CN113347384B - Video conference flow prediction method and system based on time sequence representation learning - Google Patents
Video conference flow prediction method and system based on time sequence representation learning Download PDFInfo
- Publication number
- CN113347384B CN113347384B CN202110899016.0A CN202110899016A CN113347384B CN 113347384 B CN113347384 B CN 113347384B CN 202110899016 A CN202110899016 A CN 202110899016A CN 113347384 B CN113347384 B CN 113347384B
- Authority
- CN
- China
- Prior art keywords
- time slice
- flow
- historical
- graph
- video conference
- 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.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N7/00—Television systems
- H04N7/14—Systems for two-way working
- H04N7/15—Conference systems
-
- 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
-
- 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
- H04L41/147—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
- H04L43/0888—Throughput
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N17/00—Diagnosis, testing or measuring for television systems or their details
- H04N17/004—Diagnosis, testing or measuring for television systems or their details for digital television systems
Abstract
The invention relates to a video conference flow prediction method and system based on time sequence representation learning. The method includes the steps that historical data are divided into equal-length historical time slices, feature information and flow information of each historical time slice are counted, each historical time slice is taken as a graph vertex to construct a flow in-person graph, and a characterization vector mapping function of each historical time slice are learned by adopting an inductive deep characterization learning method; introducing a long-short term memory network model for training to obtain a time sequence flow prediction model; according to the feature information of the current time slice, a flow in-person graph is fused by using feature similarity, a representation vector mapping function is combined, after a representation vector of the current time slice is obtained, the representation vector of the current time slice is input into a time sequence flow prediction model, and real-time flow is predicted, so that the problem of poor flow prediction accuracy of a video conference network can be effectively solved, more accurate real-time flow prediction is further realized, and the use efficiency of network communication resources is greatly improved.
Description
Technical Field
The invention relates to the technical field of flow prediction, in particular to a video conference flow prediction method and system based on time sequence representation learning.
Background
With the development of mobile communication technologies such as 5G and artificial intelligence technologies such as deep learning, the refinement, automation, intelligent operation and maintenance and management of the network become new challenges. To address this challenge, real-time network traffic prediction capability is one of the core technologies. Accurate network traffic prediction techniques can help improve communication network management. In the process of allocating network resources, the traditional method only depends on the current traffic use state of the network to allocate resources, lacks the prejudgment on the future state, and is easy to cause the continuous network congestion or resource waste. The accurate prediction of the network flow can help an operator to deal with the upcoming congestion as soon as possible, and the network expansion, adjustment and optimization are performed in advance. Network traffic prediction techniques may also improve communication network performance. With the large-scale deployment of 5G base stations, more micro base stations will be built in the future, requiring more power and backhaul fibers. The accurate prediction of the flow makes it possible to flexibly allocate resources according to the actual service requirements. Besides, the network flow prediction technology can be used for customizing the expanded network value-added service.
In order to improve the reliability and stability of digital media network video transmission, a digital media network video communication flow prediction system needs to be constructed, accurate prediction of digital media network video communication flow is realized in a streaming media management device, and research on a design method of a related digital media network video communication flow prediction system has important significance in the aspects of flow management, multimedia information transmission and the like. In recent years, China has been accompanied by continuous perfection of network infrastructure and continuous social digitization development. The video network traffic data has the characteristics of unstructured, nonlinear, real-time, burstiness, spatiotemporal and the like, and the traditional model is difficult to meet the actual network traffic prediction requirement.
In order to enable the network traffic of the video conference to be accurately predicted, the selection and design of the model are crucial. At present, prediction models are mainly divided into linear models and nonlinear models. The linear models mainly include an Autoregressive Model (AR), a Moving Average Model (MA), an Autoregressive Moving Average Model (ARMA) combining the characteristics of Autoregressive and Moving Average, and an Autoregressive synthesized Moving Average Model (ARIMA). The model has the characteristics that multiple parameters are required to be manually set by experience to fit data, and the model is only suitable for short-term flow prediction. Actual network traffic has many characteristics, such as nonlinearity, periodicity, self-similarity, burstiness, etc., and it is difficult to accurately fit and predict the actual network traffic by using only a linear model.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a video conference flow prediction method and system based on time sequence representation learning.
In order to achieve the purpose, the invention provides the following scheme:
a video conference flow prediction method based on time sequence characterization learning comprises the following steps:
acquiring historical data of a video conference;
dividing historical data of the video conference into historical time slices with equal length, and counting characteristic information on each historical time slice and flow information on each historical time slice;
constructing a flow marriage graph by taking each historical time slice as a graph node;
using an inductive deep characterization learning method to input the flow in-relative graph, and training a characterization learning model to obtain a characterization vector mapping function and a characterization vector of each historical time slice;
training a long-short term memory network model by taking the characterization vector as input and the flow information on each historical time slice as output to obtain a time sequence flow prediction model;
acquiring current data of a video conference, and dividing the current data of the video conference into time slices with equal length;
acquiring characteristic information of a current time slice;
according to the feature information on the current time slice, the flow in-parent graph is fused by using feature similarity, and a characterization vector of the current time slice is obtained by combining the characterization vector mapping function;
and taking the characterization vector of the current time slice as input, and obtaining the flow information of the current time slice by adopting a time sequence flow prediction model so as to complete the real-time prediction of the video conference flow.
Preferably, the constructing a traffic in-phase graph by using each historical time slice as a graph node specifically includes:
taking each historical time slice as a graph node, and determining k neighbors of each historical time slice according to the traffic information on each historical time slice; the k neighbors are historical time slices with k traffic information meeting preset traffic information;
and establishing a non-directional edge between each historical time slice and the k neighbor of the historical time slice to obtain the flow in-person graph.
Preferably, the method further comprises the following steps: and updating the flow in-relative graph in an off-line mode according to the characteristic information of the current time slice and the obtained flow information of the current time slice.
Preferably, the method further comprises the following steps: and updating the time sequence flow prediction model based on the updated flow in-relative graph.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the video conference flow prediction method based on time sequence representation learning provided by the invention comprises the steps of dividing equal-length historical time slices according to historical data, and counting characteristic information and flow information on each historical time slice; taking each historical time slice as a vertex in the network to construct a flow marriage graph; learning the characterization vector and the characterization vector mapping function of each historical time slice by adopting an inductive deep characterization learning method; introducing a dependency relationship on the long-short term memory network LSTM model modeling time sequence flow, and training to obtain a time sequence flow prediction model; according to the feature information of the current time slice, a flow in-person graph is fused by using feature similarity, a representation vector mapping function is combined, after a representation vector of the current time slice is obtained, the representation vector of the current time slice is input into a time sequence flow prediction model, and real-time flow is predicted, so that the problem of poor flow prediction accuracy of a video conference network can be effectively solved, more accurate real-time flow prediction is further realized, and the use efficiency of network communication resources is greatly improved.
Corresponding to the video conference flow prediction method based on time sequence characterization learning, the invention also provides the following implementation system:
a video conference traffic prediction system based on timing characterization learning, comprising:
the historical data acquisition module is used for acquiring historical data of the video conference;
the historical time slice dividing module is used for dividing the historical data of the video conference into historical time slices with equal length, and counting the characteristic information of each historical time slice and the flow information of each historical time slice;
the traffic marriage graph building module is used for building a traffic marriage graph by taking each historical time slice as a graph node;
the characterization vector output module is used for inputting the flow in-relative graph by using an inductive deep characterization learning method, and training a characterization learning model to obtain a characterization vector mapping function and a characterization vector of each historical time slice;
the time sequence flow prediction model training module is used for training a long-term and short-term memory network model by taking the characterization vector as input and the flow information on each historical time slice as output to obtain a time sequence flow prediction model;
the system comprises a current data acquisition module, a time slice generation module and a time slice generation module, wherein the current data acquisition module is used for acquiring current data of a video conference and dividing the current data of the video conference into time slices with equal length;
the characteristic information acquisition module is used for acquiring the characteristic information of the current time slice;
a characterization vector determining module, configured to blend the traffic in-phase map with feature similarity according to the feature information on the current time slice, and obtain a characterization vector of the current time slice by combining with the characterization vector mapping function;
and the flow prediction module is used for taking the characterization vector of the current time slice as input and obtaining the flow information of the current time slice by adopting a time sequence flow prediction model so as to complete the real-time prediction of the video conference flow.
Preferably, the flow affinity graph building module specifically includes:
a k neighbor determining unit, configured to determine a k neighbor of each historical time slice according to traffic information on each historical time slice, where each historical time slice is used as a graph node; the k neighbors are historical time slices with k traffic information meeting preset traffic information;
and the traffic in-person graph constructing unit is used for establishing a non-directional edge between each historical time slice and the k neighbor of the historical time slice to obtain the traffic in-person graph.
Preferably, the method further comprises the following steps:
and the flow marriage map updating module is used for updating the flow marriage map according to the characteristic information of the current time slice and the obtained flow information of the current time slice in an off-line mode.
Preferably, the method further comprises the following steps:
and the time sequence flow prediction model updating module is used for updating the time sequence flow prediction model based on the updated flow in-relative graph.
The technical effect achieved by the video conference flow prediction system based on the time sequence representation learning provided by the invention is the same as that achieved by the video conference flow prediction method based on the time sequence representation learning provided by the invention, so that the description is omitted here.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
Fig. 1 is a flowchart of a video conference traffic prediction method based on time series characterization learning according to the present invention;
fig. 2 is a structural framework diagram of a video conference traffic prediction method implemented based on time series characterization learning according to an embodiment of the present invention.
FIG. 3 is a block diagram of a prediction framework of a traffic prediction module according to an embodiment of the present invention;
fig. 4 is a flow chart of training and updating a traffic affinity graph and a time sequence traffic prediction model according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a video conference traffic prediction system based on timing characterization learning according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a video conference flow prediction method and system based on time sequence representation learning so as to improve the accuracy of real-time prediction of the flow of a video conference application system.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
As shown in fig. 1, a video conference traffic prediction method based on time series characterization learning provided by the present invention includes:
step 100: historical data of the video conference is acquired.
Step 101: dividing historical data of the video conference into historical time slices with equal length, and counting characteristic information of each historical time slice and flow information of each historical time slice.
Step 102: and constructing a traffic in-person graph by taking each historical time slice as a graph node. The method specifically comprises the following steps:
and taking each historical time slice as a graph node, and determining k neighbors of each historical time slice according to the traffic information on each historical time slice. The k neighbors are historical time slices in which k pieces of flow information meet preset flow information.
And establishing a non-directional edge between each historical time slice and the k neighbor of the historical time slice to obtain a flow affinity graph.
Step 103: and (3) taking the flow in-relation graph as input by using an inductive deep characterization learning method, and training a characterization learning model to obtain a characterization vector mapping function and a characterization vector of each historical time slice.
Step 104: and training the long-short term memory network model by taking the characterization vector as input and taking the flow information on each historical time slice as output to obtain a time sequence flow prediction model.
Step 105: and acquiring current data of the video conference, and dividing the current data of the video conference into time slices with equal length.
Step 106: and acquiring the characteristic information of the current time slice.
Step 107: and according to the feature information on the current time slice, fusing a flow in-relation graph by using feature similarity, and combining a feature vector mapping function to obtain a feature vector of the current time slice.
Step 108: and taking the characterization vector of the current time slice as input, and obtaining the flow information of the current time slice by adopting a time sequence flow prediction model so as to complete the real-time prediction of the video conference flow.
Further, in order to reduce a prediction error in a subsequent traffic prediction process, the video conference traffic prediction method based on the time series characterization learning provided by the invention further comprises:
and updating the flow in-relative graph in an off-line mode according to the characteristic information of the current time slice and the obtained real flow information of the current time slice.
And updating the time-series flow prediction model based on the updated flow marriage map (namely, after the flow marriage map is replaced by the updated flow marriage map, executing the steps 103 and 104).
The following describes in detail the video conference traffic prediction method based on the time series characterization learning according to the present invention, taking the implementation framework as shown in fig. 2 as an example.
Statistical characteristic data module 200: according to the method shown in the above steps 100 and 101, the historical log data is divided into time slices in equal length, and the characteristic information of each time slice is countedX t And flow datay t 。
The build traffic in-person graph module 201: constructing a undirected weightless graph according to the method of step 102 aboveg（v，ε，X) (namely traffic in-parent graph), modeling traffic similarity of each time slice, wherein each vertex represents one time slice, and constructing a connecting edge between the vertices by using k neighbors of the traffic.
Inductive characterization learning module 202: according to the step 103, the low-dimensional token vectors of all the vertices and the mapping function of the token vectors are learned and obtained by using an inductive token vector learning method (i.e. an inductive depth token learning method, such as GraphSAGE). Wherein the mapping function isz j =Φ(X j )。
The time series flow prediction modeling module 203: according to the step 104, a long-short term memory network LSTM model is introduced, and a time sequence flow prediction model is trained and established by utilizing the characterization vectors and the real network flow on the continuous time slice.
The real-time traffic prediction module 204: according to the step 108, as shown in fig. 3, according to the feature information of the current time slice acquired in the step 106, the current time slice node is merged into the traffic in-phase map by using feature similarity, the low-dimensional characterization vector of the current time slice is acquired according to the mapping function of the formula (1), and then the characterization vector of the current time slice node is obtainedz i Inputting the data into the LSTM network trained in step 104, and outputting the predicted network flowh i 。
The offline model update module 205: updating the latest time slice data to the flow relation graph in an off-line modeg(v，ε，X) And repeating the step 103 and the step 104, and updating the time sequence flow prediction model.
As shown in fig. 4, the detailed procedure of training and updating the timing characterization model and the flow prediction model of the present invention is as follows:
And 2, modeling each time slice into a graph node, and calculating k neighbors (k time slices with the most similar flow) of each time slice according to the flow information on each time slice. Each time slice and k neighbor thereof establish a non-directional edge to obtain a flow in-person graphg(v，ε，X). Wherein each vertexv∈VRepresenting a time slice, each edgee=(u，v)∈ERepresenting two time slicesuAndvthe network traffic of the network is close to each other,Xis a characteristic information matrix for all time slices.
wherein the content of the first and second substances,MSErepresents the average square root Error (Mean Squared Error),y t is a time slicetTrue network traffic.
And 6, updating the latest new data to historical data, repeating the steps 2, 3 and 4, and updating the flow prediction model.
In summary, the present invention designs a prediction method based on timing characterization learning for video conference network traffic prediction, which is specifically represented as: and regarding each time slice as a vertex in the network, counting the video conference information on each time slice as characteristic information, and constructing a flow in-person graph by using flow k neighbors. And learning to obtain the characterization vector and the characterization vector mapping function of each time slice by adopting an inductive deep characterization learning method. And modeling the dependency relationship on the time sequence flow by using the long-short term memory network LSTM model, and training to obtain a flow prediction model. And obtaining a characterization vector of the current time slice according to the characteristic information of the current time slice, and inputting the characterization vector into the LSTM model to obtain the predicted network flow.
Based on this, the invention has the following advantages:
1) compared with a traditional network flow prediction model, the method adopts a time sequence characterization learning method, designs a characterization learning method based on a flow in-relative graph, integrates characteristic information on time slices with similar flows, namely the time slices with similar flows can learn the characteristic information of each other, and a characterization vector considers the proximity of global and local node characteristics rather than directly utilizing statistical characteristic information to predict.
2) The invention provides a flow prediction framework based on a characterization vector, which combines characterization learning and time sequence flow modeling and can adaptively learn the parameters of a model.
3) The method utilizes the characterization vector to carry out time sequence flow modeling, utilizes the characterization vector to train the model, is beneficial to improving the prediction performance and generalization capability of the prediction model, and can well complete the flow prediction when the historical samples are not enough.
4) The invention provides a method for carrying out inductive characterization learning on a new node (a current time slice node), which carries out flow prediction by using a characterization vector of a new time slice, rather than predicting by statistical characteristics directly.
In conclusion, the method and the device can effectively solve the problem of traffic prediction of the video conference network, realize more accurate traffic prediction and improve the use efficiency of network communication resources.
In addition, the invention also provides a video conference flow prediction system based on the time sequence representation learning, which corresponds to the video conference flow prediction method based on the time sequence representation learning. As shown in fig. 5, the video conference traffic prediction system based on the time series representation learning includes: the flow prediction system comprises a historical data acquisition module 1, a historical time slice dividing module 2, a flow in-person graph construction module 3, a characterization vector output module 4, a time sequence flow prediction model training module 5, a current data acquisition module 6, a characteristic information acquisition module 7, a characterization vector determination module 8 and a flow prediction module 9.
The historical data acquisition module 1 is used for acquiring historical data of the video conference.
The historical time slice dividing module 2 is configured to divide historical data of the video conference into historical time slices with equal length, and count feature information in each historical time slice and traffic information in each historical time slice.
And the traffic affinity graph building module 3 is used for building a traffic affinity graph by taking each historical time slice as a graph node.
And the characterization vector output module 4 is used for training the characterization learning model to obtain a characterization vector mapping function and a characterization vector of each historical time slice by using the inductive deep characterization learning method to input the flow in-relative map.
And the time sequence flow prediction model training module 5 is used for training the long-short term memory network model by taking the characterization vector as input and taking the flow information on each historical time slice as output to obtain the time sequence flow prediction model.
The current data obtaining module 6 is configured to obtain current data of the video conference, and divide the current data of the video conference into equal-length time slices.
The characteristic information obtaining module 7 is configured to obtain characteristic information of the current time slice.
And the characterization vector determining module 8 is configured to blend the flow in-affinity graph by using feature similarity according to the feature information on the current time slice, and obtain a characterization vector of the current time slice by combining with a characterization vector mapping function.
The flow prediction module 9 is configured to obtain flow information of the current time slice by using the characterization vector of the current time slice as an input and using a time sequence flow prediction model, and further complete real-time prediction of the video conference flow.
The flow marriage map building module 3 preferably further includes: a k neighbor determination unit and a traffic affinity graph construction unit.
And the k neighbor determining unit is used for determining the k neighbor of each historical time slice according to the traffic information on each historical time slice by taking each historical time slice as a graph node. The k neighbors are historical time slices in which k pieces of flow information meet preset flow information.
And the traffic affinity graph constructing unit is used for establishing a non-directional edge between each historical time slice and the k neighbor of the historical time slice to obtain a traffic affinity graph.
Further, in order to reduce a prediction error in a subsequent flow prediction process, the video conference flow prediction system based on the time series characterization learning provided by the invention further comprises: the flow family graph updating module and the time sequence flow prediction model updating module.
And the flow marriage map updating module is used for updating the flow marriage map in an off-line mode according to the characteristic information of the current time slice and the obtained real flow information of the current time slice.
And the time sequence flow prediction model updating module is used for updating the time sequence flow prediction model based on the updated flow in-relative graph.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (6)
1. A video conference flow prediction method based on time sequence characterization learning is characterized by comprising the following steps:
acquiring historical data of a video conference;
dividing historical data of the video conference into historical time slices with equal length, and counting characteristic information on each historical time slice and flow information on each historical time slice;
constructing a flow marriage graph by taking each historical time slice as a graph node;
using an inductive deep characterization learning method to input the flow in-relative graph, and training a characterization learning model to obtain a characterization vector mapping function and a characterization vector of each historical time slice;
training a long-short term memory network model by taking the characterization vector as input and the flow information on each historical time slice as output to obtain a time sequence flow prediction model;
acquiring current data of a video conference, and dividing the current data of the video conference into time slices with equal length;
acquiring characteristic information of a current time slice;
according to the feature information on the current time slice, the flow in-parent graph is fused by using feature similarity, and a characterization vector of the current time slice is obtained by combining the characterization vector mapping function; the characteristic similarity is the cosine similarity between the current time slice node and the historical node;
taking the characterization vector of the current time slice as input, and obtaining the flow information of the current time slice by adopting a time sequence flow prediction model so as to complete the real-time prediction of the video conference flow;
constructing a traffic in-person graph by using each historical time slice as a graph node, wherein the method specifically comprises the following steps:
taking each historical time slice as a graph node, and determining k neighbors of each historical time slice according to the traffic information on each historical time slice; the k neighbors are historical time slices with k traffic information meeting preset traffic information;
and establishing a non-directional edge between each historical time slice and the k neighbor of the historical time slice to obtain the flow in-person graph.
2. The video conference traffic prediction method based on time series characterization learning according to claim 1, further comprising: and updating the flow in-relative graph in an off-line mode according to the characteristic information of the current time slice and the obtained flow information of the current time slice.
3. The video conference traffic prediction method based on time series characterization learning according to claim 2, further comprising: and updating the time sequence flow prediction model based on the updated flow in-relative graph.
4. A video conference traffic prediction system based on timing characterization learning, comprising:
the historical data acquisition module is used for acquiring historical data of the video conference;
the historical time slice dividing module is used for dividing the historical data of the video conference into historical time slices with equal length, and counting the characteristic information of each historical time slice and the flow information of each historical time slice;
the traffic marriage graph building module is used for building a traffic marriage graph by taking each historical time slice as a graph node;
the characterization vector output module is used for inputting the flow in-relative graph by using an inductive deep characterization learning method, and training a characterization learning model to obtain a characterization vector mapping function and a characterization vector of each historical time slice;
the time sequence flow prediction model training module is used for training a long-term and short-term memory network model by taking the characterization vector as input and the flow information on each historical time slice as output to obtain a time sequence flow prediction model;
the system comprises a current data acquisition module, a time slice generation module and a time slice generation module, wherein the current data acquisition module is used for acquiring current data of a video conference and dividing the current data of the video conference into time slices with equal length;
the characteristic information acquisition module is used for acquiring the characteristic information of the current time slice;
a characterization vector determining module, configured to blend the traffic in-phase map with feature similarity according to the feature information on the current time slice, and obtain a characterization vector of the current time slice by combining with the characterization vector mapping function; the characteristic similarity is the cosine similarity between the current time slice node and the historical node;
the flow prediction module is used for taking the characterization vector of the current time slice as input and obtaining the flow information of the current time slice by adopting a time sequence flow prediction model so as to complete the real-time prediction of the video conference flow;
the flow marriage graph building module specifically comprises:
a k neighbor determining unit, configured to determine a k neighbor of each historical time slice according to traffic information on each historical time slice, where each historical time slice is used as a graph node; the k neighbors are historical time slices with k traffic information meeting preset traffic information;
and the traffic in-person graph constructing unit is used for establishing a non-directional edge between each historical time slice and the k neighbor of the historical time slice to obtain the traffic in-person graph.
5. The video conference traffic prediction system based on timing characterization learning according to claim 4, further comprising:
and the flow marriage map updating module is used for updating the flow marriage map according to the characteristic information of the current time slice and the obtained flow information of the current time slice in an off-line mode.
6. The video conference traffic prediction system based on timing characterization learning according to claim 5, further comprising:
and the time sequence flow prediction model updating module is used for updating the time sequence flow prediction model based on the updated flow in-relative graph.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110899016.0A CN113347384B (en) | 2021-08-06 | 2021-08-06 | Video conference flow prediction method and system based on time sequence representation learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110899016.0A CN113347384B (en) | 2021-08-06 | 2021-08-06 | Video conference flow prediction method and system based on time sequence representation learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113347384A CN113347384A (en) | 2021-09-03 |
| CN113347384B true CN113347384B (en) | 2021-11-05 |
Family
ID=77480927
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110899016.0A Active CN113347384B (en) | 2021-08-06 | 2021-08-06 | Video conference flow prediction method and system based on time sequence representation learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113347384B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110730099A (en) * | 2019-10-21 | 2020-01-24 | 电子科技大学 | Flow prediction method based on historical flow data time sequence |
| CN112311608A (en) * | 2020-11-25 | 2021-02-02 | 中国人民解放军66136部队 | Multilayer heterogeneous network space node characterization method |
| WO2021130926A1 (en) * | 2019-12-25 | 2021-07-01 | 日本電信電話株式会社 | Flow of people prediction device, flow of people prediction method, and flow of people prediction program |
| CN113095535A (en) * | 2020-01-08 | 2021-07-09 | 普天信息技术有限公司 | Flow prediction method and device based on deep space-time residual error network |
| CN113162811A (en) * | 2021-06-01 | 2021-07-23 | 长扬科技（北京）有限公司 | Industrial control network flow abnormity detection method and device based on deep learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112396135B (en) * | 2021-01-21 | 2021-06-15 | 北京电信易通信息技术股份有限公司 | Method and system for detecting abnormal traffic of converged communication network |
-
2021
- 2021-08-06 CN CN202110899016.0A patent/CN113347384B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110730099A (en) * | 2019-10-21 | 2020-01-24 | 电子科技大学 | Flow prediction method based on historical flow data time sequence |
| WO2021130926A1 (en) * | 2019-12-25 | 2021-07-01 | 日本電信電話株式会社 | Flow of people prediction device, flow of people prediction method, and flow of people prediction program |
| CN113095535A (en) * | 2020-01-08 | 2021-07-09 | 普天信息技术有限公司 | Flow prediction method and device based on deep space-time residual error network |
| CN112311608A (en) * | 2020-11-25 | 2021-02-02 | 中国人民解放军66136部队 | Multilayer heterogeneous network space node characterization method |
| CN113162811A (en) * | 2021-06-01 | 2021-07-23 | 长扬科技（北京）有限公司 | Industrial control network flow abnormity detection method and device based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| Traffic Flow Prediction for Road Transportation;Afshin Abadi, Tooraj Rajabioun, and Petros A. Ioannou;《IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS》;20150314;第16卷(第2期);全文 * |
| 基于网络学习的网络流量分析综述;唐政治;《网络新媒体技术》;20200930;第9卷(第5期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113347384A (en) | 2021-09-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114362367B (en) | Cloud-edge-cooperation-oriented power transmission line monitoring system and method, and cloud-edge-cooperation-oriented power transmission line identification system and method | |
| CN111159425A (en) | Temporal knowledge graph representation method based on historical relationship and double-graph convolution network | |
| CN110267292B (en) | Cellular network flow prediction method based on three-dimensional convolutional neural network | |
| Zhang et al. | LNTP: An end-to-end online prediction model for network traffic | |
| CN110326323A (en) | A kind of method and apparatus obtaining emission probability, transition probability and sequence positioning | |
| CN115866658A (en) | Traffic prediction method and system based on dynamic spatiotemporal hypergraph convolutional network | |
| Liu et al. | Deep generative model and its applications in efficient wireless network management: A tutorial and case study | |
| CN115546589A (en) | Image generation method based on graph neural network | |
| Tan et al. | Toward a future network architecture for intelligence services: A cyber digital twin-based approach | |
| Mansour et al. | Design of integrated artificial intelligence techniques for video surveillance on iot enabled wireless multimedia sensor networks | |
| Gao et al. | A deep learning framework with spatial-temporal attention mechanism for cellular traffic prediction | |
| CN113347384B (en) | Video conference flow prediction method and system based on time sequence representation learning | |
| CN111460277B (en) | Personalized recommendation method based on mobile social network tree-shaped transmission path | |
| CN108270608B (en) | Link prediction model establishment and link prediction method | |
| CN113194493A (en) | Wireless network data missing attribute recovery method and device based on graph neural network | |
| Guzmán et al. | A cyberphysical system for data-driven real-time traffic prediction on the Las Vegas I-15 freeway | |
| CN110020379B (en) | Link prediction method based on deep dynamic network embedded representation model | |
| Lovén et al. | Towards EDISON: An edge-native approach to distributed interpolation of environmental data | |
| Deng et al. | Utility maximization of cloud-based in-car video recording over vehicular access networks | |
| Tao et al. | Urban mobility prediction based on LSTM and discrete position relationship model | |
| Zhang et al. | Generative AI for space-air-ground integrated networks (sagin) | |
| CN113240098B (en) | Fault prediction method and device based on hybrid gated neural network and storage medium | |
| CN115392481A (en) | Federal learning efficient communication method based on real-time response time balancing | |
| CN113469419A (en) | Commercial building power consumption prediction method and system | |
| CN113159409B (en) | National city air quality prediction method based on group perception map neural network |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |