WO2021129086A1 - 流量预测方法、装置以及存储介质 - Google Patents

流量预测方法、装置以及存储介质 Download PDF

Info

Publication number
WO2021129086A1
WO2021129086A1 PCT/CN2020/122552 CN2020122552W WO2021129086A1 WO 2021129086 A1 WO2021129086 A1 WO 2021129086A1 CN 2020122552 W CN2020122552 W CN 2020122552W WO 2021129086 A1 WO2021129086 A1 WO 2021129086A1
Authority
WO
WIPO (PCT)
Prior art keywords
component
prediction
fitted
traffic
time period
Prior art date
Application number
PCT/CN2020/122552
Other languages
English (en)
French (fr)
Inventor
韩静
张百胜
刘建伟
左兴权
谷勇浩
徐洋凡
黄锦
Original Assignee
中兴通讯股份有限公司
北京邮电大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中兴通讯股份有限公司, 北京邮电大学 filed Critical 中兴通讯股份有限公司
Priority to US17/782,993 priority Critical patent/US11863397B2/en
Priority to EP20904338.9A priority patent/EP4047878A4/en
Publication of WO2021129086A1 publication Critical patent/WO2021129086A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/142—Network analysis or design using statistical or mathematical methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/147—Network analysis or design for predicting network behaviour
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
    • H04L43/0876—Network utilisation, e.g. volume of load or congestion level

Definitions

  • the embodiments of the present application relate to the field of communications, and in particular, to a traffic prediction method, device, and storage medium.
  • base stations can provide stable and reliable wireless traffic transmission services will directly affect the services of operators. quality. Predicting the traffic of the base station cell for a period of time in the future, combined with the traffic threshold of the base station equipment, can provide a reference for the expansion and contraction of the base station, so as to ensure the reasonable allocation of resources and user service quality.
  • the purpose of the embodiments of the present application is to provide a traffic prediction method, device, and storage medium.
  • the embodiment of the present application provides a flow prediction method, which includes: obtaining flow data for a first preset time period in a historical period, and preprocessing the flow data;
  • the flow data is decomposed by empirical mode to obtain multiple component sequences;
  • the multiple component sequences are fitted by a time series prediction model, and the multiple component prediction results for the second preset time period are obtained by using the fitted time series prediction model ; Accumulate all the component prediction results to obtain the flow prediction result for the second preset time period.
  • the embodiment of the present application also provides a traffic prediction device, including: at least one processor; and, a memory communicatively connected with the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are At least one processor is executed, so that the at least one processor can execute the above-mentioned traffic prediction method.
  • the embodiment of the present application also provides a computer-readable storage medium that stores a computer program, and when the computer program is executed by a processor, the above-mentioned traffic prediction method is implemented.
  • Fig. 1 is a schematic flow chart of a traffic prediction method according to the first embodiment of the present application
  • FIG. 2 is a schematic diagram of a process flow of processing abnormal data according to the box and plot method of the first embodiment of the present application;
  • FIG. 3 is a schematic diagram of the flow of empirical mode decomposition decomposition according to the first embodiment of the present application
  • FIG. 4 is a schematic flowchart of a flow prediction method according to a second embodiment of the present application.
  • FIG. 5 is a schematic flowchart of a traffic prediction method according to a third embodiment of the present application.
  • FIG. 6 is a schematic flowchart of a flow prediction method according to a fourth embodiment of the present application.
  • Fig. 7 is a schematic structural diagram of a traffic prediction device according to a fifth embodiment of the present application.
  • the first embodiment of the present application relates to a traffic prediction method.
  • the core of this embodiment is to obtain historical traffic data and preprocess the historical traffic data; and perform empirical mode decomposition on the preprocessed historical traffic data. , Obtain multiple component sequences; use a time-series prediction model to predict each component sequence separately to obtain the component prediction result of each component sequence; accumulate all the component prediction results to obtain the flow prediction result, so that the accuracy of the flow prediction improve.
  • the following specifically describes the implementation details of the traffic prediction method of this embodiment. The following content is only provided for ease of understanding and is not necessary for the implementation of this solution.
  • Step 101 Obtain the flow data of the first preset time period in the historical period, and preprocess the flow data.
  • the traffic data of each day of the first preset time period in the historical period is acquired, and the traffic data of each day is the highest utilization rate of the physical resource block (prb) among the 24 moments of the day.
  • the flow value at the moment is not limited to the acquisition of the traffic value at the moment when the prb utilization rate is the highest among the 24 hours of each day, and other solutions may also be used.
  • Determine the flow data for each day for example, use the average flow rate within 24 hours of each day as the flow data for each day, or use the maximum flow rate within 24 hours of each day as the flow data for each day, etc. This will not be repeated one by one.
  • the flow data needs to be processed into one file.
  • the missing values and abnormal values in the flow data need to be processed.
  • the flow data when the flow data is uniformly processed into one file, the flow data needs to be filtered to obtain the field attribute information required for the flow prediction.
  • the field attribute information required for traffic prediction includes: traffic data size ([LTE] DL CELL PDCP SDU Volume (Kbyte)), traffic collection date (Date), traffic collection time (Time), collection area number ( Cell) and base station number (EnodeB), etc.
  • the two variables EnodeB and Cell can be used to identify the cell (the two variables determine the unique identification of the cell), and a num column can also be added to the determined cell for numbering.
  • the processing method for missing values in this embodiment is to remove missing values.
  • the missing value refers to the numbered cell (usually a cell with a null value in the field) that does not have the corresponding collected flow data size value, flow collection date, or flow collection time after the data is unified into one file.
  • the preprocessing process Directly removing the data of the above-mentioned cell in, can prevent the missing value from affecting the prediction result and improve the accuracy of the prediction.
  • the box plot method is used to process the abnormal value in this embodiment.
  • Outliers are outliers that deviate greatly from the normal value in the size of the flow data.
  • a total of 211 days of flow data has been collected in a certain area, and the outliers of the data flow are processed by calculating the quartile values of these flow data.
  • Step 1013 Calculate the maximum value Top and the minimum value Low.
  • the calculation formulas are:
  • Step 1014 Determine whether the data sequence X(t) is greater than the maximum value, if yes, go to step 1015, if not, go to step 1016;
  • Step 1015 Update X(t), the update formula is:
  • Step 1016 Determine whether the data sequence X(t) is less than or equal to the minimum value, if yes, go to step 1017, if not, go to step 1018;
  • Step 1017 Update X(t), the update formula is:
  • Step 102 Perform empirical mode decomposition on the preprocessed flow data to obtain multiple component sequences.
  • Empirical Mode Decomposition is performed on the preprocessed historical flow data to obtain multiple Intrinsic Mode Function (IMF) components and a residual component, And a single reconstruction of each sequence corresponding to each component.
  • IMF Intrinsic Mode Function
  • the traversal is performed from low to high, and the EMD decomposition is performed on each cell.
  • the steps of performing empirical mode decomposition on the flow data of a cell mainly include:
  • Step 1021 Obtain input preprocessed cell flow data X(t) (X(t) is a time series of flow values, where time t represents a date in a certain day as a unit);
  • Step 1022 Calculate the maximum value envelope X max (t), the minimum value envelope X min (t), and the envelope average value m(t);
  • Step 1024 Judge whether h(t) meets the IMF condition, if yes, go to step 1026, if not, go to step 1025;
  • IMF conditions include:
  • Condition (1) In the entire data set, the number of local extreme values and the number of values greater than zero are the same or the difference is equal to one;
  • Condition (2) At any point in time, the average value of the upper envelope defined by the local maximum and the lower envelope defined by the local minimum is zero;
  • Step 1026 Use h(t) as an IMF component
  • Step 1028 Judge whether r(t) has a monotonous trend, if yes, complete the EMD decomposition of the cell traffic data, if not, go to step 1029;
  • Step 103 Fit the multiple component sequences with a time series prediction model, and use the fitted time series prediction model to obtain multiple component prediction results for a second preset time period.
  • the time series prediction model is first used to fit the component sequences to obtain the fitted time series prediction model; and then the second preset time period for each component sequence is determined according to the fitted time series model The component prediction result.
  • the time series prediction model may be one of a prophet model, an autoregressive model, a moving average model, or an autoregressive moving average model.
  • a prophet model is used to fit each component sequence.
  • Step 104 Accumulate all the component prediction results to obtain a traffic prediction result for a second preset time period.
  • the component prediction result is the prediction result of the second preset time period of the component sequence obtained by empirical mode decomposition based on historical flow data
  • the result of adding the prediction results of each component is the second preset time period Traffic forecast results.
  • the embodiment of this application performs empirical modal decomposition on the preprocessed historical flow data to obtain multiple component sequences; uses a time series prediction model to fit multiple component sequences, and uses the fitted time sequence
  • the prediction model obtains multiple component prediction results for the second preset time period; accumulates all component prediction results to obtain a traffic prediction result.
  • the empirical mode decomposition method performs signal decomposition based on the time scale characteristics of the data itself, there is no need to set any basis functions in advance, and it has obvious advantages in processing non-stationary and nonlinear data. Therefore, the historical flow data is decomposed by empirical mode decomposition. Decomposition into a relatively stable component sequence reduces the impact of large fluctuations in data on the accuracy of prediction, thereby improving the accuracy of traffic prediction.
  • the second embodiment of the present application relates to a traffic prediction method.
  • the second embodiment is roughly the same as the first embodiment, but the difference is that the second embodiment uses a time series prediction model to fit the multiple component sequences, and uses the fitted time series prediction model to obtain a second preset
  • each component sequence is decomposed into the sum of trend, seasonal, and noise items; the fitted trend item and the fitted seasonal item are determined respectively, and then the simulation
  • the combined trend item and the fitted seasonal item obtain the trend item forecast result and the seasonal item forecast result of the second preset time period; accumulate the trend item forecast result, the seasonal item forecast result and the noise item to obtain the second preset time
  • the component prediction result of each component sequence of the segment is roughly the same as the first embodiment, but the difference is that the second embodiment uses a time series prediction model to fit the multiple component sequences, and uses the fitted time series prediction model to obtain a second preset
  • each component sequence is decomposed into the sum of trend, seasonal, and noise items; the fitted trend item and the fitted seasonal item are determined respectively,
  • the traffic prediction method in this embodiment is shown in Figure 3, and specifically includes:
  • Step 201 Obtain the flow data of the first preset time period in the historical period, and preprocess the flow data.
  • Step 202 Perform empirical mode decomposition on the preprocessed flow data to obtain multiple component sequences.
  • Steps 201 to 202 are substantially the same as steps 101 to 102 in the first embodiment, respectively. To avoid repetition, they will not be repeated here.
  • Step 203 Decompose each component sequence into the sum of trend items, seasonal items, and noise items;
  • the prophet model is used to predict each component sequence separately, and the component prediction result of each component sequence is obtained.
  • t is the time
  • y(t) is the component sequence
  • g(t) is the trend term
  • s(t) is the seasonal term
  • ⁇ t is the noise term.
  • Step 204 Determine the fitted trend item and the fitted seasonal item respectively, and then use the fitted trend item and the fitted seasonal item to obtain the trend item forecast result and the seasonal item forecast for the second preset time period result.
  • the known component sequence and the preset fitting function respectively determine the fitted trend item and the seasonal item, and then use the fitted trend item and the seasonal item to obtain the trend of the second preset time period Forecast results and seasonal forecast results.
  • the fitting function of the trend term g(t) is
  • t is the time parameter
  • k is the data flow growth rate
  • m is the offset parameter
  • ⁇ is the growth rate vector
  • ⁇ j is set to -s j ⁇ j to make the function continuous
  • a(t) is a custom vector and T is the transpose operator.
  • the fitting function of the trend term clearly defines the transformation point that allows the growth rate to change, and incorporates the trend change in the growth model into the trend term fitting function, which is specifically derived by the following method:
  • a growth rate vector ⁇ j is the change in rate at time s j.
  • the growth rate at any time t is the reference rate k plus all the rate changes up to that point in time:
  • the fitting function of the seasonal term s(t) is a Fourier function, specifically:
  • ⁇ (0, ⁇ 2 ) that is, ⁇ obeys a normal distribution, and the mathematical expectation of the normal distribution is 0, and the variance is ⁇ 2 ;
  • the noise term ⁇ t represents changes that the model cannot capture, and it is assumed that the noise term obeys a normal distribution.
  • Step 205 Accumulate the prediction result of the trend item, the prediction result of the seasonal item, and the noise item to obtain the component prediction result of each component sequence.
  • the prediction result of the trend item is obtained through step 203
  • Seasonal forecast results The noise term ⁇ t takes a random value that obeys the normal distribution and accumulates And ⁇ t to obtain the component prediction result of a single component sequence, namely:
  • Is the component prediction result Is the forecast result of the trend item, Is the predicted result of the seasonal term, and ⁇ t is the noise term.
  • Step 206 Accumulate all the component prediction results to obtain the traffic prediction result for the second preset time period.
  • Step 206 is substantially the same as step 104 in the first embodiment. To avoid repetition, it will not be repeated here.
  • this embodiment uses a time series model to process the component sequences separately and obtain the component prediction results of the component sequences, and specifically uses the prophet model to decompose each component sequence into trend items and seasonal items. The sum of the term and the noise term is fitted, and then the fitted prophet model is used to obtain the flow prediction result.
  • the prophet model By using the prophet model to predict the component sequence, it is possible to simultaneously predict the flow based on the trend of periodic changes and the trend of non-periodical changes in the flow data, which improves the accuracy of the flow forecast.
  • the third embodiment of the present application relates to a traffic prediction method.
  • the third embodiment is roughly the same as the second embodiment, except that, in the step of decomposing each component sequence into the sum of trend terms, seasonal terms, and noise terms, the third embodiment divides each component The sequence is decomposed into the sum of trend items, seasonal items, holiday items, and noise items; and in the subsequent steps, the fitted trend items, seasonal items, and holiday items are determined respectively, and then the fitted trend items and seasonal items are used And holiday items get the forecast results of trend items, seasonal items and holiday items in the second preset time period; accumulate the forecast results of trend items, seasonal items, forecast results of holiday items, and noise items to get the second preset time period The component prediction result of each component sequence of.
  • the traffic prediction method in this embodiment is shown in Figure 4, and specifically includes:
  • Step 301 Obtain the flow data of the first preset time period in the historical period, and preprocess the flow data.
  • Step 302 Perform empirical mode decomposition on the preprocessed historical flow data to obtain multiple component sequences.
  • Steps 301 to 302 are approximately the same as steps 101 to 102 in the first embodiment, respectively. To avoid repetition, they will not be repeated here.
  • Step 303 Decompose each component sequence into the sum of trend items, seasonal items, holiday items, and noise items;
  • the prophet model is used to predict each component sequence separately, and the component prediction result of each component sequence is obtained.
  • each component sequence is decomposed into the sum of trend items, seasonal items, and noise items, namely
  • t is the time
  • y(t) is the component sequence
  • g(t) is the trend term
  • s(t) is the seasonal term
  • ⁇ t is the noise term.
  • Step 304 Determine the fitted trend item, the fitted seasonal item and the fitted holiday item respectively, and then use the fitted trend item, the fitted seasonal item and the fitted holiday item to obtain the first 2.
  • the fitting functions of the trend term, the seasonal term, and the noise term are similar to those in step 204 of the second embodiment, and will not be repeated here.
  • the fitting function of the holiday term h(t) is as follows:
  • ⁇ is the normal distribution curve
  • Z(t) [1(t ⁇ D 1 ),...,1(t ⁇ D L )], for the Lth holiday, D L means that the holiday has an impact Time period.
  • Step 305 Accumulate the prediction result of the trend item, the prediction result of the seasonal item, the prediction result of the holiday item, and the noise item to obtain the component prediction result of each component sequence.
  • the prediction result of the trend item is obtained through step 303 Seasonal forecast results And the forecast results of the holiday items
  • the noise term ⁇ t takes a random value that obeys the normal distribution and accumulates And ⁇ t to obtain the component prediction result of a single component sequence, namely:
  • Is the component prediction result Is the forecast result of the trend item, Is the forecast result of the seasonal item, Is the prediction result of the holiday term, and ⁇ t is the noise term.
  • Step 306 Accumulate all the component prediction results to obtain a traffic prediction result.
  • Step 306 is substantially the same as step 104 in the first embodiment, and to avoid repetition, it will not be repeated here.
  • this embodiment decomposes each component sequence into the sum of trend items, seasonal items, holiday items, and noise items, and then performs fitting.
  • the prophet model obtained the flow forecast results.
  • the fourth embodiment of the present application relates to a traffic prediction method.
  • the fourth embodiment is roughly the same as the first embodiment.
  • the difference is that in the fourth embodiment, in the step of fitting the multiple component sequences with a time series prediction model, it specifically includes: according to a preset step size, All component sequences are divided into component training set and component test set; the time series prediction model is used to fit the component sequence of the component training set; according to the component test set, the prediction error of the fitted time series prediction model is determined.
  • the traffic prediction method in this embodiment is shown in Figure 5, and specifically includes:
  • Step 401 Obtain the flow data of the first preset time period in the historical period, and preprocess the flow data.
  • Step 402 Perform empirical mode decomposition on the preprocessed flow data to obtain multiple component sequences.
  • Steps 401 to 402 are substantially the same as steps 101 to 102 in the first embodiment, respectively, and to avoid repetition, they will not be repeated here.
  • Step 403 Divide all the component sequences into a component training set and a component test set according to a preset step length
  • all component sequences are divided into a component training set and a component test set according to a preset step size, which may be specifically: taking the duration of the second preset time period as the preset time period.
  • Set the step size; in all the component sequences, the set of data outside the preset step that is closest to the current time is used as the component training set; among all the component sequences, the set of data within the preset step that is closest to the current time
  • the collection of data is used as the component test set.
  • the component sequences of the last 30 days of the past 210 days of all cells are used as the component test set, and the component sequences of all the remaining cells of the previous 180 days are used as the component training set.
  • Step 404 Fit the component sequence of the component training set with a time series prediction model.
  • step 103 of the first embodiment a time series prediction model is used to fit each component sequence.
  • step 103 of the first embodiment a time series prediction model is used to fit the components in the training set. Each component sequence, and the fitted time series prediction model is obtained.
  • Step 405 Determine the prediction error of the fitted time series prediction model according to the component test set.
  • step 403 is used to divide the component sequence to obtain the component training set and the component test set.
  • step 403 determines the historical time period closest to the current time. Traffic prediction result, wherein the duration of the historical time period is the preset step size; according to the component test set and the traffic prediction result of the historical time period, the fitted time series prediction model is determined Forecast error.
  • the prediction effect is evaluated by calculating the mean absolute percentage error (MAPE):
  • I the traffic prediction result in the historical time period
  • y t the component sequence corresponding to the component test set.
  • Step 406 Use the fitted time series prediction model to obtain multiple component prediction results for the second preset time period, and accumulate all the component prediction results to obtain the traffic prediction result for the second preset time period.
  • step 404 the fitted time series prediction model obtained in step 404 is used to predict the second preset time period and obtain multiple component prediction results.
  • the remaining steps are roughly the same as step 104 in the first embodiment, in order to avoid repetition , I won’t repeat it here.
  • this embodiment divides the multiple component sequences obtained by EMD decomposition into a component training set and a component test set.
  • the component training set is used to determine the fitted time series prediction model
  • the component test set is used to Evaluate the prediction accuracy of the time series prediction model fitted by the component training set.
  • the prediction accuracy of the time series prediction model is evaluated through cross-validation, so that the model can be retrained when the model prediction accuracy is low, and the prediction accuracy can reach the ideal state.
  • the fifth embodiment of the present application relates to a traffic prediction device. As shown in FIG. 6, it includes at least one processor 501; and a memory 502 communicatively connected with at least one processor 501; wherein, the memory 502 stores at least one The instructions executed by the processor 501 are executed by the at least one processor 501, so that the at least one processor 501 can execute the foregoing traffic prediction method embodiment.
  • the memory 502 and the processor 501 are connected in a bus manner.
  • the bus may include any number of interconnected buses and bridges, and the bus connects one or more various circuits of the processor 501 and the memory 502.
  • the bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, etc., which are all known in the art, and therefore, no further description will be given herein.
  • the bus interface provides an interface between the bus and the transceiver.
  • the transceiver may be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on the transmission medium.
  • the data processed by the processor 501 is transmitted on a wireless medium through an antenna.
  • the antenna also receives the data and transmits the data to the processor 501.
  • the processor 501 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces,
  • the memory 502 may be used to store data used by the processor 501 when performing operations.
  • the embodiment of the present application also provides a computer-readable storage medium that stores a computer program, and when the computer program is executed by a processor, the foregoing embodiment of the traffic prediction method is implemented.
  • the program is stored in a storage medium and includes several instructions to enable a device ( It may be a single-chip microcomputer, a chip, etc.) or a processor (processor) to execute all or part of the steps of the methods described in the embodiments of the present application.
  • the aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks or optical disks and other media that can store program codes. .

Landscapes

  • Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Algebra (AREA)
  • Environmental & Geological Engineering (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Physics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Pure & Applied Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本申请实施例涉及通信领域，公开了一种流量预测方法，包括：获取历史时期内第一预设时间段的流量数据，并对流量数据进行预处理；对预处理后的流量数据进行经验模态分解(EMD分解)，得到多个分量序列；采用时间序列预测模型拟合多个分量序列，利用拟合后的时间序列预测模型得到第二预设时间段的多个分量预测结果；累加所有分量预测结果，得到第二预设时间段的流量预测结果。本发明还提供了一种流量预测装置和存储介质。

Description

流量预测方法、装置以及存储介质
相关申请的交叉引用
本申请基于申请号为201911359122.9、申请日为2019年12月25日的中国专利申请提出，并要求该中国专利申请的优先权，该中国专利申请的全部内容在此以引入方式并入本申请。
技术领域
本申请实施例涉及通信领域，特别涉及一种流量预测方法、装置以及存储介质。
背景技术
随着无线通信技术的发展，互联网规模在不断增大，人们上网的习惯以及对流量的需求也发生了天翻地覆的变化，基站能否提供稳定可靠的无线流量传输服务，将直接影响运营商的服务质量。预测基站小区未来一段时间的流量、结合基站设备承受的流量阈值，可以为基站扩缩容提供参考，从而保证资源的合理分配和用户服务质量。
然而，发明人发现现有技术中至少存在如下问题：在进行流量预测时，若历史流量数据中存在大幅度波动的数据，传统算法模型的预测准确度低。
发明内容
本申请实施方式的目的在于提供一种流量预测方法、装置以及存储介质。
为解决上述技术问题，本申请的实施方式提供了一种流量预测方法，包括：获取历史时期内第一预设时间段的流量数据，并对所述流量数据进行预处理；对预处理后的流量数据进行经验模态分解，得到多个分量序列；采用时间序列 预测模型拟合所述多个分量序列，利用拟合后的时间序列预测模型得到第二预设时间段的多个分量预测结果；累加所有所述分量预测结果，得到第二预设时间段的流量预测结果。
本申请的实施方式还提供了一种流量预测装置，包括：至少一个处理器；以及，与至少一个处理器通信连接的存储器；其中，存储器存储有可被至少一个处理器执行的指令，指令被至少一个处理器执行，以使至少一个处理器能够执行上述的流量预测方法。
本申请的实施方式还提供了一种计算机可读存储介质，存储有计算机程序，所述计算机程序被处理器执行时实现上述流量预测方法。
附图说明
一个或多个实施例通过与之对应的附图中的图片进行示例性说明，这些示例性说明并不构成对实施例的限定。
图1是根据本申请第一实施方式的流量预测方法的流程示意图；
图2是根据本申请第一实施方式的箱线图法处理异常数据的流程示意图；
图3是根据本申请第一实施方式的经验模态分解分解的流程示意图；
图4是根据本申请第二实施方式的流量预测方法的流程示意图；
图5是根据本申请第三实施方式的流量预测方法的流程示意图；
图6是根据本申请第四实施方式的流量预测方法的流程示意图；
图7是根据本申请第五实施方式的流量预测装置的结构示意图。
具体实施方式
为使本申请实施例的目的、技术方案和优点更加清楚，下面将结合附图对本申请的各实施方式进行详细的阐述。然而，本领域的普通技术人员可以理解，在本申请各实施方式中，为了使读者更好地理解本申请而提出了许多技术细节。但是，即使没有这些技术细节和基于以下各实施方式的种种变化和修改，也可以实现本申请所要求保护的技术方案。
本申请的第一实施方式涉及一种流量预测方法，本实施方式的核心在于， 获取历史流量数据、并对所述历史流量数据进行预处理；对预处理后的历史流量数据进行经验模态分解、得到多个分量序列；采用时间序列预测模型分别预测所述各分量序列、得到所述各分量序列的分量预测结果；累加所有所述分量预测结果、得到流量预测结果，使得流量预测的准确度提高。下面对本实施方式的流量预测方法的实现细节进行具体的说明，以下内容仅为方便理解提供的实现细节，并非实施本方案的必须。
本实施方式中的流量预测方法如图1所示：
步骤101：获取历史时期内第一预设时间段的流量数据，并对所述流量数据进行预处理。
具体地说，在实际应用场景中要对同一基站下多个小区未来一段时间的流量进行预测之前，先获取运营商采集到的小区历史时期的流量数据。
在一个示例性实施例中，获取历史时期内第一预设时间段每一天的流量数据，每一天的流量数据为当天24个时刻中物理资源块(physical resource block，简称prb)利用率最大的时刻的流量值。需要说明的是，所述获取历史时期内第一预设时间段每一天的流量数据，并不局限于获取每一天24时刻中prb利用率最大的时刻的流量值，也可以是采用其他方案来确定每一天的流量数据，例如，将每一天24个小时之内的流量平均值作为每一天的流量数据，或将每一天24个小时之内的最大流量作为每一天的流量数据等等，在此不再一一赘述。
接下来，由于流量数据是以多个文件保存的，所以需要将流量数据统一处理成一个文件，此外还需要对流量数据中的缺失值以及异常值进行处理。
在一示例性实施例中，将流量数据统一处理成一个文件时，需要对流量数据进行过滤得到流量预测所需要的字段属性信息。本实施方式中，流量预测所需要的字段属性信息包括：流量数据大小([LTE]DL CELL PDCP SDU Volume(Kbyte))、流量采集日期(Date)、流量采集时刻(Time)、采集区域编号(Cell)和基站编号(EnodeB)等。其中，EnodeB以及Cell这两个变量可以用来对小区进行标识(两个变量确定唯一标识小区)，还可以对已确定的小区增加一个num列进行编号。
在一个示例性实施例中，本实施方式中对于缺失值的处理方式是去除缺失值。缺失值指的是将数据统一成一个文件后，没有对应的采集的流量数据大小值、流量采集日期或流量采集时刻的已编号的小区(一般是含有字段为空值的 小区)，预处理过程中直接去除上述小区的数据，可以防止缺失值影响预测结果，提升预测的准确度。
在一个示例性实施例中，本实施方式中采用箱线图法对异常值进行处理。异常值指的是流量数据大小中偏离正常值很大的离群值，假设某地区总共采集了211天的流量数据，通过计算这些流量数据的四分位数值来处理数据流量的离群值。通过采用箱线图法对历史流量数据中的异常数据进行处理，使得异常数据对后续采用时间序列预测模型预测分量序列的影响降低，从而提高了流量预测的准确度。如图2所示，其具体包括以下步骤：
步骤1011：输入数据流量序列X(t)(X(t)是一条关于流量值的时间序列，其中时间t表示以某天日期为单位)，取t＝0；
步骤1012：计算上四分位数Q3、下四分位数Q1以及四分位距IQR(IQR＝Q3–Q1)；
步骤1013：计算最大值Top和最小值Low，计算公式分别为：
Top＝Q3+1.1*IQR
Low＝Q1-1.1*IQR
步骤1014：判断数据序列X(t)是否大于最大值，如果是，进入步骤1015，如果否，进入步骤1016；
步骤1015：更新X(t)，更新公式为：
X(t)＝(Top+X(t))/2
步骤1016：判断数据序列X(t)是否小于或等于最小值，如果是，进入步骤1017，如果否，进入步骤1018；
步骤1017：更新X(t)，更新公式为：
X(t)＝(Low+X(t))/2
步骤1018：判断t是否等于N(本例子中，N＝211)，如果否，进入步骤1019，如果是，则完成对流量数据序列X(t)的异常值处理；
步骤1019：t＝t+1，获取X(t)序列的下一个流量值进行处理。
通过上述步骤对流量数据中的离群值进行处理，可减少离群值对预测结果的影响。
步骤102：对预处理后的流量数据进行经验模态分解，得到多个分量序列。
具体地说，对预处理后的历史流量数据进行经验模态分解(Empirical Mode Decomposition，简称：EMD分解)，得到多个本质模态函数(Intrinsic Mode Function，简称IMF)分量和一个残差分量，并对每个分量所对应的各个序列进行单一重建。
在一个示例性实施例中，本实施方式中，按照步骤101中预处理后数据num字段编号的顺序，从低到高依次进行遍历，对每一个小区进行EMD分解。如图3所示，对一个小区的流量数据进行经验模态分解的步骤主要包括：
步骤1021：获取输入预处理后的小区流量数据X(t)(X(t)是一条关于流量值的时间序列，其中时间t表示以某天日期为单位)；
步骤1022：计算极大值包络线X max(t)、极小值包络线X min(t)以及包络线平均值m(t)；
具体地说，标记出X(t)的局部极大值和极小值，然后使用三次样条曲线拟合求出局部极大值或极小值的极大值包络线X max(t)和极小值包络线X min(t)，并计算上下包络线的平均值m(t)＝(X max(t)+X min(t))/2；
步骤1023：确定疑似IMF分量的分量序列h(t)，h(t)＝X(t)-m(t)；
步骤1024：判断h(t)是否满足IMF条件，如果是，进入步骤1026，如果否，进入步骤1025；
具体地说，IMF条件包括：
条件(1)：在整个数据集中，局部极值的数量和大于零的值的数量相同或差值等于一；
条件(2)：在任何时间点，由局部极大值定义的上包络线和由局部极小值定义的下包络线的均值为零；
步骤1025：更新X(t)：X(t)＝h(t)；
步骤1026：将h(t)作为一个IMF分量；
具体地说，如果h(t)满足步骤1024中的条件，则h(t)可作为一个IMF分量并编号IMF i其中i∈{0,1,2...n}，更新编号下标i＝i+1
步骤1027：计算残差分量r(t)，r(t)＝X(t)-h(t)；
步骤1028：判断r(t)是否趋势单调，如果是，完成对该小区流量数据的EMD分解，如果否，进入步骤1029；
步骤1029：更新X(t)：X(t)＝r(t)；
通过上述步骤进行EMD分解，最终得到n个分量序列：X(t)＝IMF 0+IMF 1+IMF 2+IMF 3+IMF 4+…+IMF n-2+r(t)。
步骤103：采用时间序列预测模型拟合所述多个分量序列，利用拟合后的时间序列预测模型得到第二预设时间段的多个分量预测结果。
具体地说，先采用时间序列预测模型对所述各分量序列进行拟合、得到拟合后的时间序列预测模型；再根据拟合后的时间序列模型，确定各分量序列第二预设时间段的分量预测结果。其中，时间序列预测模型可以是prophet模型、自回归模型、滑动平均模型或自回归滑动平均模型中的一种。在一个示例性实施例中，采用prophet模型对各分量序列进行拟合。
步骤104：累加所有所述分量预测结果，得到第二预设时间段的流量预测结果。
具体地说，由于分量预测结果为根据历史流量数据进行经验模态分解得到的分量序列的第二预设时间段的预测结果，将各分量预测结果相加后的结果为第二预设时间段的流量预测结果。
与现有技术相比，本申请实施方式对预处理后的历史流量数据进行经验模态分解、得到多个分量序列；采用时间序列预测模型拟合多个分量序列，利用拟合后的时间序列预测模型得到第二预设时间段的多个分量预测结果；累加所有分量预测结果，得到流量预测结果。由于经验模态分解方法依据数据自身的时间尺度特征来进行信号分解，无须预先设定任何基函数，在处理非平稳及非线性数据上具有明显的优势，因此通过经验模态分解将历史流量数据分解为较为平稳的分量序列，降低了数据大幅度波动对预测准确性的影响，从而提高了流量预测的准确度。
本申请的第二实施方式涉及一种流量预测方法。第二实施方式与第一施方式大致相同，不同之处在于，第二实施方式在采用时间序列预测模型拟合所述多个分量序列，利用拟合后的时间序列预测模型得到第二预设时间段的多个分量预测结果这一步骤中，将每个分量序列分解为趋势项、季节项、噪声项之和； 分别确定拟合后的趋势项和拟合后的季节项，再利用拟合后的趋势项和拟合后的季节项得到第二预设时间段的趋势项预测结果和季节项预测结果；累加趋势项预测结果、季节项预测结果以及噪声项，得到第二预设时间段的每个分量序列的分量预测结果。
本实施方式中的流量预测方法如图3所示，具体包括：
步骤201：获取历史时期内第一预设时间段的流量数据，并对所述流量数据进行预处理。
步骤202：对预处理后的流量数据进行经验模态分解，得到多个分量序列。
步骤201至步骤202分别与第一实施方式中步骤101至步骤102大致相同，为避免重复，在此不再一一赘述。
步骤203：将每个所述分量序列分解为趋势项、季节项、噪声项之和；
在本实施方式中，采用prophet模型分别预测各分量序列、得到各分量序列的分量预测结果。
具体地说，先通过局部加权回归进行季节性和趋势性分解(Seasonal and Trend decomposition using Loess，简称：STL)，将每个所述分量序列分解为趋势项、季节项、噪声项之和，即
y(t)＝g(t)+s(t)+∈ t
式中，t为时间、y(t)为分量序列、g(t)为趋势项，s(t)为季节项、∈ t为噪声项。
需要说明的是，本步骤中“利用prophet模型进行分量序列预测”的做法并非必需，在其他可变更的实施方案中，也可采用其他的时间序列模型对各个分量序列进行预测，在此不再一一赘述。
步骤204：分别确定拟合后的趋势项和拟合后的季节项，再利用拟合后的趋势项和拟合后的季节项得到第二预设时间段的趋势项预测结果和季节项预测结果。
具体地说，根据已知的分量序列和预设的拟合函数，分别确定拟合后的趋势项和季节项，再利用拟合后的趋势项和季节项得到第二预设时间段的趋势项预测结果和季节项预测结果。
在一个示例性实施例中，趋势项g(t)的拟合函数为
g(t)＝(k+a(t) Tδ)t+(m+a(t) Tγ)
式中，t为时间参数、k为数据流量增长率、m为偏移参数、δ为增长率向量、γ j设为-s jδ j来使该函数连续；
s j为prophet模型有S个变点的一个时刻，j＝1,...,S；
a(t)为自定义向量且
Figure PCTCN2020122552-appb-000001
T为转置运算符。
趋势项的拟合函数通过明确定义允许增长率变化的变换点，将增长模型中的趋势变化纳入趋势项拟合函数中，具体通过以下方法推导得到：假设分量序列y(t)有S个变点在时刻s j，j＝1,...,S。定义一个增长率向量
Figure PCTCN2020122552-appb-000002
δ j是在时刻s j速率发生的变化。任何时间t的增长速率为基准速率k加上所有到该时间点的速率变化值：
Figure PCTCN2020122552-appb-000003
定义一个向量a(t)∈{0,1} S，等式如下
Figure PCTCN2020122552-appb-000004
则任何时间t的增长速率简写为：k+a(t) Tδ。当速率k调整之后偏移量m还必须调整以连接段端点。变换点j正确的调整值计算为：
Figure PCTCN2020122552-appb-000005
在一个示例性实施例中，季节项s(t)的拟合函数为傅立叶函数，具体为：
Figure PCTCN2020122552-appb-000006
式中，β～(0,σ 2)，即β服从正态分布，且该正态分布的数学期望为0，方差为σ 2；
而噪声项∈ t代表模型无法捕捉的变化，且假设噪声项服从正态分布。
步骤205：累加所述趋势项的预测结果、所述季节项的预测结果以及所述噪声项，得到每个所述分量序列的分量预测结果。
具体地说，通过步骤203求得趋势项的预测结果
Figure PCTCN2020122552-appb-000007
季节项的预测结果
Figure PCTCN2020122552-appb-000008
噪声项∈ t取一个服从正态分布的随机值，累加
Figure PCTCN2020122552-appb-000009
和∈ t得到单个分量序列的分量预测结果，即：
Figure PCTCN2020122552-appb-000010
式中，
Figure PCTCN2020122552-appb-000011
为分量预测结果、
Figure PCTCN2020122552-appb-000012
为趋势项的预测结果、
Figure PCTCN2020122552-appb-000013
为季节项的预 测结果、∈ t为噪声项。
步骤206：累加所有所述分量预测结果，得到第二预设时间段的流量预测结果。
步骤206与第一实施方式中步骤104大致相同，为避免重复，此处不再赘述。
需要说明的是，当历史时期流量数据中，覆盖的节假日较少时(比如说流量数据的天数小于一年的天数、所覆盖的节假日数小于一年的节假日数)时，采用本方法实施例可以获得较佳预测效果。
与第一实施方式相比，本实施方式在采用时间序列模型分别处理所述各分量序列、得到所述各分量序列的分量预测结果时，具体采用prophet模型将各分量序列分解为趋势项、季节项、噪声项之和并进行拟合，进而利用拟合后的prophet模型求出流量预测结果。通过采用prophet模型对分量序列进行预测，能够同时根据流量数据周期性变化的趋势和非周期性变化的趋势进行流量预测，提升了流量预测的准确度。
本申请的第三实施方式涉及一种流量预测方法。第三实施方式与第二施方式大致相同，不同之处在于，第三实施方式在将每个所述分量序列分解为趋势项、季节项、噪声项之和这一步骤中，将每个分量序列分解为趋势项、季节项、节假日项、噪声项之和；并在之后的步骤中，分别确定拟合后的趋势项、季节项和节假日项，再利用拟合后的趋势项、季节项和节假日项得到第二预设时间段的趋势项、季节项和节假日项的预测结果；累加趋势项预测结果、季节项预测结果、节假日项的预测结果以及噪声项，得到第二预设时间段的每个分量序列的分量预测结果。
本实施方式中的流量预测方法如图4所示，具体包括：
步骤301：获取历史时期内第一预设时间段的流量数据，并对所述流量数据进行预处理。
步骤302：对预处理后的历史流量数据进行经验模态分解，得到多个分量序列。
步骤301至步骤302分别与第一实施方式中步骤101至步骤102大致相同， 为避免重复，在此不再一一赘述。
步骤303：将每个所述分量序列分解为趋势项、季节项、节假日项、噪声项之和；
在本实施方式中，采用prophet模型分别预测各分量序列、得到各分量序列的分量预测结果。
具体地说，先通过STL分解，将每个所述分量序列分解为趋势项、季节项、噪声项之和，即
y(t)＝g(t)+s(t)+h(t)+∈ t
式中，t为时间、y(t)为分量序列、g(t)为趋势项，s(t)为季节项、∈ t为噪声项。
步骤304：分别确定拟合后的趋势项、拟合后的季节项和拟合后的节假日项，再利用拟合后的趋势项、拟合后的季节项和拟合后的节假日项得到第二预设时间段的趋势项预测结果、季节项预测结果和节假日项预测结果。
具体地说，趋势项、季节项以及噪声项的拟合函数与第二实施方式步骤204中类似，在此不作赘述。
在一个示例性实施例中，节假日项h(t)的拟合函数如下：
h(t)＝Z(t)κ
式中，κ为正态分布曲线、Z(t)＝[1(t∈D 1),…,1(t∈D L)]，对于第L个节假日而言，D L表示该节假日产生影响的时间段。
步骤305：累加所述趋势项的预测结果、所述季节项的预测结果、所述节假日项的预测结果以及所述噪声项，得到每个所述分量序列的分量预测结果。
具体地说，通过步骤303求得趋势项的预测结果
Figure PCTCN2020122552-appb-000014
季节项的预测结果
Figure PCTCN2020122552-appb-000015
以及节假日项的预测结果
Figure PCTCN2020122552-appb-000016
噪声项∈ t取一个服从正态分布的随机值，累加
Figure PCTCN2020122552-appb-000017
和∈ t得到单个分量序列的分量预测结果，即：
Figure PCTCN2020122552-appb-000018
式中，
Figure PCTCN2020122552-appb-000019
为分量预测结果、
Figure PCTCN2020122552-appb-000020
为趋势项的预测结果、
Figure PCTCN2020122552-appb-000021
为季节项的预测结果、
Figure PCTCN2020122552-appb-000022
为节假日项的预测结果、∈ t为噪声项。
步骤306：累加所有所述分量预测结果，得到流量预测结果。
步骤306与第一实施方式中步骤104大致相同，为避免重复，此处不再赘述。
需要说明的是，当历史时期流量数据中，覆盖的节假日较多时(比如说流量数据的天数大于一年的天数、所覆盖的节假日数大于一年的节假日数)时，采用本方法实施例可以获得较佳预测效果。
与第二实施方式相比，本实施方式在采用prophet模型拟合分量序列时，将各分量序列分解为趋势项、季节项、节假日项和噪声项之和并进行拟合，进而根据拟合后的prophet模型求出流量预测结果。通过在采用prophet模型时保留prophet模型的节假日项，能够同时根据流量数据周期性变化的趋势、非周期性变化的趋势以及节假日对流量数据的影响进行流量预测，提升了流量预测的准确度。
本申请的第四实施方式涉及一种流量预测方法。第四实施方式与第一施方式大致相同，不同之处在于，第四实施方式在采用时间序列预测模型拟合所述多个分量序列这一步骤中，具体包括：根据预设步长，将所有分量序列划分为分量训练集和分量测试集；采用时间序列预测模型拟合分量训练集的分量序列；根据分量测试集，确定拟合后的时间序列预测模型的预测误差。
本实施方式中的流量预测方法如图5所示，具体包括：
步骤401：获取历史时期内第一预设时间段的流量数据，并对所述流量数据进行预处理。
步骤402：对预处理后的流量数据进行经验模态分解，得到多个分量序列。
步骤401至步骤402分别与第一实施方式中步骤101至步骤102大致相同，为避免重复，在此不再一一赘述。
步骤403：根据预设步长，将所有所述分量序列划分为分量训练集和分量测试集；
在一示例性实施例中，在本实施方式中，根据预设步长将所有分量序列划分为分量训练集和分量测试集，可以具体为：将所述第二预设时间段的时长作为预设步长；将所有所述分量序列中，距离当前时间最近的预设步长外的数据的集合作为分量训练集；将所有所述分量序列中，距离当前时间最近的预设步 长内的数据的集合作为分量测试集。
在实际应用场景中，假设要预测某基站管理下多个小区未来30天的流量数据，并对采集到的该批小区过去210天的历史流量数据进行了EMD分解。在本步骤中，由于目标预测步长为30天，将所有小区过去210天的最后30天的分量序列作为分量测试集，剩余的所有小区前180天的分量序列作为分量训练集。
步骤404：采用时间序列预测模型拟合所述分量训练集的分量序列。
本步骤与第一实施方式中步骤103大致相同，不同之处在于，第一实施方式步骤103中采用时间序列预测模型拟合各分量序列，本步骤中采用时间序列预测模型拟合分量训练集中的各分量序列，并得到拟合后的时间序列预测模型。
步骤405：根据所述分量测试集，确定所述拟合后的时间序列预测模型的预测误差。
在一示例性实施例中，采用步骤403对分量序列进行划分得到分量训练集和分量测试集，此时，根据所述拟合后的时间序列预测模型，确定距离当前时间最近的历史时间段的流量预测结果，其中，所述历史时间段的时长为所述预设步长；根据所述分量测试集和所述历史时间段的流量预测结果，确定所述拟合后的时间序列预测模型的预测误差。
在一示例性实施例中，通过计算平均绝对百分比误差(MAPE)来评估预测效果：
Figure PCTCN2020122552-appb-000023
式中，
Figure PCTCN2020122552-appb-000024
为历史时间段的流量预测结果，y t为分量测试集对应的分量序列。
步骤406：利用拟合后的时间序列预测模型得到第二预设时间段的多个分量预测结果，累加所有所述分量预测结果，得到第二预设时间段的流量预测结果。
具体地说，利用步骤404得到的拟合后的时间序列预测模型，预测第二预设时间段并得到多个分量预测结果，其余步骤中与第一实施方式中步骤104大致相同，为避免重复，此处不再赘述。
与第一实施方式相比，本实施方式将经过EMD分解得到多个分量序列划分为分量训练集和分量测试集，分量训练集用于确定拟合后的时间序列预测模 型，分量测试集用于评估分量训练集拟合后的时间序列预测模型的预测准确度。通过交叉验证的方式评估时间序列预测模型的预测准确度，从而可以在模型预测准确度较低的情况下重新训练模型，保证预测准确度达到理想状态。
上面各种方法的步骤划分，只是为了描述清楚，实现时可以合并为一个步骤或者对某些步骤进行拆分，分解为多个步骤，只要包括相同的逻辑关系，都在本专利的保护范围内；对算法中或者流程中添加无关紧要的修改或者引入无关紧要的设计，但不改变其算法和流程的核心设计都在该专利的保护范围内。
本申请第五实施方式涉及一种流量预测装置，如图6所示，包括至少一个处理器501；以及，与至少一个处理器501通信连接的存储器502；其中，存储器502存储有可被至少一个处理器501执行的指令，指令被至少一个处理器501执行，以使至少一个处理器501能够执行上述流量预测方法实施例。
其中，存储器502和处理器501采用总线方式连接，总线可以包括任意数量的互联的总线和桥，总线将一个或多个处理器501和存储器502的各种电路连接在一起。总线还可以将诸如***设备、稳压器和功率管理电路等之类的各种其他电路连接在一起，这些都是本领域所公知的，因此，本文不再对其进行进一步描述。总线接口在总线和收发机之间提供接口。收发机可以是一个元件，也可以是多个元件，比如多个接收器和发送器，提供用于在传输介质上与各种其他装置通信的单元。经处理器501处理的数据通过天线在无线介质上进行传输，在一示例性实施例中，天线还接收数据并将数据传送给处理器501。
处理器501负责管理总线和通常的处理，还可以提供各种功能，包括定时、***接口、
电压调节、电源管理以及其他控制功能。而存储器502可以被用于存储处理器501在执行操作时所使用的数据。
本申请的实施方式还提供了一种计算机可读存储介质，存储有计算机程序，所述计算机程序被处理器执行时实现上述流量预测方法实施例。
即，本领域技术人员可以理解，实现上述实施例方法中的全部或部分步骤是可以通过程序来指令相关的硬件来完成，该程序存储在一个存储介质中，包括若干指令用以使得一个设备(可以是单片机，芯片等)或处理器(processor) 执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
本领域的普通技术人员可以理解，上述各实施方式是实现本申请的具体实施例，而在实际应用中，可以在形式上和细节上对其作各种改变，而不偏离本申请的精神和范围。

Claims (10)

  1. 一种流量预测方法，包括：
    获取历史时期内第一预设时间段的流量数据，并对所述流量数据进行预处理；
    对预处理后的流量数据进行经验模态分解，得到多个分量序列；
    采用时间序列预测模型拟合所述多个分量序列，利用拟合后的时间序列预测模型得到第二预设时间段的多个分量预测结果；
    累加所有所述分量预测结果，得到第二预设时间段的流量预测结果。
  2. 根据权利要求1所述的流量预测方法，其中，所述获取历史时期内第一预设时间段的流量数据包括：
    获取历史时期内第一预设时间段的每一天的流量数据，其中，所述每一天的流量数据为当天24小时内物理资源块利用率最大的时刻的流量值。
  3. 根据权利要求1所述的流量预测方法，其中，对所述流量数据进行预处理，包括：
    采用箱线图法处理所述流量数据中的异常数据。
  4. 根据权利要求1所述的流量预测方法，其中，所述采用时间序列预测模型拟合所述多个分量序列，利用拟合后的时间序列预测模型得到第二预设时间段的多个分量预测结果，包括：
    将每个所述分量序列分解为趋势项、季节项以及噪声项之和；
    分别确定拟合后的趋势项和拟合后的季节项，再利用拟合后的趋势项和拟合后的季节项得到第二预设时间段的趋势项预测结果和季节项预测结果；
    累加所述趋势项预测结果、所述季节项预测结果以及所述噪声项，得到第二预设时间段的每个所述分量序列的分量预测结果。
  5. 根据权利要求1所述的流量预测方法，其中，所述采用时间序列预测模型拟合所述多个分量序列，利用拟合后的时间序列预测模型得到第二预设时间段的多个分量预测结果，包括：
    将每个所述分量序列分解为趋势项、季节项、节假日项以及噪声项之 和；
    分别确定拟合后的趋势项、拟合后的季节项和拟合后的节假日项，再利用拟合后的趋势项、拟合后的季节项以及拟合后的节假日项，得到第二预设时间段的趋势项预测结果、季节项预测结果以及节假日项预测结果；
    累加所述趋势项的预测结果、所述季节项的预测结果、所述节假日项的预测结果以及所述噪声项，得到第二预设时间段的每个所述分量序列的分量预测结果。
  6. 根据权利要求1所述的流量预测方法，其中，所述采用时间序列预测模型拟合所述多个分量序列，包括：
    根据预设步长，将所有所述分量序列划分为分量训练集和分量测试集；
    采用时间序列预测模型拟合所述分量训练集的分量序列；
    根据所述分量测试集，确定所述拟合后的时间序列预测模型的预测误差。
  7. 根据权利要求6所述的流量预测方法，其中，所述根据预设步长，将所有所述分量序列划分为分量训练集和分量测试集，包括：
    将所述第二预设时间段的时长作为预设步长；
    将所有所述分量序列中，距离当前时间最近的预设步长外的数据的集合作为分量训练集；
    将所有所述分量序列中，距离当前时间最近的预设步长内的数据的集合作为分量测试集。
  8. 根据权利要求7所述的流量预测方法，其中，所述根据所述分量测试集，确定所述拟合后的时间序列预测模型的预测误差，包括：
    根据所述拟合后的时间序列预测模型，确定距离当前时间最近的历史时间段的流量预测结果，其中，所述历史时间段的时长为所述预设步长；
    根据所述分量测试集和所述历史时间段的流量预测结果，确定所述拟合后的时间序列预测模型的预测误差。
  9. 一种流量预测装置，包括：
    至少一个处理器；以及，
    与至少一个处理器通信连接的存储器；其中，
    存储器存储有可被至少一个处理器执行的指令，指令被至少一个处理器执行，以使至少一个处理器能够执行如权利要求1至8中任一项所述的流量预测方法。
  10. 一种计算机可读存储介质，存储有计算机程序，其中，所述计算机程序被处理器执行时实现如权利要求1至8中任一项所述的流量预测方法。
PCT/CN2020/122552 2019-12-25 2020-10-21 流量预测方法、装置以及存储介质 WO2021129086A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US17/782,993 US11863397B2 (en) 2019-12-25 2020-10-21 Traffic prediction method, device, and storage medium
EP20904338.9A EP4047878A4 (en) 2019-12-25 2020-10-21 TRAFFIC PREDICTION METHOD, DEVICE AND STORAGE MEDIA

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911359122.9A CN113037531A (zh) 2019-12-25 2019-12-25 流量预测方法、装置以及存储介质
CN201911359122.9 2019-12-25

Publications (1)

Publication Number Publication Date
WO2021129086A1 true WO2021129086A1 (zh) 2021-07-01

Family

ID=76458550

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/122552 WO2021129086A1 (zh) 2019-12-25 2020-10-21 流量预测方法、装置以及存储介质

Country Status (4)

Country Link
US (1) US11863397B2 (zh)
EP (1) EP4047878A4 (zh)
CN (1) CN113037531A (zh)
WO (1) WO2021129086A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113792931A (zh) * 2021-09-18 2021-12-14 北京京东振世信息技术有限公司 数据预测方法、装置、物流货量预测方法、介质和设备
CN114978956A (zh) * 2022-04-11 2022-08-30 北京邮电大学 智慧城市网络设备性能异常突变点检测方法及装置
CN116562471A (zh) * 2023-07-10 2023-08-08 安徽中科海奥电气股份有限公司 一种基于stl数据分解的stl-sarima-gru功率预测方法

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114786190A (zh) * 2022-05-09 2022-07-22 中国联合网络通信集团有限公司 一种流量的预测方法、装置及存储介质
CN115426201B (zh) * 2022-11-03 2023-01-17 湖南大佳数据科技有限公司 一种面向网络靶场的数据采集方法和***
CN116011680A (zh) * 2023-02-20 2023-04-25 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) 一种f10.7指数的预测方法及设备
CN116320043B (zh) * 2023-05-24 2023-08-01 深圳龙电华鑫控股集团股份有限公司 一种提高多载波通信***传输效率的方法及***
CN117734392B (zh) * 2024-02-21 2024-04-16 上海万柔汽车电子有限公司 基于nfc技术的汽车香氛交互方法及***

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107026763A (zh) * 2017-06-02 2017-08-08 广东电网有限责任公司中山供电局 一种基于流量分解的数据通信网流量预测方法
US20190028537A1 (en) * 2017-07-18 2019-01-24 Wipro Limited Method and system for dynamically handling load on a computer network
CN109802862A (zh) * 2019-03-26 2019-05-24 重庆邮电大学 一种基于集合经验模态分解的组合型网络流量预测方法
CN110381523A (zh) * 2019-06-17 2019-10-25 盐城吉大智能终端产业研究院有限公司 一种基于tvf-emd-lstm模型的蜂窝基站网络流量预测方法

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102568205B (zh) * 2012-01-10 2013-12-04 吉林大学 非常态下基于经验模态分解和分类组合预测的交通参数短时预测方法
US10531251B2 (en) * 2012-10-22 2020-01-07 United States Cellular Corporation Detecting and processing anomalous parameter data points by a mobile wireless data network forecasting system
US10867421B2 (en) * 2016-02-29 2020-12-15 Oracle International Corporation Seasonal aware method for forecasting and capacity planning
CN107809737B (zh) 2016-08-29 2020-08-07 华为技术服务有限公司 一种确定基站的网络流量的方法和装置
CN107426026B (zh) * 2017-07-31 2020-05-22 山东省计算中心（国家超级计算济南中心） 一种基于eemd-arima的云计算服务器负载短期预测方法
CN109657831A (zh) * 2017-10-11 2019-04-19 顺丰科技有限公司 一种业务量预测方法、装置、设备、存储介质
CN107730887B (zh) * 2017-10-17 2020-05-15 海信集团有限公司 实现交通流量预测的方法及装置、可读存储介质
CN110175711A (zh) 2019-05-17 2019-08-27 北京市天元网络技术股份有限公司 一种基于联合lstm基站小区流量预测方法以及装置
CN110210658B (zh) * 2019-05-22 2023-10-03 东南大学 基于小波变换的Prophet与高斯过程用户网络流量预测方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107026763A (zh) * 2017-06-02 2017-08-08 广东电网有限责任公司中山供电局 一种基于流量分解的数据通信网流量预测方法
US20190028537A1 (en) * 2017-07-18 2019-01-24 Wipro Limited Method and system for dynamically handling load on a computer network
CN109802862A (zh) * 2019-03-26 2019-05-24 重庆邮电大学 一种基于集合经验模态分解的组合型网络流量预测方法
CN110381523A (zh) * 2019-06-17 2019-10-25 盐城吉大智能终端产业研究院有限公司 一种基于tvf-emd-lstm模型的蜂窝基站网络流量预测方法

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
KIT YAN CHAN ; THARAM DILLON ; ELIZABETH CHANG ; JAIPAL SINGH: "Prediction of Short-Term Traffic Variables Using Intelligent Swarm-Based Neural Networks", IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY., IEEE SERVICE CENTER, NEW YORK, NY., US, vol. 21, no. 1, 1 January 2013 (2013-01-01), US, pages 263 - 274, XP011482703, ISSN: 1063-6536, DOI: 10.1109/TCST.2011.2180386 *
LIAN YIPING: "Electronic Design Engineering 第 27 卷 Traffic flow forecast based on EMD and LSSVM combination model", ELECTRONIC DESIGN ENGINEERING, SHIJIE ZHISHI CHUBANSHE, CN, vol. 27, no. 11, 1 June 2019 (2019-06-01), CN, pages 47 - 55, XP055827154, ISSN: 1674-6236, DOI: 10.14022/j.cnki.dzsjgc.2019.11.011 *
See also references of EP4047878A4 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113792931A (zh) * 2021-09-18 2021-12-14 北京京东振世信息技术有限公司 数据预测方法、装置、物流货量预测方法、介质和设备
CN114978956A (zh) * 2022-04-11 2022-08-30 北京邮电大学 智慧城市网络设备性能异常突变点检测方法及装置
CN114978956B (zh) * 2022-04-11 2024-04-09 北京邮电大学 智慧城市网络设备性能异常突变点检测方法及装置
CN116562471A (zh) * 2023-07-10 2023-08-08 安徽中科海奥电气股份有限公司 一种基于stl数据分解的stl-sarima-gru功率预测方法
CN116562471B (zh) * 2023-07-10 2023-10-24 安徽中科海奥电气股份有限公司 一种基于stl数据分解的stl-sarima-gru功率预测方法

Also Published As

Publication number Publication date
EP4047878A1 (en) 2022-08-24
US11863397B2 (en) 2024-01-02
US20230028574A1 (en) 2023-01-26
EP4047878A4 (en) 2022-12-14
CN113037531A (zh) 2021-06-25

Similar Documents

Publication Publication Date Title
WO2021129086A1 (zh) 流量预测方法、装置以及存储介质
CN107480028B (zh) 磁盘可使用的剩余时长的获取方法及装置
KR102468316B1 (ko) 과거 데이터에 기초한 시계열 예측 방법 및 장치
CN111294812B (zh) 一种资源扩容规划的方法及***
CN109116444B (zh) 基于PCA-kNN的空气质量模式PM2.5预报方法
US11693380B1 (en) Distribution methods and distribution systems for greenspace irrigation water of smart cities based on internet of things
JP2020154785A (ja) 予測方法、予測プログラムおよびモデル学習方法
CN115829024B (zh) 一种模型训练方法、装置、设备及存储介质
US11197175B2 (en) Forcasting time series data
CN114694379B (zh) 一种基于自适应动态图卷积的交通流预测方法及***
CN117909888B (zh) 智慧人工智能气候预测方法
CN115409153A (zh) 一种基于注意力lstm的畜牧指标预测方法及预测***
CN111598457A (zh) 一种电力无线网络质量的确定方法及装置
CN114648156A (zh) 一种基于卷积神经网络的时间序列负荷预测方法
CN108509984A (zh) 激活值量化训练方法及装置
CN112423327B (zh) 一种容量预测方法及设备、存储介质
CN114154768A (zh) 天然气需求预测方法、装置、电子设备及介质
CN116528270B (zh) 一种基站节能潜力评估方法、装置、设备及存储介质
CN112686416A (zh) 一种风电功率概率预测方法和***
CN115225546B (zh) 一种网络流量的预测方法、装置及设备
CN117578428A (zh) 风机功率递归预测方法、装置、计算机设备和存储介质
CN118261628A (zh) 一种研发成本可视化管理方法及***
CN115423211A (zh) 一种街区随机功率预测方法、装置、设备及介质
CN116894528A (zh) 一种冷热站负荷预测方法、装置和计算机设备
CN115169692A (zh) 一种时间序列预测方法及相关装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20904338

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020904338

Country of ref document: EP

Effective date: 20220517

NENP Non-entry into the national phase

Ref country code: DE