CN114205853A - Flow prediction method, flow prediction model establishing method, and storage medium - Google Patents
Flow prediction method, flow prediction model establishing method, and storage medium Download PDFInfo
- Publication number
- CN114205853A CN114205853A CN202010909565.7A CN202010909565A CN114205853A CN 114205853 A CN114205853 A CN 114205853A CN 202010909565 A CN202010909565 A CN 202010909565A CN 114205853 A CN114205853 A CN 114205853A
- Authority
- CN
- China
- Prior art keywords
- value
- flow
- sequence
- prediction
- network
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 147
- 238000003860 storage Methods 0.000 title claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 106
- 230000015654 memory Effects 0.000 claims description 28
- 238000012545 processing Methods 0.000 claims description 24
- 238000000605 extraction Methods 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 7
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 3
- 210000004027 cell Anatomy 0.000 description 93
- 230000000875 corresponding effect Effects 0.000 description 50
- 230000008569 process Effects 0.000 description 19
- 238000004364 calculation method Methods 0.000 description 14
- 238000004891 communication Methods 0.000 description 9
- 230000006870 function Effects 0.000 description 9
- 238000009826 distribution Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 230000002159 abnormal effect Effects 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 230000006403 short-term memory Effects 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 230000007787 long-term memory Effects 0.000 description 4
- 238000012300 Sequence Analysis Methods 0.000 description 3
- 238000007781 pre-processing Methods 0.000 description 3
- 238000012896 Statistical algorithm Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 230000023402 cell communication Effects 0.000 description 2
- 238000010295 mobile communication Methods 0.000 description 2
- BDEDPKFUFGCVCJ-UHFFFAOYSA-N 3,6-dihydroxy-8,8-dimethyl-1-oxo-3,4,7,9-tetrahydrocyclopenta[h]isochromene-5-carbaldehyde Chemical compound O=C1OC(O)CC(C(C=O)=C2O)=C1C1=C2CC(C)(C)C1 BDEDPKFUFGCVCJ-UHFFFAOYSA-N 0.000 description 1
- 210000004460 N cell Anatomy 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008602 contraction Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 230000007257 malfunction Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 230000007723 transport mechanism Effects 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/06—Testing, supervising or monitoring using simulated traffic
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Environmental & Geological Engineering (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention provides a flow prediction method, a flow prediction model establishing method and a storage medium, wherein the flow prediction method comprises the following steps: acquiring a first characteristic value sequence and a predicted time step length, wherein the first characteristic value sequence is obtained according to a first time period and a first network flow value corresponding to the first time period; and inputting the first characteristic value sequence and the prediction time step into a traffic prediction model to obtain a network traffic prediction value within the prediction time step, wherein the traffic prediction model is obtained by training a deep autoregressive model according to training set data, and the training set data comprises a plurality of cell identification information, a plurality of second time periods corresponding to the cell identification information and second network traffic values corresponding to the second time periods one by one. The deep autoregressive model is adopted for training, and can be conveniently brought into additional features and can be used for predicting the future network traffic of the cell more accurately, so that the accuracy of network traffic prediction can be improved.
Description
Technical Field
Embodiments of the present invention relate to, but not limited to, the field of information processing technologies, and in particular, to a traffic prediction method, a traffic prediction model establishment method, a traffic prediction apparatus, a traffic prediction model establishment apparatus, a traffic prediction system, and a computer-readable storage medium.
Background
With the advent of 5G networks, the demand for wireless traffic will show a more diversified trend, which will result in a large amount of wirelessly transmitted traffic data. And traffic prediction of a base station plays an important role in wireless communication technology. The quality of user experience is considered as one of the most prominent evaluation indexes, and the quality of service is gradually replaced to measure the user experience. In order to ensure the experience quality of users, network equipment providers pay more attention to traffic prediction and traffic trend, and guide operators to dynamically expand and contract base station cell communication equipment through prediction of base station cell traffic. However, the existing base station cell flow prediction method often depends on data stationarity highly, is not good at predicting long period time sequences, or lacks adaptability, so that the existing prediction method is difficult to accurately predict the base station cell flow, thereby directly influencing the expansion and contraction capacity of operators on base station cell communication equipment, and further influencing the user service level and quality.
Disclosure of Invention
The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.
The embodiment of the invention provides a flow prediction method, a flow prediction model establishing method, a flow prediction device, a flow prediction model establishing device, a flow prediction system and a computer readable storage medium, which can improve the accuracy of network flow prediction.
In a first aspect, an embodiment of the present invention provides a traffic prediction method, including:
acquiring a first characteristic value sequence and a predicted time step length, wherein the first characteristic value sequence is obtained according to a first time period and a first network flow value corresponding to the first time period;
inputting the first characteristic value sequence and the prediction time step into a traffic prediction model to obtain a network traffic prediction value within the prediction time step, wherein the traffic prediction model is obtained by training a deep autoregressive model according to training set data, and the training set data comprises a plurality of cell identification information, a plurality of second time periods corresponding to the cell identification information and second network traffic values corresponding to the second time periods in a one-to-one mode.
In a second aspect, an embodiment of the present invention further provides a method for establishing a traffic prediction model, including:
acquiring training set data, wherein the training set data comprises a plurality of cell identification information, a plurality of second time periods corresponding to the cell identification information and second network flow values corresponding to the second time periods in a one-to-one manner;
processing the second time period and the second network flow value to obtain an initial sequence corresponding to the training set data;
performing feature extraction on the initial sequence to obtain a second characteristic value sequence;
and training a depth autoregressive model by adopting the initial sequence and the second characteristic value sequence to obtain a flow prediction model.
In a third aspect, an embodiment of the present invention further provides a traffic prediction apparatus, including:
the system comprises a first acquisition unit, a second acquisition unit and a time step prediction unit, wherein the first acquisition unit is used for acquiring a first characteristic value sequence and a predicted time step, and the first characteristic value sequence is obtained according to a first time period and a first network flow value corresponding to the first time period;
and the traffic prediction unit is used for inputting the first characteristic value sequence and the prediction time step into a traffic prediction model to obtain a network traffic prediction value within the prediction time step, wherein the traffic prediction model is obtained by training a deep autoregressive model according to training set data, and the training set data comprises a plurality of cell identification information, a plurality of second time periods corresponding to the cell identification information and second network traffic values corresponding to the second time periods in a one-to-one mode.
In a fourth aspect, an embodiment of the present invention further provides a device for establishing a flow prediction model, where the device includes:
a second obtaining unit, configured to obtain training set data, where the training set data includes multiple cell identification information, multiple second time periods corresponding to the cell identification information, and second network traffic values corresponding to the second time periods in a one-to-one manner;
the processing unit is used for processing the second time period and the second network flow value to obtain an initial sequence corresponding to the training set data;
the characteristic extraction unit is used for extracting the characteristics of the initial sequence to obtain a second characteristic value sequence;
and the training unit is used for training the depth autoregressive model by adopting the initial sequence and the second characteristic value sequence to obtain a flow prediction model.
In a fifth aspect, an embodiment of the present invention further provides a flow prediction system, including: a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the flow prediction method according to the first aspect when executing the computer program, or implementing the flow prediction model building method according to the second aspect when executing the computer program.
In a sixth aspect, the present invention further provides a computer-readable storage medium storing computer-executable instructions for performing the flow prediction method according to the first aspect, or performing the flow prediction model building method according to the second aspect.
The embodiment of the invention comprises the following steps: acquiring a first characteristic value sequence and a predicted time step length, wherein the first characteristic value sequence is obtained according to a first time period and a first network flow value corresponding to the first time period; and then inputting the first characteristic value sequence and the prediction time step into a traffic prediction model to obtain a network traffic prediction value within the prediction time step, wherein the traffic prediction model is obtained by training a deep autoregressive model according to training set data, and the training set data comprises a plurality of cell identification information, a plurality of second time periods corresponding to the cell identification information and second network traffic values corresponding to the second time periods in a one-to-one mode. According to the technical scheme of the embodiment of the invention, the deep autoregressive model is adopted to train according to the cell identification information, the second time period and the second network flow value in the training set data, and as the deep autoregressive model can conveniently incorporate extra features and can more accurately predict the future network flow of the cell, the embodiment of the invention can improve the accuracy of network flow prediction.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the example serve to explain the principles of the invention and not to limit the invention.
FIG. 1 is a schematic diagram of a system architecture platform for implementing a traffic prediction model building method or a traffic prediction method according to an embodiment of the present invention;
FIG. 2 is a flow chart of a method for flow prediction modeling according to an embodiment of the present invention;
FIG. 3 is a flow chart of a method for flow prediction modeling according to another embodiment of the present invention;
FIG. 4 is a flow chart of a method for flow prediction modeling according to another embodiment of the present invention;
FIG. 5 is a flow chart of a method for flow prediction modeling according to another embodiment of the present invention;
FIG. 6 is a flow chart of a method for flow prediction modeling according to another embodiment of the present invention;
FIG. 7 is a flow chart of a method for flow prediction modeling according to another embodiment of the present invention;
FIG. 8 is a flow chart of a method for flow prediction modeling according to another embodiment of the present invention;
fig. 9 is a specific flowchart for extracting the second eigenvalue sequence and the first eigenvalue sequence according to an embodiment of the present invention;
FIG. 10 is a flow chart of a method for flow prediction modeling according to another embodiment of the present invention;
FIG. 11 is a detailed flow chart of gate calculations for the long-short term memory neural network according to an embodiment of the present invention;
FIG. 12 is a detailed flow chart for training a model according to an embodiment of the present invention;
FIG. 13 is a flow chart of a flow prediction method provided by one embodiment of the present invention;
FIG. 14 is a detailed flow chart of predicting network traffic provided by one embodiment of the present invention;
FIG. 15 is a flow chart of a flow prediction method according to another embodiment of the present invention;
FIG. 16 is a schematic diagram of a flow prediction model building apparatus according to an embodiment of the present invention;
fig. 17 is a schematic diagram of a flow prediction apparatus according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
It should be noted that although functional blocks are partitioned in a schematic diagram of an apparatus and a logical order is shown in a flowchart, in some cases, the steps shown or described may be performed in a different order than the partitioning of blocks in the apparatus or the order in the flowchart. The terms "first," "second," and the like in the description, in the claims, or in the drawings described above, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order.
In the related art, the prediction of network traffic for a base station cell for a future period of time depends on historical network traffic data for different base station cells. Due to the development of wireless technology and different daily habits, people have different internet access requirements, so that infinite traffic data presents different distribution characteristics, which increases the complexity of a model or an algorithm. At present, for a time series model for predicting base station cell traffic, for example, statistical modeling methods such as an AR (Autoregressive) model, an MA (Moving Average) model, an ARMA (Auto-Regressive Moving Average) model, and a mixture of some joint wavelet decompositions, etc., although short-term prediction can be well performed on a unary time series in a strongly correlated scene. However, the above existing algorithm models often have some problems, such as: the ARMA, AR or MA model has higher dependence on data stationarity, is not good at predicting long-period time sequences, neglects certain trends of week time and the like; in addition, wavelet analysis requires wavelet basis functions, the selection of which has a great influence on the results of the entire wavelet analysis, and once the wavelet basis functions are determined, they cannot be replaced in the entire analysis process, even though the wavelet basis functions may be optimal globally but may not be optimal locally, so the wavelet basis functions of wavelet analysis lack adaptability.
Based on the above situation, the present invention provides a traffic prediction method, a traffic prediction model building method, a traffic prediction apparatus, a traffic prediction model building apparatus, a traffic prediction system, and a computer-readable storage medium, wherein the traffic prediction model building method includes: firstly, acquiring training set data, wherein the training set data comprises a plurality of cell identification information, a plurality of second time periods corresponding to the cell identification information and second network flow values corresponding to the second time periods one by one; then processing the second time period and the second network flow value to obtain an initial sequence corresponding to the training set data; then, carrying out feature extraction on the initial sequence to obtain a second characteristic value sequence; and finally, training the depth autoregressive model by adopting the initial sequence and the second characteristic value sequence to obtain a flow prediction model.
In addition, after the traffic prediction model is established, a traffic prediction method is further required to be adopted to predict the network traffic of the base station cell for a period of time in the future, and the traffic prediction method includes: firstly, acquiring a first characteristic value sequence and a predicted time step length, wherein the first characteristic value sequence is obtained according to a first time period and a first network flow value corresponding to the first time period; and then inputting the first characteristic value sequence and the prediction time step into a traffic prediction model to obtain a network traffic prediction value within the prediction time step, wherein the traffic prediction model is obtained by training a deep autoregressive model according to training set data, and the training set data comprises a plurality of cell identification information, a plurality of second time periods corresponding to the cell identification information and second network traffic values corresponding to the second time periods in a one-to-one mode.
For the technical scheme of the embodiment of the invention, firstly, the embodiment of the invention can process the second time period and the second network flow value, thereby reducing the influence of abnormal data in the training set data on the prediction result; secondly, the embodiment of the invention can establish a uniform flow prediction model for all cells, thereby improving the convenience of the flow prediction model during the actual use period; in addition, the embodiment of the invention can effectively distinguish the cells by selecting the characteristic values of the cells, thereby having good prediction accuracy; finally, the deep autoregressive model is adopted to train according to the cell identification information, the second time period and the second network flow value in the training set data, and because the deep autoregressive model can conveniently incorporate extra features and can more accurately predict the future network flow of the cell, the accuracy of network flow prediction can be improved.
The embodiments of the present invention will be further explained with reference to the drawings.
As shown in fig. 1, fig. 1 is a schematic diagram of a system architecture platform for performing a traffic prediction model building method or a traffic prediction method according to an embodiment of the present invention.
In the example of fig. 1, the system architecture platform includes a traffic prediction system 100, wherein the traffic prediction system 100 is provided with a processor 110 and a memory 120, wherein the processor 110 and the memory 120 may be connected by a bus or by other means, and fig. 1 takes the example of being connected by a bus as an example.
The memory 120, which is a non-transitory computer readable storage medium, may be used to store non-transitory software programs as well as non-transitory computer executable programs. Further, the memory 120 may include high speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, memory 120 optionally includes memory located remotely from processor 110, which may be connected to the system architecture platform via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
It can be understood by those skilled in the art that the system architecture platform can be applied to a 3G communication network system, an LTE communication network system, a 5G communication network system, a mobile communication network system that is evolved later, and the like, and this embodiment is not limited in particular.
Those skilled in the art will appreciate that the system architecture platform illustrated in FIG. 1 does not constitute a limitation on embodiments of the invention, and may include more or fewer components than those illustrated, or some components may be combined, or a different arrangement of components.
In the system architecture platform shown in fig. 1, the processor 110 may call a flow prediction model building program stored in the memory 120 to perform a flow prediction model building method. Alternatively, the processor 110 may call a flow prediction program stored in the memory 120 to perform the flow prediction method.
Based on the above system architecture platform, the following provides various embodiments of the traffic prediction model establishment method of the present invention.
As shown in fig. 2, fig. 2 is a flowchart of a flow prediction model building method according to an embodiment of the present invention, which includes, but is not limited to, step S100, step S200, step S300, and step S400.
Step S100, training set data is obtained, wherein the training set data comprises a plurality of cell identification information, a plurality of second time periods corresponding to the cell identification information, and second network flow values corresponding to the second time periods in a one-to-one mode.
In an embodiment, the present invention may collect, as training set data, real network traffic data collected by a communication operator at different times in each cell, where the training set data includes cell identification information of each cell and second network traffic values corresponding to a plurality of different second time periods in each cell.
It should be noted that one cell may be correspondingly provided with one or more base station devices, and therefore, the embodiment of the present invention may acquire a network traffic value from the base station device as the network traffic value of the cell.
For each cell, for example, the communications operator may select, as the downlink data traffic of the cell on the current day, the network traffic at a time instant when the utilization rate of a PRB (Physical Resource Block) is the maximum among multiple time instants per day, and use the downlink data traffic as the network traffic value of the cell on the current day; and repeating the operation to obtain the network flow value of each cell for a plurality of continuous days.
It should be noted that, in the actual collection process, for example, cell traffic from 1000 collected base stations provided by a communication operator is obtained, since data is stored in multiple files, data of a data set needs to be processed uniformly and merged into one file, and after the collected real data traffic files are merged, field attribute information of each data traffic includes, but is not limited to, a collection traffic date, a collection area number, and data traffic.
It should be understood that the network traffic value may be a wireless data traffic value, a wired data traffic value, or a sum of the wireless data traffic value and the wired data traffic value. The cell identification information is not limited to number information or character information.
And step S200, processing the second time period and the second network flow value to obtain an initial sequence corresponding to the training set data.
In an embodiment, because the second time period and the second network traffic value in the acquired training set data may be abnormal data or useless data that does not contribute to subsequent training and prediction, the second time period and the second network traffic value in the training set data need to be preprocessed, so that the influence of the abnormal data in the training set data on the subsequent training and prediction can be reduced, and the accuracy of the prediction result can be improved, or the useless data in the training set data can be deleted, and the data size of the training set data and the complexity of subsequent model training can be reduced. And finally, obtaining an initial sequence corresponding to the training set data according to the preprocessed second time period and the second network flow value.
And step S300, performing feature extraction on the initial sequence to obtain a second characteristic value sequence.
In an embodiment, after the initial sequence is obtained, since the initial sequence corresponds to the second time period and the second network traffic value after the preprocessing, the embodiment of the present invention may perform feature extraction on the second time period and the second network traffic value in the initial sequence, so as to obtain a second feature value sequence.
It is to be understood that, regarding the second characteristic value sequence described above, it may be one of a mean value sequence, a variance sequence, and a standard deviation sequence.
And S400, training the depth autoregressive model by adopting the initial sequence and the second characteristic value sequence to obtain a flow prediction model.
In an embodiment, after the initial sequence and the second feature value sequence are obtained, a Deep AutoregRessive (Deep AutoregRessive) model is used for training according to the initial sequence and the second feature value sequence, and compared with an LSTM (Long Short-Term Memory), the Deep AutoregRessive model can conveniently incorporate additional features, so that compared with a traditional statistical algorithm model, the Deep AutoregRessive model used in the embodiment of the present invention can more accurately predict the future network traffic of a cell.
In an embodiment, since the traffic prediction method includes the above step S100, step S200, step S300, and step S400, first, the embodiment of the present invention can process the second time period and the second network traffic value, so as to reduce the influence of abnormal data in the training set data on the prediction result; secondly, the embodiment of the invention can establish a uniform flow prediction model for all cells, thereby improving the convenience of the flow prediction model during the actual use period; in addition, the embodiment of the invention can effectively distinguish the cells by selecting the characteristic values of the cells, thereby having good prediction accuracy; finally, the deep autoregressive model is adopted to train according to the cell identification information, the second time period and the second network flow value in the training set data, and the deep autoregressive model can conveniently incorporate extra features and can more accurately predict the future network flow of the cell, so that the accuracy of network flow prediction can be improved.
It should be noted that, regarding the processing of the second time period and the second network traffic value in the above step S200, it may be: removing some cells containing null values, adopting a box-line graph method to carry out self-adaptive adjustment on the second network flow value, and deleting the second network flow value in holidays.
In an embodiment, regarding the above-mentioned removing some cells with null values, for example, after collecting cell traffic of 1000 base stations, the cells may be uniquely identified by uniformly numbering the cells, and the numbering of the cells is recorded by adding a num column to the data set. Some cells containing nulls are then removed, for example: some cells contain a field Nan (null). Finally, all cell number intervals are 0 to 841, and thus relevant traffic data of 842 cells is obtained.
In addition, the present invention provides a plurality of embodiments as described in detail below with respect to the above-mentioned adaptive adjustment of the second network traffic value using the box-plot method and the deletion of the second network traffic value for holidays.
As shown in fig. 3, fig. 3 is a flowchart of a flow prediction model building method according to another embodiment of the present invention. In an embodiment, regarding the processing of the second time period and the second network traffic value in step S200, the processing includes, but is not limited to, step S510 and step S520.
Step S510, processing the second network flow value through a box diagram method to obtain an upper flow limit value and a lower flow limit value;
and step S520, adjusting the second network flow value according to the flow upper limit value and the flow lower limit value to obtain the adjusted second network flow value.
In one embodiment, due to some unavoidable malfunctions in the robotic or human acquisition process, some individual outliers that deviate significantly from the normal values, i.e., the above-mentioned abnormal data, may be acquired. Due to the existence of outliers, influences of different degrees are generally generated in the subsequent initial sequence analysis, and the prediction accuracy of the established flow prediction model is further influenced. Therefore, the embodiment of the invention adopts a box line method to process the second network flow value, so as to obtain the upper flow limit value and the lower flow limit value, and adaptively adjust the second network flow value corresponding to each cell in each day according to the obtained upper flow limit value and the obtained lower flow limit value.
Specifically, step S520 described above includes, but is not limited to, the method steps as in fig. 4 or fig. 5.
As shown in fig. 4, fig. 4 is a flowchart of a flow prediction model building method according to another embodiment of the present invention. In an embodiment, the step S520 includes, but is not limited to, the step S610.
And step S610, when the second network flow value is larger than the flow upper limit value, adjusting the second network flow value to be the average value of the sum of the second network flow value and the flow upper limit value.
In an embodiment, after the flow upper limit value is obtained by a box line method, if the second network flow value of a certain day is greater than the flow upper limit value, the second network flow value of the certain day is updated to an average value of the sum of the original second network flow value and the flow upper limit value, so that an outlier exceeding the flow upper limit value can be adjusted, the influence degree generated in subsequent initial sequence analysis is reduced, and the prediction accuracy of the established flow prediction model is improved.
As shown in fig. 5, fig. 5 is a flowchart of a flow prediction model building method according to another embodiment of the present invention. In an embodiment, the step S520 includes, but is not limited to, the step S620.
And step S620, when the second network flow value is smaller than the flow lower limit value, adjusting the second network flow value to be the average value of the sum of the second network flow value and the flow lower limit value.
In an embodiment, after the flow lower limit value is obtained by a box line method, if the second network flow value of a certain day is smaller than the flow lower limit value, the second network flow value of the certain day is updated to an average value of the sum of the original second network flow value and the flow lower limit value, so that an outlier lower than the flow lower limit value can be adjusted, the influence degree generated in subsequent initial sequence analysis is reduced, and the prediction accuracy of the established flow prediction model is improved.
It will be appreciated that if the second network traffic value for a day is between the upper and lower traffic limits, no adjustment is made to the second network traffic value for that day.
Based on the above embodiments of fig. 4 and 5, another exemplary embodiment of the present invention provides a specific flowchart for adaptively adjusting the second network traffic value by a box-line graph method, as shown in fig. 6, the method includes, but is not limited to, step S710, step S720, step S730, step S740, step C110, step S750, step C120, step S760, step C130, step S770, and step S780.
Step S710, training set data is obtained and processed to obtain a sequence { z }i,tH, wherein i represents the number of cells and t represents the number of days of a certain day;
step S720, calculating an upper quartile Q3 and a lower quartile Q1;
step S730, calculating a quartile range IQR-Q3-Q1;
step S740, calculating a flow upper limit top Q3+1.1 i qr and a flow lower limit low Q1-1.1 i qr, and executing step C110;
step C110, judging zi,tWhether it is greater than upper limit top of flow when zi,tIf the value is more than top, executing the step S750, otherwise executing the step C120;
step S750, update zi,t＝(top+zi,t)/2；
Step C120, judging zi,tIf it is less than the lower limit value of flow low, when zi,tIf the value is less than top, executing step S760, otherwise executing step C130;
step S760, updating zi,t＝(low+zi,t)/2；
Step C130, judging whether T is equal to T or not, if T is not equal to T, executing step S770, otherwise executing step S780, wherein T is the total number of collected days;
step S770, adjusting t to t +1, and executing step C110;
step S780 ends.
In one embodiment, for example, if the wireless data traffic in the area has 175 days in total, the outlier of the data traffic is processed by using a box plot method, and the algorithm is described as follows: the input is training set data Tdata [ i ] { i ═ 1,2,3 …,175}, total days 175, and the output is data result [ i ] { i ═ 1,2,3, …,175}, which is adaptively adjusted by a box-line graph method, and the specific process is as follows:
as shown in fig. 7, fig. 7 is a flowchart of a flow prediction model building method according to another embodiment of the present invention. In an embodiment, regarding the processing of the second time period and the second network traffic value in step S200, the processing includes, but is not limited to, step S810 and step S820.
Step S810, screening out a second time period for representing holidays and a second network flow value corresponding to the second time period for representing holidays from the second time period;
and step S820, deleting the second time period for representing the holiday and the second network flow value corresponding to the second time period for representing the holiday from the training set data.
In an embodiment, since the cell base station traffic usage is an appearance of the production activities of people, the activity areas of people and the network traffic usage during holidays are changed more greatly than non-holidays from the experience of people activities, and if the time interval to be predicted is only a normal working day, the cell traffic usage of the holiday base station in the training set data interferes with the prediction, so that the holiday interval in the training set data needs to be eliminated in this case. Therefore, when the holidays only appear in the training set data, the holiday intervals in the training set data are removed, otherwise, the holiday intervals are not removed.
As shown in fig. 8, fig. 8 is a flowchart of a flow prediction model building method according to another embodiment of the present invention. In an embodiment, the step S300 includes, but is not limited to, the step S900.
And step S900, performing feature extraction on the initial sequence to obtain a second feature value corresponding to the initial sequence, and calculating the second feature value in a moving average mode to obtain a second feature value sequence.
In one embodiment, the preprocessed initial sequence may be further processed to obtain a second feature value sequence, which includes a mean feature sequence, a variance feature sequence, and a standard deviation feature sequence.
It should be noted that, feature extraction performed on the initial sequence may obtain a total feature value sequence, where the total feature value sequence includes the second feature value sequence and the first feature value sequence, and a time interval included in the total feature value sequence is a sum of the training set time interval and the prediction time interval. The characteristic value sequence used in the model training stage is a sequence corresponding to a training set time interval, namely a second characteristic value sequence; and the stage of outputting the prediction result uses the sequence of the characteristic value as the sequence corresponding to the prediction interval, namely the first sequence of the characteristic value. The embodiment of the invention dynamically calculates the characteristic sequence by using a local sliding mode, and the characteristic sequence calculated by the method can enable input step data and prediction step data to establish characteristic relation corresponding to an average value, a variance or a standard deviation in the model training process, so that when the characteristic sequence of a corresponding category is input in the prediction process, a result can be more accurately predicted.
Illustratively, the initial sequence z after preprocessing is acquiredi,tAfter this, where i is 1, …, N denotes the number of N cells of the experiment, T is 1, …, T denotes the time step. Then for the initial sequence zi,tExtracting features, calculating the feature sequence of each type of feature in a sliding average mode, wherein the feature value types comprise mean feature, variance feature and standard deviation feature
Where l represents the number of features, l' represents the predicted step size, and T represents the total step size of the sequence.
Based on the embodiment of FIG. 9, illustratively, for the input sequence { z }t1, …,175, the prediction step length l' is 31 for this example, the input step length is 62, the characteristic values of each class are calculated by means of moving average respectively, and a characteristic value sequence { x is obtainedt1, …,206, where l denotes the sequence ztLength of, cl represents input step size, pl represents prediction step size, and f represents eigenvalue. Wherein the algorithm is described as follows: input as an initial sequence z t1, …,175 for a total number of days, and a second sequence of eigenvalues { x } is output t1, …,206 for a total number of days 206. The specific process is as follows:
as shown in fig. 10, fig. 10 is a flowchart of a flow prediction model building method according to another embodiment of the present invention. In one embodiment, the depth autoregressive model is trained using the initial sequence and the second feature value sequence in step S400, including but not limited to step S1010 and step S1020.
Step S1010, processing the initial sequence and the second characteristic value sequence through a long-term and short-term memory neural network to obtain a hidden variable corresponding to a network layer in a deep autoregressive model;
and step S1020, performing maximum log-likelihood calculation on the hidden variables to obtain training parameters of the depth autoregressive model.
In one embodiment, a DeepAR model and an initial sequence { z ] are usedi,tAnd a second sequence of eigenvalues { x }i,tThe sequence is trained. Assuming that the target value for each time series obeys a probability distribution l (z)i,t|θi,t) Wherein thetai,t＝θ(Hi,t) Representing a probability distribution parameter. Firstly, an LSTM unit is used for calculating an implicit variable H of the current time stepit＝LSTM(Hit-1,zit-1,xit) Finally by maximizing the log-likelihood
Learning network parameters to obtain a final model, and obtaining a predicted value y of each cell in the next M days through the trained model and the predicted step length M parameteri，i＝1,…,N。
Illustratively, when the initial sequence z is obtainedi,tAnd a second sequence of eigenvalues { x }i,tThe calculation of each gate of the long-short term memory neural network is shown in fig. 11, wherein i is 1, …,842, t is 1, …,175, and the process is described in detail as follows: the inputs of the long and short term memory gates are all current time step inputs zi,tAnd last time hidden state Hi,t-1The output is calculated by the full-link layer of the sigmoid function of the activation function, so that the three gate elements are all [0,1 ]]. Assuming that the number of hidden units is h, the input of a small batch at a given time step t
Characteristic sequence
(represents the ith cell time step as t), where n represents the number of samples, the input number is d, and the previous time step hidden state
Input gate of time step t
Forgetting door
And output gate
The following are calculated respectively:
Ii,t＝σ(zi,tWzj+xi,tWxj+Hi,t-1Whj+bj)
Fi,t＝σ(zi,tWzf+xi,tWxf+Hi,t-1Whf+bf)
Oi,t＝σ(zi,tWzo+xi,tWzxo+Hi,t-1Who+bo)
wherein the content of the first and second substances,
and
is a weight parameter that is a function of,
is a bias parameter.
Next, candidate memory cells of the long-term and short-term memory neural networks are calculated
And memory cells
The calculation process is as follows: candidate memory cells and memory cells using a range of [ -1,1 [ -1]As an activation function, for the candidate memory cells of time step t
The calculation formula of (2) is as follows:
wherein
And
is a weight parameter that is a function of,
is inclined toAnd setting parameters. Then pass through
To calculate memory cell CI,t. Finally, by Hi,t＝Oi,t⊙tanh(Ci,t) The layer hidden state is calculated.
Then calculating to obtain Hi,tThe network layer is used for training the deep AR model, the model training process is shown in FIG. 12, and the training process is described as follows: assuming that the target value for each time series obeys a Gaussian probability distribution l (z)i,t|θi,t) Wherein thetai,t＝θ(Hi,t) Representing a probability distribution parameter. The embodiment of the invention can maximize the log-likelihood
Calculating to obtain the optimal training parameter theta of the model, wherein the parameter comprises two parts: one part is the parameters in the RNN (Recurrent Neural) network, such as the weight parameter W and the bias parameter b, and the probability distribution parameter θ (·) until the model training is finished.
Based on the trained traffic prediction model, the following provides various embodiments of the traffic prediction method of the present invention.
As shown in fig. 13, fig. 13 is a flowchart of a flow prediction method according to an embodiment of the present invention, which includes, but is not limited to, step S1100 and step S1200.
Step S1100, a first characteristic value sequence and a predicted time step length are obtained, wherein the first characteristic value sequence is obtained according to a first time period and a first network flow value corresponding to the first time period;
step S1200, inputting the first characteristic value sequence and the prediction time step into a traffic prediction model to obtain a network traffic prediction value within the prediction time step, wherein the traffic prediction model is obtained by training a deep autoregressive model according to training set data, and the training set data comprises a plurality of cell identification information, a plurality of second time periods corresponding to the cell identification information and second network traffic values corresponding to the second time periods in a one-to-one mode.
In an embodiment, firstly, the embodiment of the invention can process the second time period and the second network flow value, thereby reducing the influence of abnormal data in the training set data on the prediction result; secondly, the embodiment of the invention can establish a uniform flow prediction model for all cells, thereby improving the convenience of the flow prediction model during the actual use period; in addition, the embodiment of the invention can effectively distinguish the cells by selecting the characteristic values of the cells, thereby having good prediction accuracy; finally, the deep autoregressive model is adopted to train according to the cell identification information, the second time period and the second network flow value in the training set data, and because the deep autoregressive model can conveniently incorporate extra features and can more accurately predict the future network flow of the cell, the accuracy of network flow prediction can be improved.
It should be noted that, the first feature value sequence may be obtained in step S900 or obtained in other manners. In addition, the first characteristic value sequence may be a mean value sequence, a variance sequence, or a standard deviation sequence.
For example, when a trained traffic prediction model is obtained, the trained traffic prediction model can be used for prediction, and the prediction flow is shown in fig. 14, and the detailed process is described as follows: assuming that the step size range of model training is T1, … T, the prediction step size is l ', and l' step is predicted for the ith cell. When t' is 0, initializing the parameter
And
is that
Then pass through
Is calculated to obtain
Further result in
Then sequentially iterate to obtain
Due to the fact that
Obedience distribution
Can be obtained by
Obtaining a predicted value y 'of the corresponding prediction step length from the median of the corresponding distribution function'i,t′。
As shown in fig. 15, fig. 15 is a flowchart of a flow prediction method according to another embodiment of the present invention. The flow prediction method further includes, but is not limited to, step S1300.
And step S1300, obtaining the accuracy of the flow prediction method according to the predicted value of the network flow and the actual value of the network flow within the prediction time step.
In an embodiment, the embodiment of the present invention may evaluate the model prediction effect by using the network traffic predicted value and the network traffic true value, so as to evaluate the accuracy of the overall prediction of all cells.
Illustratively, the embodiment of the present invention may be calculated by an RMSLE (Root Mean Squared Error) formula, that is, a formula is adopted
To evaluate the accuracy of the overall prediction of all cells, where l' represents the total step size of the prediction, yt′Is true value, y't′Is a predicted value.
Compared with the prior art, the flow prediction model establishing method and the flow prediction method have the following advantages that: firstly, a data preprocessing method is adopted to remove the interference of holiday flow on prediction, and meanwhile, a box plot method is utilized to reduce the influence of outliers on model training. Secondly, the existing wireless traffic prediction method establishes a prediction model for each cell, but the invention establishes a uniform prediction model for all cells instead of establishing a prediction model for each cell, thereby improving the convenience of the model during the actual use. Thirdly, when the key problem of establishing a prediction model for all cells is that the characteristics of each cell need to be selected, the embodiment of the invention selects the flow mean, the variance and the standard deviation of each cell as the characteristics of the cells through repeated trial and comparison experiment analysis, and experiments show that the characteristics can effectively distinguish the cells to achieve good prediction accuracy. Fourthly, the embodiment of the present invention obtains the eigenvalue series by using a local sliding eigenvalue calculation method. Fifthly, the flow prediction of the cell is carried out by using the deep AR for the first time. Unlike LSTM, DeepAR predicts the probability distribution of the value of a traffic sequence at each time step. The deep ar may conveniently incorporate additional features compared to the LSTM. In addition, experiments show that the embodiment of the invention can more accurately predict the future flow of the base station cell compared with the traditional statistical algorithm model.
Based on the above-described trained traffic prediction model establishment method, the following provides various embodiments of the traffic prediction model establishment apparatus of the present invention.
As shown in fig. 16, fig. 16 is a schematic diagram of a flow prediction model building apparatus 2000 according to an embodiment of the present invention. The device 2000 for establishing the flow prediction model includes, but is not limited to, a second obtaining unit 2100, a processing unit 2200, a feature extracting unit 2300 and a training unit 2400.
Specifically, the second obtaining unit 2100 is configured to obtain training set data, where the training set data includes a plurality of cell identification information, a plurality of second time periods corresponding to each cell identification information, and second network traffic values corresponding to the second time periods in a one-to-one manner; the processing unit 2200 is configured to process the second time period and the second network traffic value to obtain an initial sequence corresponding to the training set data; the feature extraction unit 2300 is configured to perform feature extraction on the initial sequence to obtain a second feature value sequence; the training unit 2400 is configured to train the deep autoregressive model by using the initial sequence and the second feature value sequence to obtain a traffic prediction model.
Referring to fig. 16, the processing unit 2200 described above includes, but is not limited to, a flow threshold value calculation unit 2210 and an adjustment unit 2220. The flow threshold value calculation unit 2210 is configured to process the second network flow value through a box plot method, so as to obtain an upper flow value and a lower flow value. The adjusting unit 2220 is configured to adjust the second network traffic value according to the traffic upper limit value and the traffic lower limit value, so as to obtain an adjusted second network traffic value.
Referring to fig. 16, the adjusting unit 2220 includes, but is not limited to, a first adjusting unit 2221 and a second adjusting unit 2222. The first adjusting unit 2221 is configured to, when the second network traffic value is greater than the traffic upper limit value, adjust the second network traffic value to an average value of a sum of the second network traffic value and the traffic upper limit value. The second adjusting unit 2222 is configured to adjust the second network traffic value to an average value of a sum of the second network traffic value and the traffic lower limit value when the second network traffic value is smaller than the traffic lower limit value.
Referring to fig. 16, the processing unit 2200 described above further includes, but is not limited to, a filtering unit 2230 and a deleting unit 2240. Wherein the screening unit 2230 is configured to screen out, from the second time period, a second time period for characterizing holidays and a second network flow value corresponding to the second time period for characterizing holidays. The deleting unit 2240 is configured to delete the second time period for characterizing the holiday and the second network traffic value corresponding to the second time period for characterizing the holiday from the training set data.
Referring to fig. 16, feature extraction section 2300 is further configured to extract features of the initial sequence to obtain a second feature value corresponding to the initial sequence, and calculate a second feature value sequence by a moving average of the second feature value.
Referring to fig. 16, training section 2400 includes, but is not limited to implicit variable calculation section 2410 and training parameter calculation section 2420. The hidden variable calculating unit 2410 is configured to process the initial sequence and the second eigenvalue sequence through the long-term and short-term memory neural network to obtain a hidden variable corresponding to a network layer in the deep autoregressive model; the training parameter calculation unit 2420 is configured to perform maximum log-likelihood calculation on the hidden variable to obtain a training parameter of the deep autoregressive model.
It should be noted that the technical effect of the traffic prediction model building apparatus 2000 according to the embodiment of the present invention can be implemented by referring to the above-mentioned traffic prediction model building method.
Based on the above-described flow rate prediction method, various embodiments of the flow rate prediction apparatus of the present invention are set forth below.
As shown in fig. 17, fig. 17 is a flow rate prediction apparatus 3000 according to an embodiment of the present invention. The flow prediction apparatus 3000 includes, but is not limited to, a first acquisition unit 3100 and a flow prediction unit 3200.
Specifically, the first obtaining unit 3100 is configured to obtain a first characteristic value sequence and a predicted time step, where the first characteristic value sequence is obtained according to a first time period and a first network traffic value corresponding to the first time period; the traffic prediction unit 3200 is configured to input the first feature value sequence and the prediction time step into a traffic prediction model to obtain a network traffic prediction value within the prediction time step, where the traffic prediction model is obtained by training a deep autoregressive model according to training set data, where the training set data includes a plurality of cell identification information, a plurality of second time periods corresponding to the cell identification information, and second network traffic values corresponding to the second time periods one to one.
Referring to fig. 17, the traffic prediction apparatus 3000 further includes an accuracy calculation unit 3300, where the accuracy calculation unit 3300 is configured to obtain an accuracy of the traffic prediction method according to the predicted value of the network traffic and the true value of the network traffic within the prediction time step.
It should be noted that, the technical effect of the flow rate prediction device 3000 according to the embodiment of the present invention can be referred to the embodiment of the flow rate prediction method.
Based on the above flow prediction model building method and flow prediction method, the following respectively proposes various embodiments of the flow prediction system and the computer readable storage medium of the present invention.
In addition, an embodiment of the present invention provides a flow prediction system including: a memory, a processor, and a computer program stored on the memory and executable on the processor.
The processor and memory may be connected by a bus or other means.
It should be noted that the flow rate prediction system in the present embodiment may correspond to the flow rate prediction system in the system architecture platform in the embodiment shown in fig. 1, and may form a part of the system architecture platform in the embodiment shown in fig. 1, and both of them belong to the same inventive concept, so both of them have the same implementation principle and beneficial effect, and will not be described in detail herein.
The non-transitory software programs and instructions required to implement the flow prediction model building method or the flow prediction method of the above-described embodiments are stored in the memory, and when executed by the processor, perform the flow prediction model building method of the above-described embodiments, for example, perform the above-described method steps S100 to S400 in fig. 2, method steps S510 to S520 in fig. 3, method step S610 in fig. 4, method step S620 in fig. 5, method steps S710 to S780 and method steps C110 to C130 in fig. 6, method steps S810 to S820 in fig. 7, method step S900 in fig. 8, the respective method steps in fig. 9, and method steps S1010 to S1020 in fig. 10. Still alternatively, when executed by a processor, the flow prediction method of the above-described embodiment is performed, for example, the method steps S1100 to S1200 in fig. 13, and the method step S1300 in fig. 15 described above are performed.
The above-described embodiments of the apparatus are merely illustrative, wherein the units illustrated as separate components may or may not be physically separate, i.e. may be located in one place, or may also be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment.
Furthermore, an embodiment of the present invention further provides a computer-readable storage medium, which stores computer-executable instructions, when the computer-executable instructions are used for executing the above-mentioned flow prediction model building method, for example, the method steps S100 to S400 in fig. 2, the method steps S510 to S520 in fig. 3, the method step S610 in fig. 4, the method step S620 in fig. 5, the method steps S710 to S780 and the method steps C110 to C130 in fig. 6, the method steps S810 to S820 in fig. 7, the method step S900 in fig. 8, the respective method steps in fig. 9, and the method steps S1010 to S1020 in fig. 10 are executed. Still alternatively, when the computer-executable instructions are used to perform the above-described flow prediction method, for example, the method steps S1100 to S1200 in fig. 13 and the method step S1300 in fig. 15 described above are performed.
One of ordinary skill in the art will appreciate that all or some of the steps, systems, and methods disclosed above may be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on computer readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). The term computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data, as is well known to those of ordinary skill in the art. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, Digital Versatile Disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can accessed by a computer. In addition, communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media as known to those skilled in the art.
While the preferred embodiments of the present invention have been described in detail, it will be understood by those skilled in the art that the foregoing and various other changes, omissions and deviations in the form and detail thereof may be made without departing from the scope of this invention.
Claims (14)
1. A traffic prediction method, comprising:
acquiring a first characteristic value sequence and a predicted time step length, wherein the first characteristic value sequence is obtained according to a first time period and a first network flow value corresponding to the first time period;
inputting the first characteristic value sequence and the prediction time step into a traffic prediction model to obtain a network traffic prediction value within the prediction time step, wherein the traffic prediction model is obtained by training a deep autoregressive model according to training set data, and the training set data comprises a plurality of cell identification information, a plurality of second time periods corresponding to the cell identification information and second network traffic values corresponding to the second time periods in a one-to-one mode.
2. The method of claim 1, further comprising:
and obtaining the accuracy of the flow prediction method according to the network flow predicted value and the network flow true value within the prediction time step.
3. The method of claim 1 or 2, wherein the first sequence of eigenvalues is one of a sequence of mean values, a sequence of variance values and a sequence of standard deviation values.
4. A flow prediction model building method comprises the following steps:
acquiring training set data, wherein the training set data comprises a plurality of cell identification information, a plurality of second time periods corresponding to the cell identification information and second network flow values corresponding to the second time periods in a one-to-one manner;
processing the second time period and the second network flow value to obtain an initial sequence corresponding to the training set data;
performing feature extraction on the initial sequence to obtain a second characteristic value sequence;
and training a depth autoregressive model by adopting the initial sequence and the second characteristic value sequence to obtain a flow prediction model.
5. The method of claim 4, wherein the processing the second time period and the second network traffic value comprises:
processing the second network flow value by a box curve graph method to obtain an upper flow limit value and a lower flow limit value;
and adjusting the second network flow value according to the flow upper limit value and the flow lower limit value to obtain the adjusted second network flow value.
6. The method of claim 5, wherein the adjusting the second network traffic value according to the upper traffic value and the lower traffic value to obtain an adjusted second network traffic value comprises at least one of:
when the second network flow value is larger than the flow upper limit value, the second network flow value is adjusted to be the average value of the sum of the second network flow value and the flow upper limit value;
and when the second network flow value is smaller than the flow lower limit value, adjusting the second network flow value to be the average value of the sum of the second network flow value and the flow lower limit value.
7. The method of claim 4, wherein the processing the second time period and the second network traffic value comprises:
screening out a second time period used for representing holidays and a second network flow value corresponding to the second time period used for representing holidays from the second time period;
deleting the second time period for characterizing holidays and the second network traffic value corresponding to the second time period for characterizing holidays from the training set data.
8. The method of claim 4, wherein the feature extracting the initial sequence to obtain a second feature value sequence comprises:
and performing feature extraction on the initial sequence to obtain a second feature value corresponding to the initial sequence, and calculating the second feature value in a moving average manner to obtain a second feature value sequence.
9. The method of claim 4, wherein training a depth autoregressive model using the initial sequence and the second sequence of feature values comprises:
processing the initial sequence and the second characteristic value sequence through a long-short term memory neural network to obtain a hidden variable corresponding to a network layer in a deep autoregressive model;
and obtaining the training parameters of the depth autoregressive model by calculating the hidden variables through the maximum log-likelihood.
10. The method of any one of claims 4 to 9, wherein the second sequence of eigenvalues is one of a sequence of mean values, a sequence of variance values and a sequence of standard deviation values.
11. A flow prediction device, comprising:
the system comprises a first acquisition unit, a second acquisition unit and a time step prediction unit, wherein the first acquisition unit is used for acquiring a first characteristic value sequence and a predicted time step, and the first characteristic value sequence is obtained according to a first time period and a first network flow value corresponding to the first time period;
and the traffic prediction unit is used for inputting the first characteristic value sequence and the prediction time step into a traffic prediction model to obtain a network traffic prediction value within the prediction time step, wherein the traffic prediction model is obtained by training a deep autoregressive model according to training set data, and the training set data comprises a plurality of cell identification information, a plurality of second time periods corresponding to the cell identification information and second network traffic values corresponding to the second time periods in a one-to-one mode.
12. A flow prediction model creation device, comprising:
a second obtaining unit, configured to obtain training set data, where the training set data includes multiple cell identification information, multiple second time periods corresponding to the cell identification information, and second network traffic values corresponding to the second time periods in a one-to-one manner;
the processing unit is used for processing the second time period and the second network flow value to obtain an initial sequence corresponding to the training set data;
the characteristic extraction unit is used for extracting the characteristics of the initial sequence to obtain a second characteristic value sequence;
and the training unit is used for training the depth autoregressive model by adopting the initial sequence and the second characteristic value sequence to obtain a flow prediction model.
13. A flow prediction system, comprising: memory, processor and computer program stored on the memory and executable on the processor, the processor implementing the flow prediction method according to any one of claims 1 to 3 when executing the computer program, or implementing the flow prediction model building method according to any one of claims 4 to 10 when executing the computer program.
14. A computer-readable storage medium storing computer-executable instructions for performing the flow prediction method according to any one of claims 1 to 3 or the flow prediction model building method according to any one of claims 4 to 10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010909565.7A CN114205853A (en) | 2020-09-02 | 2020-09-02 | Flow prediction method, flow prediction model establishing method, and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010909565.7A CN114205853A (en) | 2020-09-02 | 2020-09-02 | Flow prediction method, flow prediction model establishing method, and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114205853A true CN114205853A (en) | 2022-03-18 |
Family
ID=80644339
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010909565.7A Pending CN114205853A (en) | 2020-09-02 | 2020-09-02 | Flow prediction method, flow prediction model establishing method, and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114205853A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018202297A1 (en) * | 2017-05-04 | 2018-11-08 | Huawei Technologies Co., Ltd. | Network computing device, method and system for time-series forecasting based on a moving-median algorithm |
| CN109873712A (en) * | 2018-05-18 | 2019-06-11 | 新华三信息安全技术有限公司 | A kind of network flow prediction method and device |
| CN110474808A (en) * | 2019-08-20 | 2019-11-19 | 中国联合网络通信集团有限公司 | A kind of method for predicting and device |
| CN110858973A (en) * | 2018-08-23 | 2020-03-03 | 中国移动通信集团山东有限公司 | Method and device for predicting network traffic of cell |
| CN111148118A (en) * | 2019-12-18 | 2020-05-12 | 福建省南鸿通讯科技有限公司 | Flow prediction and carrier turn-off method and system based on time sequence |
| CN111200531A (en) * | 2020-01-02 | 2020-05-26 | 国网冀北电力有限公司信息通信分公司 | Communication network traffic prediction method and device and electronic equipment |
-
2020
- 2020-09-02 CN CN202010909565.7A patent/CN114205853A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018202297A1 (en) * | 2017-05-04 | 2018-11-08 | Huawei Technologies Co., Ltd. | Network computing device, method and system for time-series forecasting based on a moving-median algorithm |
| CN109873712A (en) * | 2018-05-18 | 2019-06-11 | 新华三信息安全技术有限公司 | A kind of network flow prediction method and device |
| CN110858973A (en) * | 2018-08-23 | 2020-03-03 | 中国移动通信集团山东有限公司 | Method and device for predicting network traffic of cell |
| CN110474808A (en) * | 2019-08-20 | 2019-11-19 | 中国联合网络通信集团有限公司 | A kind of method for predicting and device |
| CN111148118A (en) * | 2019-12-18 | 2020-05-12 | 福建省南鸿通讯科技有限公司 | Flow prediction and carrier turn-off method and system based on time sequence |
| CN111200531A (en) * | 2020-01-02 | 2020-05-26 | 国网冀北电力有限公司信息通信分公司 | Communication network traffic prediction method and device and electronic equipment |
Non-Patent Citations (3)
| Title |
|---|
| DAVID SALINAS 等: "DeepAR: Probabilistic Forecasting with Autoregressive Recurrent Networks", 《HTTP://ARXIV.ORG/ABS/1704.04110》, 13 April 2017 (2017-04-13), pages 2 - 5 * |
| 何勇 等: "基于向量自回归模型的移动通信基站流量预测", 《工业工程与管理》, 10 August 2017 (2017-08-10), pages 79 - 84 * |
| 张家晨: "基站小区无线网络流量预测方法的设计与实现", 《中国优秀硕士学位论文全文数据库 (信息科技辑)》, 15 January 2022 (2022-01-15), pages 3 - 4 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110647900B (en) | Intelligent safety situation prediction method, device and system based on deep neural network | |
| CN112532439B (en) | Network flow prediction method based on attention multi-component space-time cross-domain neural network model | |
| CN110968426B (en) | Edge cloud collaborative k-means clustering model optimization method based on online learning | |
| CN110324170B (en) | Data analysis equipment, multi-model co-decision system and method | |
| CN109816221A (en) | Decision of Project Risk method, apparatus, computer equipment and storage medium | |
| CN109104731B (en) | Method and device for building cell scene category division model and computer equipment | |
| CN110717535B (en) | Automatic modeling method and system based on data analysis processing system | |
| CN112529204A (en) | Model training method, device and system | |
| CN111294812B (en) | Resource capacity-expansion planning method and system | |
| CN108934016B (en) | Method and device for dividing cell scene categories, computer equipment and storage medium | |
| CN109344992B (en) | Modeling method for user control behavior habits of smart home integrating time-space factors | |
| CN117391641A (en) | Pilatory production flow management method and system | |
| CN113783717A (en) | Smart city network flow prediction method and system | |
| CN110621026A (en) | Base station flow multi-time prediction method | |
| CN108960486B (en) | Interactive set evolution method for predicting adaptive value based on gray support vector regression | |
| CN113205223A (en) | Electric quantity prediction system and prediction method thereof | |
| CN117575663A (en) | Fitment cost estimation method and system based on deep learning | |
| CN115018170A (en) | Short-term wind power prediction method, device, equipment and storage medium | |
| CN113159109B (en) | Wireless network flow prediction method based on data driving | |
| CN117422936A (en) | Remote sensing image classification method and system | |
| CN114372561A (en) | Network traffic prediction method based on depth state space model | |
| CN114205853A (en) | Flow prediction method, flow prediction model establishing method, and storage medium | |
| CN110390041B (en) | Online learning method and device and computer readable storage medium | |
| CN116562364A (en) | Deep learning model collaborative deduction method, device and equipment based on knowledge distillation | |
| EP4216073A1 (en) | Data management method, data management apparatus, and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |