Network index prediction method and device
The method improves network indicator prediction accuracy by employing a time-series model for time-influenced periods and scene-specific models for lesser-influenced scenarios, ensuring accurate network indicator predictions.
Patent Information
- Application Number
- CN202510357896.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-25
- Publication Date
- 2025-07-15
AI Technical Summary
In the prior art, network index prediction accuracy is poor, and it is difficult to meet the needs of complex and changeable practical application scenarios.
Using a method combining a time series prediction model and a specified scenario index model, we use a method to obtain network index prediction needs and historical data, judge the degree of fluctuation in network indexes, and select an appropriate prediction model to improve accuracy.
When the network index data is greatly affected by the time law, the results of the time series prediction model are output, and when the impact is small, the results of the scene index model are output, thereby improving the accuracy of network index prediction.
Smart Images
Figure CN120321132A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of artificial intelligence, and particularly to a method and device for predicting network metrics. Background Art
[0002] In the field of communication technology, with the growth of the number of mobile users and communication devices, the service requirements of communication networks are becoming increasingly complex. In actual application scenarios, network metrics are often affected by various factors, and the network metrics in different time periods and different scenarios may be affected by different factors, which makes the prediction accuracy of network metrics poor and difficult to meet the accuracy requirements for network metric prediction in actual application scenarios.
[0003] How to improve the prediction accuracy of network metrics is the technical problem to be solved by this application. Summary of the Invention
[0004] The purpose of the embodiments of this application is to provide a method and device for predicting network metrics to improve the prediction accuracy of network metrics.
[0005] In a first aspect, a method for predicting network metrics is provided, including: Obtain the network metric prediction requirement and the historical network metric data associated with the network metric prediction requirement, where the network metric prediction requirement includes a target prediction time period and a target prediction scenario; Input the historical network metric data and the target prediction time period into a time series prediction model to obtain a first prediction result predicted by the time series prediction model for the target prediction time period; If the fluctuation degree of the network metrics in the first prediction result is greater than a first preset fluctuation degree, then determine the first prediction result as the network metric prediction result for the target prediction time period; If the fluctuation degree of the network metrics in the first prediction result is less than a second preset fluctuation degree, then input the historical network metric data and the target prediction time period into a specified scenario metric model matched with the target prediction scenario, and determine the second prediction result output by the specified scenario metric model as the network metric prediction result for the target prediction time period, where the second preset fluctuation degree is less than or equal to the first preset fluctuation degree.
[0006] In a second aspect, a device for predicting network metrics is provided, including: An acquisition module, which acquires the network metric prediction requirement and the historical network metric data associated with the network metric prediction requirement, where the network metric prediction requirement includes a target prediction time period and a target prediction scenario; An input module that inputs the historical network metric data and the target prediction period into a time series prediction model to obtain a first prediction result predicted by the time series prediction model for the target prediction period; A first determination module that, if the degree of network metric fluctuation of the first prediction result is greater than a first preset fluctuation degree, determines the first prediction result as the network metric prediction result for the target prediction period; A second determination module that, if the degree of network metric fluctuation of the first prediction result is less than a second preset fluctuation degree, inputs the historical network metric data and the target prediction period into a specified scenario metric model matching the target prediction scenario, and determines the second prediction result output by the specified scenario metric model as the network metric prediction result for the target prediction period, where the second preset fluctuation degree is less than or equal to the first preset fluctuation degree.
[0007] In a third aspect, an electronic device is provided. The electronic device includes a processor, a memory, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, the steps of the method in the first aspect are implemented.
[0008] In a fourth aspect, a computer-readable storage medium is provided. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps of the method in the first aspect are implemented.
[0009] In a fifth aspect, a computer program product is provided. The computer program product includes a non-transitory computer-readable storage medium storing a computer program. The computer program is operable to cause a computer to execute some or all of the steps of the method in the first aspect.
[0010] In an embodiment of the present application, first, obtain a network metric prediction requirement and historical network metric data associated with the network metric prediction requirement. The network metric prediction requirement includes a target prediction period and a target prediction scenario, providing a data basis for subsequent predictions. Then, input the historical network metric data and the target prediction period into a time series prediction model to obtain a first prediction result predicted by the time series prediction model for the target prediction period. The time series prediction model can highlight data characteristics that change over time, so as to reflect whether the target prediction period is affected by time patterns through the first prediction result. If the fluctuation degree of the network metric in the first prediction result is greater than a first preset fluctuation degree, it indicates that the target prediction period is greatly affected by time patterns, and then the first prediction result is determined as the network metric prediction result for the target prediction period. If the fluctuation degree of the network metric in the first prediction result is less than a second preset fluctuation degree, it indicates that the target prediction period is less affected by time patterns, and then the historical network metric data and the target prediction period are input into a specified scenario metric model matching the target prediction scenario, and the second prediction result output by the specified scenario metric model is determined as the network metric prediction result for the target prediction period, where the second preset fluctuation degree is less than or equal to the first preset fluctuation degree. When the target prediction period is less affected by time patterns, the historical network metric data and the target prediction period are further input into the specified scenario metric model according to the target prediction scenario, so as to perform prediction using an appropriate scenario metric model for the target prediction scenario, making the second prediction result fit the target prediction scenario, thereby improving the accuracy of the finally output network metric prediction result. Through the solution provided by the embodiment of the present application, the first prediction result predicted based on the time series prediction model can be output when the network metric data is greatly affected by time patterns, and the second prediction result predicted based on the scenario metric model can be output when the network metric data is less affected by time patterns, so that the output network metric prediction result fits the real data, effectively improving the accuracy of network metric prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0011] The drawings described herein are used to provide a further understanding of the present application and form a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings: Figure 1 is one of the flow diagrams of a network metric prediction method according to an embodiment of the present application; Figure 2 is another flow diagram of a network metric prediction method according to an embodiment of the present application; Figure 3 is yet another flow diagram of a network metric prediction method according to an embodiment of the present application; Figure 4It is the fourth flowchart diagram of a network metric prediction method according to an embodiment of the present application; Figure 5 It is the fifth flowchart diagram of a network metric prediction method according to an embodiment of the present application; Figure 6 It is the sixth flowchart diagram of a network metric prediction method according to an embodiment of the present application; Figure 7 It is the seventh flowchart diagram of a network metric prediction method according to an embodiment of the present application; Figure 8 It is the structural diagram of a network metric prediction device according to an embodiment of the present application. Detailed implementation manners
[0012] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application. The accompanying drawing numbers in the present application are only used to distinguish each step in the solution and do not limit the execution order of each step. The specific execution order shall be subject to the description in the specification.
[0013] In the field of communication technology, the rapid growth of wireless services has put forward higher requirements for mobile operators in aspects such as reasonably allocating base station resources and improving user experience. Reliable network metric prediction is beneficial for network operation and maintenance personnel to understand the trend change of network quality as early as possible, make optimization adjustments and resource deployments to the network as early as possible, so as to ensure the user experience. In addition, based on reliable network metric prediction, the operating state of the base station can be flexibly adjusted to achieve the effect of energy conservation and emission reduction.
[0014] In the application scenario of network metric prediction, there may be defects such as a large prediction time range and the inability to accurately predict metrics at a daily granularity or a finer granularity. Moreover, in actual applications, holidays often have a greater impact on network metrics, and a prediction scheme that ignores the holiday effect cannot effectively achieve network metric prediction. In addition, the network metrics can also be predicted by pre-training a machine learning model. However, the pre-trained model is often applicable to a specific scenario, and the network metrics in the actual application scenario are flexible and changeable. It is difficult for the model pre-trained in the specified scenario to meet the accuracy requirements for network metric prediction in the actual application scenario.
[0015] To solve the problems existing in the related technologies, the embodiments of the present application provide a network metric prediction method, as Figure 1 shown, including: S11: Obtain the network metric prediction requirements and the historical network metric data associated with the network metric prediction requirements, where the network metric prediction requirements include the target prediction period and the target prediction scenario.
[0016] In this step, obtaining the network metric prediction requirements and the historical network metric data provides a reliable data basis for subsequent network metric prediction.
[0017] Among them, the network metric prediction requirements include the target prediction period and the target prediction scenario. The target prediction period represents which period of network metrics needs to be predicted, and the target prediction period can be continuous or multiple segments. The target prediction scenario can be related to the inherent attributes of the network metrics or the environment where they are located. For example, the prediction scenarios can be divided into single network metric prediction, multiple network metric prediction, single network metric prediction for multiple base stations, multiple network metric prediction for multiple base stations, etc. In addition to the number of network metrics and the number of base stations, attribute dimensions such as the environment where the base stations are located, the types and quantities of network users involved in the network metrics can all be used as the attribute dimensions of the prediction scenario.
[0018] The above historical network metric data is associated with the network metric prediction requirements.
[0019] Optionally, in terms of the inherent attributes of the network metrics, the historical network metric data includes network metric data of the same type as the network metric prediction requirements. That is, data of the network metric to be predicted is obtained for the historical period of that network metric.
[0020] Optionally, in terms of the environmental attributes of the network metrics, the historical network metric data and the network metric prediction requirements are in the same or similar network environment. For example, the historical network metric data is the network metric data of a specified base station, and the network metric prediction requirements indicate that the network metrics of the above-specified base station need to be predicted.
[0021] Optionally, in terms of the time dimension, the historical period in which the historical network metric data is located is the historical period before the target prediction period.
[0022] In some application scenarios, the actually obtained network metric prediction requirements and historical network metric data may have the defect of poor data quality. Optionally, performing preprocessing on the obtained network metric prediction requirements and historical network metric data can effectively improve the data quality, and thus improve the accuracy of the prediction results in subsequent steps.
[0023] Specifically, real operation and maintenance data usually has a large amount of sudden noise. If this noise is not extracted and stripped, it will severely limit the fitting performance of the time series prediction model. In this solution, the data can be preprocessed through methods such as missing outlier processing, time alignment, and data segmentation preprocessing to improve the effectiveness of historical network metric data.
[0024] Among them, missing outlier processing can specifically include filling missing values, which can be achieved by interpolation. Among them, interpolation can be implemented in ways such as nearest neighbor interpolation and mean interpolation. In addition, extreme outliers can also be identified and processed, specifically by using the box plot method to identify and correct or delete outliers.
[0025] Time alignment processing is used to ensure that the values in the historical network metric data are sorted according to the correct time step.
[0026] Optionally, for the preprocessed historical network metric data, the model training set and test set can be divided through data segmentation processing for pre-training and evaluation of the time series prediction model to ensure that the prediction ability of the pre-trained time series prediction model meets the required requirements. Optionally, 80% of the historical network metric data is used as the training sample set, and the remaining data is used as the test sample set. During the model application process, the model can be iteratively trained after obtaining the historical network metric data to update the model parameters so that the time series prediction model meets the prediction requirements of the latest network metric data.
[0027] S12: Input the historical network metric data and the target prediction period into the time series prediction model to obtain the first prediction result predicted by the time series prediction model for the target prediction period.
[0028] In this step, the historical network metric data and the target prediction period are input into the time series prediction model, thereby instructing the time series prediction model to predict the network metric data within the target prediction period based on the change characteristics of the historical network metric data.
[0029] Among them, the time series prediction model can be pre-trained based on the network metric data of the historical period.
[0030] In this step, the historical network metric data and the target prediction period can be directly input into the time series prediction model for prediction. Before performing this input, a special date judgment is first made on the target prediction period. If the judgment result meets the input requirements, the historical network metric data and the target prediction period are input into the time series prediction model for prediction.
[0031] Optionally, a list of annual alignments of holidays and special dates is preset, which is used to indicate which dates belong to holidays and special dates. Before inputting historical network metric data and a target prediction period into a time series prediction model, first determine whether the target prediction period belongs to a holiday or special date based on this list. If it does, input the historical network metric data and the target prediction period into the time series prediction model for prediction. If it does not, the historical network metric data and the target prediction period can be directly input into a specified scenario metric model that matches the target prediction scenario for prediction.
[0032] Through the above list of annual alignments of holidays and special dates, it is possible to effectively determine whether the target prediction period belongs to a holiday or special date, thereby preliminarily predicting the degree to which the target prediction period is affected by the holiday effect. Furthermore, in the case of a large degree of influence by the holiday effect, input it into the time series prediction model for prediction, effectively improving the prediction effectiveness and overall execution efficiency.
[0033] In practical applications, the target prediction period can be, for example, one day, one week, one month, one year, etc. predicted backward based on the period where the historical network metric data is located. The list of annual alignments of holidays and special dates can be presented in the form of a dataset, which is used to clearly mark the dates of special events such as holidays and promotional activities, so as to incorporate the holiday effect into the time series prediction model. Optionally, the input time format is uniformly of the datetime type to improve the processing efficiency of time-dimensional data.
[0034] Taking the Chinese Lunar New Year as an example, the starting dates of the Spring Festival from 2022 to 2024 (in this example, the New Year's Eve is used as the starting date) are January 31st, January 21st, and February 9th respectively. The definitions of the Chinese Lunar New Year from 2022 to 2024 are starting from January 31st, 2022, January 21st, 2023, and February 9th, 2024 respectively, and the Spring Festival lasts for 7 days, thus achieving the time alignment of the Chinese Lunar New Year. Another example is that this special date period of summer vacation can be postponed by 60 days starting from July 1st every year. In practical applications, the specific date definitions and data storage methods of the list of annual alignments of holidays and special dates can be flexibly set according to requirements.
[0035] Optionally, the time series prediction model in this solution can, for example, select the Prophet (a procedure for forecasting time series data based on an additive model, additive time series prediction algorithm) model. It is a long-term prediction model based on time, which can predict according to the data time series law, and can especially highlight the influence of special dates such as holidays on the data, so that the prediction results conform to the network metric prediction results of special dates in the real scenario.
[0036] After obtaining the first prediction result, subsequent steps are executed according to the degree of fluctuation of the network metrics of the first prediction result. Specifically, the degree of fluctuation of the network metrics can characterize the impact of special dates on the network metrics. Generally speaking, the network metrics on non-special dates are relatively stable and do not have significant fluctuations. While the network metrics on special dates are affected by the holiday effect and often show large fluctuations. Therefore, based on the degree of fluctuation of the network metrics, the impact of the holiday effect on the first prediction result can be effectively evaluated.
[0037] S13: If the degree of fluctuation of the network metrics of the first prediction result is greater than the first preset degree of fluctuation, then determine the first prediction result as the network metrics prediction result for the target prediction period.
[0038] If the degree of fluctuation of the network metrics of the first prediction result is greater than the first preset degree of fluctuation, it indicates that the first prediction result is greatly affected by the holiday effect, and other factors other than the holiday effect have a relatively small impact on the network metrics. Based on this, the first prediction result is used as the network metrics prediction result for output.
[0039] S14: If the degree of fluctuation of the network metrics of the first prediction result is less than the second preset degree of fluctuation, then input the historical network metrics data and the target prediction period into the specified scenario metrics model matching the target prediction scenario, and determine the second prediction result output by the specified scenario metrics model as the network metrics prediction result for the target prediction period, where the second preset degree of fluctuation is less than or equal to the first preset degree of fluctuation.
[0040] If the degree of fluctuation of the network metrics of the first prediction result is less than the second preset degree of fluctuation, it indicates that the first prediction result is less affected by the holiday effect, and other factors such as scenarios other than the holiday effect often have a greater impact on the network metrics. Based on this, the historical network metrics data and the target prediction period are input into the specified scenario metrics model matching the target prediction scenario, so as to perform network metrics prediction for the application scenario and obtain the second prediction result.
[0041] In this step, for the prediction period with little impact from holidays and special dates, according to the actual scenario requirements for prediction, a specified scenario metrics model matching the scenario is selected. The scenario models for actual application can be flexibly set according to actual needs. For example, for a single metric prediction scenario, the LightGBM prediction model can be used; for a multi-metric correlation scenario, the LSTM prediction model can be used; for a massive network element spatio-temporal scenario, the DeepAR prediction model can be used.
[0042] The above-specified scenario metric model is pre-trained based on the historical network metric data of the matching prediction scenario, that is, each scenario metric model is pre-trained to reach the optimal parameters, so as to select the matching model according to the scenario for prediction in actual applications.
[0043] For example, the correspondence between the target prediction scenario and the corresponding network metric data features is as follows:
[0044] This solution jointly applies a time series prediction model and a specified scenario metric model. The time series prediction model is used to achieve long-term prediction, and the specified scenario metric model is integrated to improve the prediction performance and accuracy. Among them, the degree to which the first prediction result of the time series prediction model is affected by the holiday effect is judged, so as to flexibly select whether to output the first prediction result or to make further predictions for the specified scenario, and the network metric prediction result can be flexibly determined according to the actual network metric data features and the actual prediction scenario, meeting the complex and changeable network metric data prediction requirements.
[0045] In the embodiment of the present application, first, obtain the network metric prediction requirements and the historical network metric data associated with the network metric prediction requirements. The network metric prediction requirements include the target prediction period and the target prediction scenario, providing a data basis for subsequent predictions. Then, input the historical network metric data and the target prediction period into the time series prediction model to obtain the first prediction result predicted by the time series prediction model for the target prediction period. The time series prediction model can highlight the data characteristics that change over time, so as to reflect whether the target prediction period is affected by time rules through the first prediction result. If the fluctuation degree of the network metric of the first prediction result is greater than the first preset fluctuation degree, it indicates that the target prediction period is greatly affected by time rules, and then the first prediction result is determined as the network metric prediction result of the target prediction period. If the fluctuation degree of the network metric of the first prediction result is less than the second preset fluctuation degree, it indicates that the target prediction period is less affected by time rules, and then the historical network metric data and the target prediction period are input into the specified scenario metric model matching the target prediction scenario, and the second prediction result output by the specified scenario metric model is determined as the network metric prediction result of the target prediction period, where the second preset fluctuation degree is less than or equal to the first preset fluctuation degree. When the target prediction period is less affected by time rules, according to the target prediction scenario, the historical network metric data and the target prediction period are further input into the specified scenario metric model, so as to perform predictions using an appropriate scenario metric model for the target prediction scenario, making the second prediction result fit the target prediction scenario, thereby improving the accuracy of the finally output network metric prediction result. Through the solution provided by the embodiment of the present application, when the network metric data is greatly affected by time rules, the first prediction result predicted by the time series prediction model can be output, and when the network metric data is less affected by time rules, the second prediction result predicted by the scenario metric model can be output, so that the output network metric prediction result fits the real data and effectively improves the accuracy of network metric prediction.
[0046] Through the solution provided by the embodiment of the present application, the network metric prediction can be efficiently realized, and the obtained network metric prediction result plays a crucial role in the actual application scenario. For example, it can effectively help the optimization personnel understand the network development trend, guide the network optimization personnel to discover potential future hazards, and perform preventive maintenance. For instance, in the scenario of dealing with the network element capacity problem, knowing in advance the trend changes of the network element traffic and users is beneficial for achieving precise load balancing and carrier scheduling for high-load network elements, deploying scheduling resources in advance, which is of great significance and practical value to the network development.
[0047] Based on the solution provided by the above embodiment, optionally, as Figure 2As shown, after the above step S12, that is, after inputting the historical network metric data and the target prediction period into the time series prediction model to obtain the first prediction result predicted by the time series prediction model for the target prediction period, the following steps are further included: S21: Determine the network metric fluctuation coefficient of the first prediction result based on the first fluctuation test data and at least one second fluctuation test data, where the first fluctuation test data includes the network metric prediction data of the first fluctuation test period in the first prediction result, and the second fluctuation test data includes the historical network metric data of the historical same period of the first fluctuation test period.
[0048] In this step, the network metric fluctuation degree of the first prediction result is evaluated based on the historical network metric data of the historical same period.
[0049] For example, through the historical network metric data of the same holiday in the previous year and the previous month, calculate the impact of the holiday on the network traffic volume on the day of the holiday, so as to obtain the sensitivity coefficient, which is used to characterize the network metric fluctuation degree.
[0050] For instance, the Dragon Boat Festival in 2023 is from June 22nd to 24th, which are Thursday, Friday, and Saturday respectively. The traffic volume on June 22nd (Thursday) is represented as a, and the daily traffic volumes of the four Thursdays, namely June 15th, June 8th, June 1st, and May 25th, are represented as b1, b2, b3, and b4 respectively. In this step, the mathematical statistical method can be flexibly selected, and based on statistical characteristic values such as the average, median, and mode, the fluctuation degree of a can be evaluated based on b1, b2, b3, and b4.
[0051] S22: If the network metric fluctuation coefficient of the first prediction result is greater than the preset network metric fluctuation threshold, it is determined that the network metric fluctuation degree of the first prediction result is greater than the first preset fluctuation degree.
[0052] In this step, the preset network metric fluctuation threshold can be flexibly set according to the actual application scenario. This preset network metric fluctuation threshold can be statistically obtained based on the network metric fluctuation situation in the historical period, and is used to distinguish the data fluctuation degree affected by the holiday effect and the data fluctuation degree not affected by the holiday effect.
[0053] If the network metric fluctuation coefficient of the first prediction result is greater than the above preset network metric fluctuation threshold, it indicates that the first prediction result is greatly affected by the holiday effect. Furthermore, the first prediction result can be output as the network metric prediction result for the target prediction period.
[0054] Through the solution provided by the embodiments of the present application, based on the historical network metric data in the same historical period, the network metric fluctuation degree of the first prediction result can be objectively and effectively evaluated, so as to effectively evaluate the influence degree of the first prediction result affected by the holiday effect, and then determine whether to output the first prediction result as the network metric prediction result for the target prediction period, thereby improving the effectiveness of the output network metric prediction result.
[0055] Based on the solution provided by the above embodiment, optionally, the number of the second fluctuation test data is multiple; Among them, as Figure 3 shown, in the above step S21, determining the network metric fluctuation coefficient of the first prediction result according to the first fluctuation test data and at least one second fluctuation test data includes: S31: Determining the ratio of the first fluctuation test data to the average value of the test data as the network metric fluctuation coefficient of the first prediction result, where the average value of the test data includes the average values of multiple second fluctuation test data.
[0056] Based on the examples given in the above embodiment, taking b1, b2, b3, and b4 as the second fluctuation test data, and recording the average value of b1, b2, b3, and b4 as the average value of the test data. Determining the ratio of the first fluctuation test data a to the average value of the test data as the network metric fluctuation coefficient A of the first prediction result, which can be expressed as: A = a / [(b1 + b2 + b3 + b4) / 4].
[0057] The solution provided by the embodiments of the present application constructs a Prophet prediction model based on long-term historical network metric data, effectively covering index trends, seasonality, and holiday effects, and predicting traffic metrics in the future long cycle. By setting the sensitivity coefficient, the prediction result can be compared with the sensitivity system to perform stability analysis on the prediction result.
[0058] In this solution, the historical network metric data is used to perform stationarity analysis on the metrics of holidays and special dates, that is, to calculate the fluctuation of the business volume on the holiday day relative to the daily business volume, and set the sensitivity threshold θ for the influence of holidays and special dates on the metrics. Optionally, the threshold can be set to 40%, and different regions can be specifically analyzed and determined according to the network situation. If the network metric fluctuation coefficient of the first prediction result is greater than θ, it is a sensitive type, otherwise it is not sensitive. When the sensitivity value is greater than the threshold θ, it indicates that holidays and special dates have a greater impact on the metric fluctuation, and the final prediction result is the Prophet prediction result. When the sensitivity value is less than the threshold θ, it indicates that holidays and special dates have an insignificant impact on the metric fluctuation. To improve the accuracy of the prediction value, it is necessary to further perform scenario matching prediction according to the requirements of the prediction scenario.
[0059] Based on the solution provided by the above embodiment, optionally, asFigure 4 As shown in Figure 4 , in the above step S14, inputting the historical network metric data and the target prediction period into the specified scenario metric model matching the target prediction scenario, and determining the second prediction result output by the specified scenario metric model as the network metric prediction result for the target prediction period includes: S41: If the target prediction scenario is a scenario for predicting a single network metric, input the historical network metric data and the target prediction period into a Light Gradient Boosting Machine (LightGBM) model, and determine the second prediction result output by the LightGBM model as the network metric prediction result for the target prediction period.
[0060] In the embodiments of the present application, the Light Gradient Boosting Machine (LightGBM) algorithm is used to perform predictions on single metrics. It can effectively improve the prediction accuracy for single metrics, making the second prediction result output highlight the characteristics of the single network metric in the historical network metric data.
[0061] Based on the solution provided in the above embodiments, optionally, as Figure 5 As shown in Figure 5 , in the above step S14, inputting the historical network metric data and the target prediction period into the specified scenario metric model matching the target prediction scenario, and determining the second prediction result output by the specified scenario metric model as the network metric prediction result for the target prediction period includes: S51: If the target prediction scenario is a scenario for predicting multiple network metrics, input the historical network metric data and the target prediction period into a Long Short-Term Memory (LSTM) recurrent neural network model, and determine the second prediction result output by the LSTM model as the network metric prediction result for the target prediction period.
[0062] In the embodiments of the present application, the Long Short-Term Memory (LSTM) recurrent neural network is used to perform predictions on multiple network metrics. It can effectively handle important events and long-term dependencies in time series, and is good at achieving effective predictions for multiple network metrics, making the second prediction result output highlight the characteristics of the multiple network metrics in the historical network metric data.
[0063] Based on the solution provided in the above embodiments, optionally, as Figure 6 As shown in Figure 6 , in the above step S14, inputting the historical network metric data and the target prediction period into the specified scenario metric model matching the target prediction scenario, and determining the second prediction result output by the specified scenario metric model as the network metric prediction result for the target prediction period includes: S61: If the target prediction scenario is a scenario of predicting multiple network metrics of multiple network elements respectively, input the historical network metric data and the target prediction period into the autoregressive recurrent network DeepAR model, and determine the second prediction result output by the DeepAR model as the network metric prediction result for the target prediction period.
[0064] In the embodiments of the present application, the DeepAR (Probabilistic Forecasting with Autoregressive Recurrent Networks) autoregressive recurrent network model is used to perform predictions on multiple network metrics of multiple network elements. It is suitable for processing multivariate time series data, can achieve probabilistic forecasting, and has the advantage of high efficiency.
[0065] In one example, combining the above three machine learning models of LightGBM, LSTM, and DeepAR, automatically select the optimal model to output the prediction result according to the long-term prediction stability analysis result and the prediction demand scenario, which can meet the diverse actual application requirements.
[0066] Figure 7 The flowchart of this example is shown.
[0067] First, obtain the network metric prediction demand and the historical network metric data associated with the network metric prediction demand. Among them, the network metric prediction demand includes the target prediction period (i.e., the prediction time period) and the target prediction scenario. Judge whether the above target prediction period contains holidays and special dates according to the pre-set annual alignment list of holidays and special dates. If it does not contain, select a matching model for scenario prediction according to the target prediction scenario. If it contains, input the historical network metric data and the target prediction period into the time series prediction model (i.e., the Prophet long-term prediction model) to obtain the first prediction result (i.e., the Prophet prediction value) predicted by the time series prediction model for the target prediction period.
[0068] Subsequently, judge whether the index fluctuation sensitivity coefficient is greater than the threshold θ according to the first prediction result. If it is greater, it indicates that the network metric fluctuation degree of the first prediction result is greater than the first preset fluctuation degree, and then determine the first prediction result as the network metric prediction result for the target prediction period.
[0069] If it is not greater than the threshold θ, it indicates that the fluctuation degree of the network metric of the first prediction result is less than the second preset fluctuation degree. Then, the historical network metric data and the target prediction period are input into the specified scenario metric model matching the target prediction scenario, and the second prediction result output by the specified scenario metric model is determined as the network metric prediction result for the target prediction period. In this example, the first preset fluctuation degree and the second preset fluctuation degree are equal, both represented by the above threshold θ.
[0070] Specifically, if the target prediction scenario is a single-metric prediction scenario, the LightGBM model is selected to perform the prediction. If the target prediction scenario is a multi-metric correlation scenario, the LSTM model is selected to perform the prediction. If the target prediction scenario is a massive network element spatio-temporal scenario, the DeepAR model is selected to perform the prediction. Then, the second prediction result output by the scenario model is used as the network metric prediction result for the target prediction period.
[0071] The solution provided by the embodiments of the present application first predicts based on long-term historical network metric data through the Prophet model, and analyzes the sensitivity to holidays according to the prediction results. Subsequently, according to the predicted scenario requirements, combined with three machine learning models, namely LightGBM, LSTM, and DeepAR, deep prediction is performed, which is applicable to single-metric, multi-metric, and massive network element spatio-temporal prediction, and can meet diverse actual application requirements, and is used to guide the optimization and adjustment of wireless networks and resource deployment.
[0072] To solve the problems existing in the related art, an embodiment of the present application provides a network metric prediction device 80, as Figure 8 shown, including: An acquisition module 81, which acquires the network metric prediction requirement and the historical network metric data associated with the network metric prediction requirement, where the network metric prediction requirement includes a target prediction period and a target prediction scenario; An input module 82, which inputs the historical network metric data and the target prediction period into a time series prediction model, and obtains a first prediction result obtained by the time series prediction model for predicting the target prediction period; A first determination module 83, if the fluctuation degree of the network metric of the first prediction result is greater than the first preset fluctuation degree, then determines the first prediction result as the network metric prediction result for the target prediction period; A second determination module 84, if the fluctuation degree of the network metric of the first prediction result is less than the second preset fluctuation degree, then inputs the historical network metric data and the target prediction period into the specified scenario metric model matching the target prediction scenario, and determines the second prediction result output by the specified scenario metric model as the network metric prediction result for the target prediction period, where the second preset fluctuation degree is less than or equal to the first preset fluctuation degree.
[0073] Through the device provided by the embodiments of the present application, first, obtain the network metric prediction requirement and the historical network metric data associated with the network metric prediction requirement. Among them, the network metric prediction requirement includes the target prediction period and the target prediction scenario, providing a data basis for subsequent predictions. Then, input the historical network metric data and the target prediction period into the time series prediction model to obtain the first prediction result predicted by the time series prediction model for the target prediction period. Among them, the time series prediction model can highlight the data characteristics that change over time, so as to reflect whether the target prediction period is affected by the time law through the first prediction result. If the fluctuation degree of the network metric of the first prediction result is greater than the first preset fluctuation degree, it indicates that the target prediction period is greatly affected by the time law, and then the first prediction result is determined as the network metric prediction result of the target prediction period. If the fluctuation degree of the network metric of the first prediction result is less than the second preset fluctuation degree, it indicates that the target prediction period is less affected by the time law, and then the historical network metric data and the target prediction period are input into the specified scenario metric model matching the target prediction scenario, and the second prediction result output by the specified scenario metric model is determined as the network metric prediction result of the target prediction period, where the second preset fluctuation degree is less than or equal to the first preset fluctuation degree. When the target prediction period is less affected by the time law, the historical network metric data and the target prediction period are further input into the specified scenario metric model according to the target prediction scenario, so as to perform predictions using an appropriate scenario metric model for the target prediction scenario, making the second prediction result fit the target prediction scenario, thereby improving the accuracy of the finally output network metric prediction result. Through the solution provided by the embodiments of the present application, when the network metric data is greatly affected by the time law, the first prediction result predicted based on the time series prediction model can be output. When the network metric data is less affected by the time law, the second prediction result predicted based on the scenario metric model can be output, so that the output network metric prediction result fits the real data and effectively improves the accuracy of network metric prediction.
[0074] Among them, the above modules in the device provided by the embodiments of the present application can also implement the method steps provided by the above method embodiments. Or, the device provided by the embodiments of the present application may further include other modules in addition to the above modules to implement the method steps provided by the above method embodiments. And the device provided by the embodiments of the present application can achieve the technical effects that the above method embodiments can achieve.
[0075] Preferably, an embodiment of the present application further provides an electronic device, including a processor, a memory, and a computer program stored on the memory and operable on the processor. When the computer program is executed by the processor, it implements each process of the above embodiment of the network metric prediction method and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.
[0076] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by the processor, it implements each process of the above embodiment of the network metric prediction method and can achieve the same technical effect. To avoid repetition, it will not be elaborated here. Among them, the computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc, etc.
[0077] An embodiment of the present application further provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program. The computer program is operable to cause a computer to execute some or all of the steps of the above embodiment of the network metric prediction method and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.
[0078] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0079] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the specified functions in one Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.
[0080] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in one process Figure 1 or more processes and / or blocks Figure 1 or more blocks specified in a block or blocks.
[0081] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are performed on the computer or other programmable device to produce a computer-implemented process, thereby providing steps for implementing the functions specified in one process Figure 1 or more processes and / or blocks Figure 1 or more blocks specified in a block or blocks.
[0082] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.
[0083] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.
[0084] Computer-readable media includes both permanent and non-permanent, removable and non-removable media and can be implemented by any method or technology for storing information. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile discs (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.
[0085] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, commodity or device comprising a series of elements not only includes those elements but also other elements not expressly listed, or elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the phrase "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising said element.
[0086] Those skilled in the art should understand that the embodiments of the present application may be provided as a method, a system or a computer program product. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0087] The above are only embodiments of the present application and are not used to limit the present application. For those skilled in the art, the present application may have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.
Claims
1. A network metric prediction method, characterized in that, Including: Obtain the network metric prediction requirement and the historical network metric data associated with the network metric prediction requirement, where the network metric prediction requirement includes a target prediction time period and a target prediction scenario; Input the historical network metric data and the target prediction time period into a time series prediction model, and obtain a first prediction result predicted by the time series prediction model for the target prediction time period; If the network metric fluctuation degree of the first prediction result is greater than a first preset fluctuation degree, then determine the first prediction result as the network metric prediction result for the target prediction time period; If the network metric fluctuation degree of the first prediction result is less than a second preset fluctuation degree, then input the historical network metric data and the target prediction time period into a specified scenario metric model matching the target prediction scenario, and determine the second prediction result output by the specified scenario metric model as the network metric prediction result for the target prediction time period, where the second preset fluctuation degree is less than or equal to the first preset fluctuation degree.
2. The method according to claim 1, characterized in that, After inputting the historical network metric data and the target prediction time period into a time series prediction model and obtaining a first prediction result predicted by the time series prediction model for the target prediction time period, it further includes: Determine the network metric fluctuation coefficient of the first prediction result according to first fluctuation test data and at least one second fluctuation test data, where the first fluctuation test data includes the network metric prediction data of a first fluctuation test time period in the first prediction result, and the second fluctuation test data includes the historical network metric data of a historical same period of the first fluctuation test time period; If the network metric fluctuation coefficient of the first prediction result is greater than a preset network metric fluctuation threshold, then determine that the network metric fluctuation degree of the first prediction result is greater than the first preset fluctuation degree.
3. The method according to claim 2, wherein The number of the second fluctuation test data is multiple; Among them, determining the network metric fluctuation coefficient of the first prediction result according to first fluctuation test data and at least one second fluctuation test data includes: Determine the ratio of the first fluctuation test data to the average value of the test data as the network metric fluctuation coefficient of the first prediction result, where the average value of the test data includes the average value of multiple second fluctuation test data.
4. The method according to any one of claims 1 to 3, characterized in that, Input the historical network metric data and the target prediction time period into a specified scenario metric model matching the target prediction scenario, and determine the second prediction result output by the specified scenario metric model as the network metric prediction result for the target prediction time period, including: If the target prediction scenario is a scenario for predicting a single network metric, then input the historical network metric data and the target prediction time period into a LightGBM (Light Gradient Boosting Machine) model, and determine the second prediction result output by the LightGBM model as the network metric prediction result for the target prediction time period.
5. The method according to any one of claims 1 to 3, characterized in that Input the historical network metric data and the target prediction period into the specified scenario metric model matching the target prediction scenario, and determine the second prediction result output by the specified scenario metric model as the network metric prediction result for the target prediction period, including: If the target prediction scenario is a scenario for predicting multiple network metrics, input the historical network metric data and the target prediction period into a long short-term memory recurrent neural network (LSTM) model, and determine the second prediction result output by the LSTM model as the network metric prediction result for the target prediction period.
6. The method according to any one of claims 1 to 3, characterized in that Input the historical network metric data and the target prediction period into the specified scenario metric model matching the target prediction scenario, and determine the second prediction result output by the specified scenario metric model as the network metric prediction result for the target prediction period, including: If the target prediction scenario is a scenario for predicting multiple network metrics of multiple network elements respectively, input the historical network metric data and the target prediction period into an autoregressive recurrent network (DeepAR) model, and determine the second prediction result output by the DeepAR model as the network metric prediction result for the target prediction period.
7. A network metric prediction device, characterized in that, Including: An acquisition module that acquires a network metric prediction requirement and historical network metric data associated with the network metric prediction requirement, where the network metric prediction requirement includes a target prediction period and a target prediction scenario; An input module that inputs the historical network metric data and the target prediction period into a time series prediction model to obtain a first prediction result of the time series prediction model for the target prediction period; A first determination module that, if the degree of fluctuation of the network metrics of the first prediction result is greater than a first preset degree of fluctuation, determines the first prediction result as the network metric prediction result for the target prediction period; A second determination module that, if the degree of fluctuation of the network metrics of the first prediction result is less than a second preset degree of fluctuation, inputs the historical network metric data and the target prediction period into the specified scenario metric model matching the target prediction scenario, and determines the second prediction result output by the specified scenario metric model as the network metric prediction result for the target prediction period, where the second preset degree of fluctuation is less than or equal to the first preset degree of fluctuation.
8. An electronic device, characterized in that, Including: A memory, a processor, and a computer program stored on the memory and executable on the processor, where the computer program, when executed by the processor, implements the steps of the method according to any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.
10. A computer program product, characterized in that, The computer program product includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute the steps of the method according to any one of claims 1 to 6.