. This communication network can be Ethernet, a wireless access network, a wireless local area network (WLAN), etc. Communication interface 1104 may include a receiving unit for receiving data and a transmitting unit for transmitting data.

[0186] In one possible implementation, the communication interface in the electronic device provided in this application embodiment can also be integrated into the processor.

[0187] Through the above description of the embodiments, those skilled in the art will clearly understand that, for the sake of convenience and brevity, only the division of the above functional units is used as an example. In practical applications, the above functions can be assigned to different functional units as needed, that is, the internal structure of the device can be divided into different functional units to complete all or part of the functions described above. The specific working process of the system, device, and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0188] This application also provides a computer-readable storage medium storing instructions. When a computer executes these instructions, the computer performs each step of the method flow shown in the above-described method embodiments.

[0189] Embodiments of this application provide a computer program product containing instructions that, when executed on a computer, cause the computer to perform a network performance prediction model training method as described in the above method embodiments.

[0190] The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), registers, hard disks, optical fibers, compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof, or any other form of computer-readable storage medium in the art. An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium may also be a component of the processor. The processor and the storage medium may reside in an application-specific integrated circuit (ASIC). In the embodiments of this application, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0191] Since the electronic devices, computer-readable storage media, and computer program products in the embodiments of this application can be applied to the above methods, the technical effects they can achieve can also be referred to the above method embodiments. The embodiments of this application will not be repeated here.

[0192] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be covered within the scope of protection of this application.

Claims

1. A method for training a network performance prediction model, characterized in that, The method includes: A training dataset comprising multiple network metric data is obtained. These multiple network metric data include: network metric data corresponding to a first unit time period for each historical date among multiple adjacent historical dates; and network metric data corresponding to each unit time period within a preset duration preceding the target unit time period. The first unit time period is a unit time period in each historical date that is identical to the target unit time period. The network metric data includes: network parameters, private network characteristic parameters, and time features. The network parameters include: user traffic, packet loss rate, service latency, and user rate. The private network characteristic parameters indicate the network type, and the time features indicate whether the time corresponding to the network metric data belongs to a preset date. The multiple network indicator data are respectively input into the network performance prediction model to determine the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period. Based on the difference between the predicted network parameters corresponding to the second unit time period and the actual network parameters corresponding to the second unit time period, the model parameters of the network performance prediction model are adjusted to obtain the trained network performance prediction model. The network performance prediction model includes a private network feature learning module; the method further includes: The private network features corresponding to each of the multiple network indicator data are input into the private network feature learning module in the network performance prediction model to determine the private network feature parameters corresponding to each network indicator data. The private network features include: industry type, private network type and geographical location.

2. The method according to claim 1, characterized in that, The network performance prediction model further includes at least one of the following: a long-term learning module and a short-term learning module; The step of inputting the multiple network indicator data into the network performance prediction model to determine the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period includes: The network indicator data corresponding to the first unit time period included in each of the multiple adjacent historical dates is input into the long-term learning module in the network performance prediction model to determine the first parameter corresponding to the network indicator data corresponding to the target unit time period. And / or, input the network indicator data corresponding to each of the multiple unit time periods included in the preset duration adjacent to the target unit time period into the short-term learning module in the network performance prediction model to determine the second parameter corresponding to the network indicator data corresponding to the target unit time period; Based on the first parameter and / or the second parameter corresponding to the target unit time period, determine the prediction network parameters corresponding to the second unit time period adjacent to the target unit time period.

3. The method according to claim 2, characterized in that, The network performance prediction model includes: the long-term learning module, the short-term learning module, and the feature fusion module; The step of determining the prediction network parameters corresponding to the second unit time period adjacent to the target unit time period based on the first parameter and / or the second parameter corresponding to the target unit time period includes: The first parameter and the second parameter corresponding to the target unit time period are input into the feature fusion module in the network performance prediction model to determine the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period.

4. The method according to claim 1, characterized in that, The method further includes: Obtain the target data corresponding to the third unit time period, wherein the third unit time period is the time period after the current time, and the target data is the network indicator data corresponding to multiple unit time periods that are the same as the third unit time period included in each of multiple historical dates before the current time. The target data is input into the trained network performance prediction model to obtain the predicted network parameters corresponding to the third time unit.

5. A network performance prediction model training device, characterized in that, The network performance prediction model training device includes: an acquisition unit, a determination unit, and a processing unit; The acquisition unit is used to acquire a training dataset including multiple network indicator data. The multiple network indicator data includes: network indicator data corresponding to a first unit time period for each historical date among multiple adjacent historical dates; and network indicator data corresponding to each unit time period among multiple unit time periods within a preset duration preceding the target unit time period. The first unit time period is a unit time period in each historical date that is the same as the target unit time period. The network indicator data includes: network parameters, private network characteristic parameters, and time characteristics. The network parameters include: user traffic, packet loss rate, service latency, and user rate. The private network characteristic parameters are used to indicate the network type, and the time characteristics are used to indicate whether the time corresponding to the network indicator data belongs to a preset date. The determining unit is used to input the multiple network indicator data into the network performance prediction model respectively, and determine the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period. The processing unit is used to adjust the model parameters of the network performance prediction model based on the difference between the predicted network parameters corresponding to the second unit time period and the actual network parameters corresponding to the second unit time period, so as to obtain the trained network performance prediction model. The network performance prediction model includes a private network feature learning module; The determining unit is used to input the private network feature corresponding to each of the multiple network indicator data into the private network feature learning module in the network performance prediction model, and determine the private network feature parameter corresponding to each network indicator data. The private network feature includes at least one of the following: industry type, private network type and geographical location.

6. The network performance prediction model training device according to claim 5, characterized in that, The network performance prediction model further includes at least one of the following: a long-term learning module and a short-term learning module; The determining unit is used to input the network indicator data corresponding to the first unit time period included in each of the multiple adjacent historical dates into the long-term learning module in the network performance prediction model, and determine the first parameter corresponding to the network indicator data corresponding to the target unit time period. And / or, the determining unit is used to input the network indicator data corresponding to each of the multiple unit time periods included in the preset duration adjacent to the target unit time period into the short-term learning module in the network performance prediction model, respectively, to determine the second parameter corresponding to the network indicator data corresponding to the target unit time period; The determining unit is configured to determine the prediction network parameters corresponding to the second unit time period adjacent to the target unit time period based on the first parameter and / or the second parameter corresponding to the target unit time period.

7. The network performance prediction model training device according to claim 6, characterized in that, The network performance prediction model includes: the long-term learning module, the short-term learning module, and the feature fusion module; The determining unit is used to input the first parameter and the second parameter corresponding to the target unit time period into the feature fusion module in the network performance prediction model to determine the predicted network parameters corresponding to the second unit time period adjacent to the target unit time period.

8. The network performance prediction model training device according to claim 5, characterized in that, The acquisition unit is used to acquire target data corresponding to a third unit time period, wherein the third unit time period is the time period after the current time, and the target data is network indicator data corresponding to multiple unit time periods that are the same as the third unit time period included in each of multiple historical dates before the current time. The determining unit is used to input the target data into the trained network performance prediction model to obtain the predicted network parameters corresponding to the third unit time period.

9. An electronic device, characterized in that, include: A processor and a memory; wherein the memory is used to store one or more programs, the one or more programs including computer execution instructions, and when the electronic device is running, the processor executes the computer execution instructions stored in the memory to cause the electronic device to execute a network performance prediction model training method according to any one of claims 1-4.

10. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computer, cause the computer to perform a network performance prediction model training method according to any one of claims 1-4.

Images