CN109586950A - Network scenarios recognition methods, network management device, system and storage medium - Google Patents
Network scenarios recognition methods, network management device, system and storage medium Download PDFInfo
- Publication number
- CN109586950A CN109586950A CN201811216526.8A CN201811216526A CN109586950A CN 109586950 A CN109586950 A CN 109586950A CN 201811216526 A CN201811216526 A CN 201811216526A CN 109586950 A CN109586950 A CN 109586950A
- Authority
- CN
- China
- Prior art keywords
- network
- identified
- tree
- application scenarios
- decision
- 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.)
- Granted
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/06—Management of faults, events, alarms or notifications
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The embodiment of the present application provides a kind of network scenarios recognition methods, network management device, system and storage medium.In the embodiment of the present application, the network data reported according to the network equipment in network to be identified generates the network characterization vector of the network to be identified, is no longer dependent on manual identification, and the efficiency of network scenarios identification can be improved;In turn, according to the network characterization vector of network to be identified, the affiliated probability of network to be identified and plurality of application scenes is calculated, and the application scenarios of network to be identified are determined according to probability, the precision of scene Recognition can be improved, and then help to carry out suitability optimization to network, improve network performance.
Description
Technical field
This application involves Internet communication technology fields more particularly to a kind of network scenarios recognition methods, network management to set
Standby, system and storage medium.
Background technique
It lives indispensable a part with the rapid development of information technology, network has become people.In order to adapt to
New development needs to carry out the network optimization to network application scene, to promote the online experience of user.The range of the network coverage relates to
And different application scenarios, such as hotel's scene, office scene, shops's scene etc..For different network scenarios, need to match
It sets different network parameters and carries out the network optimization.It, could be to it therefore, it is necessary to which the application scenarios of network belonging are recognized accurately
Carry out the optimization of suitability.
Currently, carrying out application scenarios identification to network depends on manual identification, accuracy rate is lower, this undoubtedly can
The subsequent collocation degree optimized to network is influenced, network performance is influenced, user's online experience is poor.
Summary of the invention
The many aspects of the application provide network scenarios recognition methods, network management device, system and storage medium, use
With the precision that network scenarios identifies, and then help to carry out suitability optimization to network, improves network performance.
The embodiment of the present application provides a kind of network scenarios recognition methods, comprising:
According to the network data that at least network equipment in network to be identified reports, the network to be identified is generated
Network characterization vector;
According to the network characterization vector of the network to be identified, the network to be identified and at least one applied field are predicted
Affiliated probability between scape;
According to the affiliated probability of the network to be identified and at least one application scenarios, answered from at least one
With the application scenarios for determining the network belonging to be identified in scene.
The embodiment of the present application also provides a kind of network management device, comprising: memory, processor and communication component;
The memory, for storing computer program;
The processor is coupled with the memory, for executing the computer program, to be used for:
According to the network data that at least network equipment in the received network to be identified of the communication component reports,
Generate the network characterization vector of the network to be identified;According to the network characterization vector of the network to be identified, prediction it is described to
Identify the affiliated probability between network and at least one application scenarios;It is answered according to the network to be identified at least one
With the affiliated probability of scene, the application scenarios of the network belonging to be identified are determined from least one application scenarios.
It includes: at least net in network to be identified that the embodiment of the present application, which also provides a kind of network scenarios identifying system,
Network equipment and server;Wherein,
An at least network equipment, for reporting the network data of the network to be identified to the server;
The server, is used for: according to the network data of the network to be identified, generating the net of the network to be identified
Network feature vector;According to the network characterization vector of the network to be identified, the network to be identified and at least one application are predicted
Affiliated probability between scene;According to the affiliated probability of the network to be identified and at least one application scenarios, from described
The application scenarios of the network belonging to be identified are determined at least one application scenarios.
The embodiment of the present application also provides a kind of computer readable storage medium for being stored with computer program, and feature exists
In the computer program is performed the step, it can be achieved that in the above method.
In the embodiment of the present application, the network data reported according to the network equipment in network to be identified, generating should be wait know
The network characterization vector of other network, is no longer dependent on manual identification, and the efficiency of network scenarios identification can be improved;In turn, according to
The network characterization vector for identifying network calculates the affiliated probability of network to be identified and plurality of application scenes, and according to probability come really
The application scenarios of fixed network to be identified, can be improved the precision of scene Recognition, and then facilitate excellent to network progress suitability
Change, improves network performance.
Detailed description of the invention
The drawings described herein are used to provide a further understanding of the present application, constitutes part of this application, this Shen
Illustrative embodiments and their description please are not constituted an undue limitation on the present application for explaining the application.In the accompanying drawings:
Fig. 1 is a kind of structural schematic diagram for network scenarios identifying system that one exemplary embodiment of the application provides;
Fig. 2 is a kind of flow diagram for network scenarios recognition methods that one exemplary embodiment of the application provides;
Fig. 3 is a kind of structural schematic diagram for network management device that one exemplary embodiment of the application provides.
Specific embodiment
To keep the purposes, technical schemes and advantages of the application clearer, below in conjunction with the application specific embodiment and
Technical scheme is clearly and completely described in corresponding attached drawing.Obviously, described embodiment is only the application
A part of the embodiment, instead of all the embodiments.Based on the embodiment in the application, those of ordinary skill in the art are not having
Every other embodiment obtained under the premise of creative work is made, shall fall in the protection scope of this application.
Lead to recognition efficiency and the lower skill of accuracy rate dependent on manual identification for existing network scene Recognition mode
Art problem, the embodiment of the present application provide a solution, the network number reported according to the network equipment in network to be identified
According to, generate the network characterization vector of the network to be identified, be no longer dependent on manual identification, can be improved network scenarios identification effect
Rate;In turn, according to the network characterization vector of network to be identified, the affiliated probability of network to be identified and plurality of application scenes is calculated,
And the application scenarios of network to be identified are determined according to probability, the precision of scene Recognition can be improved, and then facilitate to network
Suitability optimization is carried out, network performance is improved.
Below in conjunction with attached drawing, the technical scheme provided by various embodiments of the present application will be described in detail.
Fig. 1 is a kind of structural schematic diagram for network scenarios identifying system that one exemplary embodiment of the application provides.Such as figure
Shown in 1, which includes: server 10a and at least one group of networks for being managed by server 10a.Its
In, each group of networks includes an at least network equipment 10b.The network equipment 10b and server 10a presented in Fig. 1 be
Exemplary illustration does not limit the way of realization of the two.
Wherein, wired or wireless connection is used between network equipment 10b and server 10a.Optionally, network equipment 10b
Can be communicated to connect by mobile network and server 10a, correspondingly, the network formats of mobile network can for 2G (GSM),
Appointing in 2.5G (GPRS), 3G (WCDMA, TD-SCDMA, CDMA2000, UTMS), 4G (LTE), 4G+ (LTE+), WiMax etc.
It anticipates one kind.Optionally, network equipment 10b can also pass through the modes such as bluetooth, WiFi, infrared ray and server 10a communication link
It connects.
In the present embodiment, network equipment 10b, which refers to, connects together each wired or wireless network terminal device,
Then by the equipment of wired and wireless network access network based on ethernet.For example, it may be hub, interchanger, router, light cat, nothing
The equipment that line access points (Wireless Access Point, AP) etc. have routing function, but not limited to this.
Wherein, refer to what user used with the network equipment 10b terminal device being connect, have user needed for calculate, on
The equipment of the functions such as net, communication, such as can be smart phone, tablet computer, PC, wearable device etc..Terminal device
Generally include at least one processing unit and at least one processor.The quantity of processing unit and memory is set depending on terminal
Standby configuration and type.Memory may include volatibility, such as RAM, also may include non-volatile, such as read-only
Memory (Read-Only Memory, ROM), flash memory etc., or can also simultaneously include two kinds of.Lead in memory
It often is stored with operating system (Operating System, OS), one or more application software, also can store program number
According to etc..Other than processing unit and memory, terminal device also will include network card chip, IO bus, audio-video component etc.
Basic configuration.Optionally, according to the way of realization of terminal device, terminal device also may include some peripheral equipments, such as key
Disk, mouse, input pen, printer etc..These peripheral equipments are well known in the art, and this will not be repeated here.
In the present embodiment, server 10a refers to the hardware infrastructure for managing network.Server 10a can be
One, it is also possible to more.The way of realization of the present embodiment not Limited service device 10a.For example, server 10a can be often
Advise the server apparatus such as server, Cloud Server, cloud host, virtual center.Wherein, the composition of server 10a equipment is mainly wrapped
It is similar with general computer architecture to include processor, hard disk, memory, system bus etc..
In the present embodiment, the server 10a whole network managed is known as whole net, the access in a physical region
The network equipment 10b of server 10a belongs to a logic groups for server 10a, which is defined as network
Group.In the present embodiment, scene Recognition mainly identifies the application scenarios of group of networks.For example, being hotel in physical region
When, server 10a identifies the network in the physical region, and determines that the network belonging scene is hotel's scene.Example again
Such as, when physical region is office block, server 10a identifies the network in the physical region, and determines the network
Affiliated scene is office scene.
It optionally, in the present embodiment, can be according to the physical region where whole Webweb network equipment, in whole net
The network equipment is divided according to physical region, to form at least one group of networks;And from least one group of networks determine to
It identifies network, the application scenarios belonging to it is identified.Alternatively, can also be according to the business function achieved by the network equipment
Can, the network equipment with identical services function is divided into a group of networks, so that at least one group of networks is formed, and from
Network to be identified is determined at least one group of networks, and the application scenarios belonging to it are identified.
Optionally, in the embodiment of the present application, in network to be identified at least a network equipment 10b can be to server
The request of 10a sending application scene Recognition.Correspondingly, server 10b receives identification request, and is asked according to application scenarios identification
It asks, using group of networks belonging to the network equipment for issuing identification request as network to be identified.
Further, user can operate the associated button or corresponding on at least network equipment 10b in network to be identified
Software interface control assembly etc., the relevant operation of at least network equipment 10b response user sends to server 10a
Application scenarios identification request.Alternatively, identification request sending cycle can be preset on an at least network equipment 10b, and start one
A timer or counter carry out timing to the request sending cycle.When reaching identification request sending cycle, net to be identified
An at least network equipment 10b in network is requested to server 10a sending application scene Recognition.
Alternatively, can preset in server 10a the scene Recognition period in each embodiment of the application, and start one
Timer or counter carry out timing to the scene Recognition period.It, should by current period whenever reaching in the scene Recognition period
The network group of identification is as network to be identified.
Based on above-mentioned analysis, in the present embodiment, it is under the jurisdiction of network equipment 10b in network to be identified to server 10a
Send relevant network data.These network datas include but are not limited to: the online number of at least one network equipment 10b, to
Identify the retention ratio of terminal device in network, offline user behaviors log data on the terminal device in network to be identified, to be identified
Radio frequency environment data etc. in the device type of terminal device in network, network to be identified, but not limited to this.
Correspondingly, server 10a receives the network data that an at least network equipment 10b is reported in network to be identified, and
The network characterization vector of the network to be identified is generated according to the network data;Later according to the network characterization of the network to be identified
Vector predicts the affiliated probability between the network to be identified and at least one application scenarios;And according to network to be identified and at least
A kind of affiliated probability between application scenarios determines the application scenarios of network belonging to be identified from least one application scenarios.
In the present embodiment, the network data reported according to the network equipment in network to be identified generates the net to be identified
The network characterization vector of network, is no longer dependent on manual identification, and the efficiency of network scenarios identification can be improved;Moreover, according to be identified
The network characterization vector of network calculates the affiliated probability of network to be identified and plurality of application scenes, and determined according to probability to
The application scenarios for identifying network, can be improved the precision of scene Recognition, and then help to carry out suitability optimization to network, mention
High network performance.
In the present embodiment, network application scene includes but is not limited to: office scene, shops's scene, hotel's scene,
Teaching scene, Internet bar's scene etc..Network to be identified may be implemented as any one of these scenes.Different are answered
It, therefore, in the present embodiment, can be according to net to be identified with the network characterization vector difference for the network to be identified that scene is covered
The network characterization vector of network predicts that network to be identified is implemented as the probability of these application scenarios, and according to network quilt to be identified
It is embodied as the probability of each application scenarios, determines the application scenarios belonging to it.
In the embodiment of the present application, the network data that at least network equipment 10b in network to be identified is reported, one
As include a plurality of types of data, such as a simple numerical statistic feature (at least network equipment 10b in network such as to be identified
Online number of the terminal device reported etc.), proportional-type feature is (as at least a network equipment 10b is reported in network to be identified
Three days retention ratios of terminal device etc.), the sequential character (terminal that an at least network equipment 10b is reported in network such as to be identified
Online times time difference etc. of the equipment in adjacent time interval), a text-type feature (at least network equipment 10b in network to be identified
The TF-IDF feature of the terminal device reported) etc., but not limited to this.
Based on above-mentioned analysis, server 10a is reported according at least network equipment 10b in network to be identified
Network data when generating the network characterization vector of network to be identified, a network can at least will be set according to the classification dimension of setting
The network data of every network equipment is classified in standby 10b;According to network of the every network equipment under each classification dimension
Data obtain K low order feature vector of network to be identified, wherein K is the positive integer more than or equal to 2.
Optionally, server 10a can be using K low order feature vector of network to be identified as the network of network to be identified
Feature vector.
Further, in order to improve the accuracys rate of the application scenarios to network belonging to be identified, can to K low order feature to
Amount is carried out from coded treatment, and then obtains Q high-order feature vector of network to be identified;And by K low order feature vector and Q
A high-order feature vector is together as the network characterization vector of network to be identified, and wherein Q is positive integer, and specific value can basis
Actual demand carries out flexible setting.Wherein, when carrying out K low order feature vector from coded treatment, noise reduction can be used from encoding
Device carries out from coded treatment, to obtain Q high-order feature vector of network to be identified K low order feature vector.
Further, it is contemplated that the distribution situation of network data of the every network equipment 10b under each classification dimension, every
The classification of a classification dimension is different, in order to improve the accuracy rate of the application scenarios to network belonging to be identified, server
10a generates the net of network to be identified in the network data reported according at least network equipment 10b in network to be identified
Before network feature vector, correction can be carried out to the network data that an at least network equipment 10b is reported or numeralization pre-processes.
It is illustrated below with reference to several optional embodiments:
Embodiment 1: there may be abnormal feelings for the network data reported in view of an at least network equipment 10b
Condition, in order to further increase the accuracy rate of the application scenarios to network belonging to be identified, server 10a can be according to every network
Network data distribution situation of the equipment 10b under each classification dimension, carries out at correction the feature under each classification dimension
Reason, to obtain the network data after the correction under each classification dimension.Optionally, if every network equipment 10b is at some
The mode of median numbers filling can be used by the number of missing in network data the case where there are partial data missings under dimension of classifying
According to polishing;If every network equipment 10b is in the network data under each classification dimension, there are abnormal datas, can be by abnormal number
According to filtering out.
Embodiment 2: in view of there may be text class fields by every network equipment 10b, such as in network to be identified
Device type etc., but not limited to this, then numeralization processing can be carried out to text class field.Optionally, if every network is set
The standby network data classified under dimension at some includes the device type field of the terminal device in network to be identified, then to this
Device type field carries out dummy variable processing.
Wherein, dummy variable processing refers to qualitative variable quantizes, if some is selected because being known as n kind, by it with mute
Variable processing when, n-1 dummy variable is set, to avoid complete multicollinearity, wherein n >=2 and be n positive integer.Example
Such as, in embodiment 2, dummy variable processing is carried out to the device type field of terminal device and is referred to device type Field Count
Value.Assuming that certain network equipment includes 4 terminal devices in network to be identified in the network data under some dimension
Device type field is respectively type A, B, C, D, then needs to be arranged 3 dummy variables i.e. variable 1,2,3.If device type word
The bright device type of segment table is type A, then sets 1 for variable 1, variable 2 and 3 is set as 0;If device type is equipment class
The bright device type of type-word segment table is type B, then sets 1 for variable 2, variable 1 and 3 is set as 0;If device type is to set
Standby type field shows that device type is Type C, then sets 1 for variable 3, variable 1 and 2 is set as 0;If device type
Show that device type is type D for device type field, is then all 0 by variable 1-3 setting.To 4 terminal equipment types into
The processing of row dummy variable is as shown in table 1 below:
Table 1 carries out dummy variable to 4 terminal equipment types and handles list
Device type | Variable 1 | Variable 2 | Variable 3 |
Type A | 1 | 0 | 0 |
Type B | 0 | 1 | 0 |
Type C | 0 | 0 | 1 |
Type D | 0 | 0 | 0 |
Embodiment 3: if network data of the every network equipment 10b under each classification dimension includes timing field,
It is at predetermined intervals then that unit carries out sliding-model control by the timing field.
It is worth noting that for the above embodiment 1,2,3, server 10a is to an at least network equipment 10b
When the network data reported is pre-processed, can according to it is each classification dimension under data distribution and field type into
Row selectivity is implemented, such as can be pre-processed using above embodiment 1-3, can also be using one such or two
Kind pre-process to the network data under each classification dimension.
Since scene Recognition is as the classification problem in machine learning field, often realized using disaggregated model.?
In the embodiment of the present application, model training can be carried out by machine learning, obtain decision-tree model, and utilize decision-tree model pair
The application scenarios of network belonging to be identified are predicted and are identified.The mistake of decision-tree model training is carried out to server 10a below
Journey illustrates.
Step 1: the network data of P network of samples is obtained.Wherein, P network of samples respectively belonging to application scenarios be
It is known, it is assumed that the classification number of application scenarios belonging to P network of samples is M, wherein M≤P, and P and M are positive integer.
Wherein, P improves the precision for training the decision-tree model come, it may be preferable that P sample net to meet probability statistics principle
The data volume of the network data of network can be big as far as possible, to reflect the network characterization of P network of samples comprehensively.Optionally, it can obtain
The historical network data of certain time of P network of samples is taken, for example, one week, one month, two months, 1 year etc., Huo Zhegeng
For a long time.
Step 2: according to the network data of P network of samples, the network characterization vector of P network of samples is obtained respectively.
Optionally, it when above-mentioned server 10a can be used being identified to the application scenarios of network belonging to be identified, obtains
The mode of the network characterization vector of network to be identified, to obtain the network characterization vector of P network of samples, specific descriptions can join
See above-mentioned related content, details are not described herein.
It further, can also be according to P sample before the network characterization vector that server 10a obtains P network of samples
The network data distribution situation of network, is pre-processed, details are not described herein using above embodiment 1-3.
Step 3: according to the network characterization vector of P network of samples, building represents M application belonging to P network of samples
The M stalk decision tree of scene, to generate decision-tree model.Wherein, every stalk decision tree represents an application scenarios.For every
Stalk decision tree, P sample are respectively fallen in the leaf node of the sub-tree, then for each leaf node, are corresponded to
Numerical value represent the possibility of the application scenarios that application scenarios belonging to the network of samples in the leaf node represent as the sub-tree
Property.For example, the result of decision of the sub-tree is belonging to P network of samples if a stalk decision tree represents shops's scene
A possibility that application scenarios are respectively shops's scene.
Optionally, weight, which can be used, indicates the numerical value of each leaf node.
For every stalk decision tree in the M stalk decision tree in decision-tree model, building process is identical.Under
Face is illustrated by taking the first sub-tree in M stalk decision tree as an example.Wherein, the first sub-tree can be M
Any sub-tree in sub-tree, and the first sub-tree represents the first application scenarios.For in M stalk decision tree
First sub-tree can promote decision tree (Gradient using gradient according to the network characterization vector of P network sample
Boosting Decision Tree, GBDT) algorithm or extreme gradient promote decision tree (eXtreme Gradient
Boosting Tree, XGBOOST) algorithm building represents the first sub-tree of the first application scenarios.
It wherein, can be according to the classification of setting in above-mentioned step 2 in order to improve constructed decision-tree model accuracy
Dimension classifies the network data of P network of samples, according to network number of the P network of samples under each classification dimension
According to, obtain the low order feature vector of each network of samples, wherein each classification dimension correspond to a low price feature vector;And it is right
The low order feature vector of each network of samples is carried out from coded treatment, to obtain the high-order feature vector of each network of samples;
And using the low price feature vector of each network of samples and high-order feature vector as the network characterization vector of each network of samples,
And decision-tree model is constructed using the high-order feature vector of P network sample and low price feature vector.
Further, in order to improve the robustness and accuracy of constructed decision-tree model, noise reduction self-encoding encoder can be used
The low price feature vector of each network of samples is carried out from coding, and then obtains the high-order feature vector of each network of samples.Its
In, the number of low price feature vector is determined that the two quantity is equal by the number of the classification dimension of network data.High-order feature
The number of vector can carry out flexible setting according to actual needs.
Below by taking the network characterization vector of P network sample includes low-level features vector sum high-order feature vector as an example, and
Assuming that low price feature vector is 4, the number of high-order feature vector is 2, and assumes P=5, and the first sub-tree represents net
Scene, illustrating wherein the first sub-tree to the detailed process of the first sub-tree of building includes N CART
It sets, wherein N >=2, and is positive integer, then the first sub-tree indicates the prediction output of N number of function phase Calais, expression formula are as follows:
Wherein, XiIndicate i-th of network sample, i=1,2 ... P, in this example, and P=5, i.e. i=1,2,3,4,5; fn
(Xi) indicate the function of n-th CART tree, n=i=1,2 ... N, N for the CART tree in the first stalk decision tree number;
Indicate the predicted value predicted using the first sub-tree some network sample.
S1: according to the 4 of 5 network samples low order feature vectors and 2 high-order feature vectors, the first sub- decision is selected
The split vertexes of first CART tree in tree.
Optionally, sample variance can be used and measure split vertexes purity, node is more impure, node-classification or prediction
Effect is poorer.Sample variance is bigger, indicates that the data of the node are more dispersed, the effect of prediction is poorer.Preferably for
The split vertexes of first CART tree, selection are saved using 4 low order feature vectors and 2 high-order feature vectors as division
When point, the smallest feature vector of sample variance is as split vertexes.
S2: according to the split vertexes of first CART tree, first CART tree is generated.Wherein, 5 network sample difference
The leaf node of first CART tree is fallen into, and calculates weight f of each network sample under first CART tree1(Xi)。
Optionally, cost complexity (Cost-Complexity Pruning, CCP) pruning method can be used.Select
The smallest non-leaf nodes of node surface error rate yield value, deletes the left and right child node of the non-leaf nodes, if having multiple non-
The surface error rate yield value of leaf node is identical small, then the non-leaf nodes for selecting non-leaf nodes child nodes number most
Carry out beta pruning.
S3: the weight f using corresponding activation primitive to each network sample under first CART tree1(Xi) counted
Value processing, obtains the Probability p that each network sample belongs to Internet bar's scene1(Xi)。
Optionally, activation primitive can be for softmax function, sigmoid function, Relu function or tanh function etc., but not
It is limited to this.Correspondingly, weight of each network sample under first CART tree can be brought into activation primitive, calculates each net
Network sample belongs to the Probability p of Internet bar's scene1(Xi)。
S4: the Probability p that each network sample belongs to Internet bar's scene is calculated1(Xi) with the real scene pair of each network sample
The encoded radio y answerediDifference absolute value.Optionally, for the first sub-tree, the application scenarios represented (Internet bar
Scape) it is encoded to 1, remaining application scenarios is encoded to 0.
S5: regularization objective function is used2nd CART tree is trained.Wherein, regularization objective function
Are as follows:
Wherein, regularization term isHelp to prevent over-fitting, wherein T is in every CART tree
The number of leaf node;The value of T leaf node constitutes T dimensional vector a w, w=fn(Xi).For first CART tree, w=f1
(Xi);γ and λ is respectively two T and w of Regularization function2Canonical penalty term, indicate specific gravity shared by this two difference.
L is loss function, indicates predicted valueWith the encoded radio y of application scenarios belonging to network sampleiDeviation size, and l be can
Micro- convex function.
S6: and so on, until the N CART tree training in the first sub-tree is completed.
Optionally, using above-mentioned regularization objective function, the first sub-tree is optimized using addition training, that is, is divided
Optimization order regularization objective function, first first CART tree of optimization, second CART tree of re-optimization after being over, until
N tree is optimized.
Further, server 10a can obtain the network data of S verifying network sample, to the decision tree trained
Model is verified, wherein S >=1, and be positive integer.Correspondingly, the network data that sample can be verified according to S, generates S
The network characterization vector of network sample is verified, and is folded and is intersected using K using the network characterization vector of S verifying network sample
The mode of verifying carries out cross validation to the above-mentioned decision-tree model trained, to continue to optimize every in decision-tree model
Sub-tree.Wherein, the detailed process that K rolls over cross validation belongs to techniques known, and details are not described herein.
Based on the above-mentioned decision-tree model trained, server 10a is pre- according to the network characterization vector of network to be identified
It, can be by the network characterization vector of network to be identified when surveying the affiliated probability between network to be identified and at least one application scenarios
It is sent into decision-tree model, and then is obtained between network to be identified and at least one application scenarios as input parameter
Affiliated probability;The decision-tree model includes M stalk decision tree, and every stalk decision tree represents a kind of application scenarios.
Further, since the above-mentioned decision-tree model trained includes M stalk decision tree, and every stalk decision tree packet
Containing N CART tree, when server 10a calculates the affiliated probability between network to be identified and at least one application scenarios,
The every stalk decision tree that can be sent into using the network characterization vector of network to be identified as input parameter in decision-tree model;And benefit
With N CART tree in every stalk decision tree, network to be identified application scenarios representated by every stalk decision tree are calculated
Under predicted value;Later, using the activation primitive in every stalk decision tree to corresponding predicted value carry out numerical value calculating, obtain to
Identify the affiliated probability between application scenarios representated by network and every stalk decision tree.
Optionally, activation primitive can be for softmax function, sigmoid function, Relu function or tanh function etc., but not
It is limited to this.Correspondingly, predicted value of the network to be identified under every stalk decision tree can be brought into corresponding activation primitive, calculated
Each network sample belongs to the probability of application scenarios representated by the sub-tree.
Further, every a kind of application scenarios of stalk decision tree representation in decision-tree model, and every stalk decision tree is predicted
The prediction process of the probability of the application scenarios of network belonging to be identified is identical.Below with the first sub- decision in M stalk decision tree
For tree, process, which illustrates, to be predicted to it.Wherein the first sub-tree is that any son in M stalk decision tree is determined
Plan tree.For the first sub-tree in M stalk decision tree, server 10a is utilizing N CART in the first sub-tree
Tree, when calculating the predicted value under network to be identified application scenarios representated by first sub-tree, using the first son
N CART tree in decision tree, calculates separately weight of the network to be identified under each CART tree;And by N CART tree
Gained weight results are summed, and the predicted value under network to be identified application scenarios representated by the first sub-tree is obtained.
Optionally, after calculating network to be identified using M stalk decision tree and belonging to the probability of each application scenarios, according to
Statistical principle, the network application scene of prediction and the true affiliated application scenarios of network to be identified are more close or identical, should
Probability belonging to the application scenarios of prediction is higher.Therefore, server 10a can be by network to be identified and at least one application scenarios
Between affiliated probability in maximum probability corresponding to application scenarios, the application scenarios as network belonging to be identified.
Alternatively, minimum probability threshold value can be set, server 10a by network to be identified and at least one application scenarios it
Between affiliated probability in be greater than or equal to preset probability threshold value probability corresponding to application scenarios, as network to be identified
Affiliated application scenarios.Preferably, in order to improve the accuracy rate to network belonging application scenarios to be identified, minimum probability threshold value
General setting is higher, for example, 90% or 90% or more each probability etc..
Other than network scenarios identifying system provided by the above embodiment, the embodiment of the present application also provides a kind of network
Scene recognition method is illustrated network scenarios recognition methods provided herein below from the angle of server.
Fig. 2 is a kind of flow diagram for network scenarios recognition methods that one exemplary embodiment of the application provides.The party
Method is suitable for server.As shown in Fig. 2, this method comprises:
201, the network data reported according at least network equipment in network to be identified, generates network to be identified
Network characterization vector.
202, according to the network characterization vector of network to be identified, predict network to be identified and at least one application scenarios it
Between affiliated probability.
203, according to the affiliated probability between network to be identified and at least one application scenarios, from least one applied field
The application scenarios of network belonging to be identified are determined in scape.
In step 201, the network equipment received in network to be identified sends relevant network data, and according to these
Network data generates the network characterization vector of network to be identified.These network datas include but are not limited to: an at least network
The online number of equipment, the retention ratio of terminal device in network to be identified, offline row on the terminal device in network to be identified
For the radio frequency environment data etc. in the device type of the terminal device in daily record data, network to be identified, network to be identified, but
It is without being limited thereto.
Then, in step 202, according to the network characterization vector of network to be identified, network to be identified and at least one are predicted
Affiliated probability between kind application scenarios, that is, predict that the application scenarios of network belonging to be identified are respectively these application scenarios
Probability is predicted than the application scenarios belonging to directly according to the network data of network to be identified to it, and it is accurate that prediction can be improved
Rate and forecasting efficiency.
When predicting the affiliated probability between network to be identified and at least one application scenarios, in step 203, just
The application of network belonging to be identified can be determined according to the affiliated probability between network to be identified and at least one application scenarios
Scene is any at least one application scenarios, and then identifies the application scenarios of network belonging to be identified.Later, just
The application scenarios of network to be identified can be directed to, using the network parameter being adapted to its application scenarios, network to be identified are carried out excellent
Change, the performance of network to be identified can be improved, and then improve user's online experience.
In the present embodiment, the network data reported according to the network equipment in network to be identified generates the net to be identified
The network characterization vector of network, is no longer dependent on manual identification, and the efficiency of network scenarios identification can be improved;In turn, according to be identified
The network characterization vector of network calculates the affiliated probability of network to be identified and plurality of application scenes, and determined according to probability to
The application scenarios for identifying network, can be improved the precision of scene Recognition, and then help to carry out suitability optimization to network, mention
High network performance.
In each embodiment of the application, the whole network of server admin is known as whole net, connecing in a physical region
The network equipment for entering server belongs to a logic groups for server, which is defined as group of networks.At this
In embodiment, scene Recognition mainly identifies the application scenarios of group of networks.
It is alternatively possible to according to the physical region where whole Webweb network equipment, to the network equipment in whole net according to object
Reason region is divided, to form at least one group of networks;And network to be identified is determined from least one group of networks, to it
Affiliated application scenarios are identified.Alternatively, can also will be had identical according to the business function achieved by the network equipment
The network equipment of business function is divided into a group of networks, to form at least one group of networks, and from least one group of networks
Middle determination network to be identified, identifies the application scenarios belonging to it.
Further, can be known according to the application scenarios that at least a network equipment is sent in the network to be identified received
It does not invite and asks, using group of networks belonging to the network equipment for issuing identification request as network to be identified.Wherein, network to be identified
In at least network equipment sending application scene Recognition request embodiment can be found in the phase in the above system embodiment
Description is closed, details are not described herein.
Alternatively, the scene Recognition period can be preset, and starts a timer or counter and the scene Recognition period is carried out
Timing.Whenever reaching in the scene Recognition period, the network group that current period should be identified is as network to be identified.
In an alternative embodiment, it is contemplated that the network data that at least network equipment in network to be identified reports,
A plurality of types of data are generally comprised, such as simple numerical statistic feature is (in network such as to be identified on an at least network equipment
Online number of the terminal device of report etc.), the proportional-type feature (terminal that an at least network equipment reports in network such as to be identified
Three days retention ratios of equipment etc.), (terminal device that at least a network equipment reports in network such as to be identified exists sequential character
Online times time difference of adjacent time interval etc.), the text-type feature (end that an at least network equipment reports in network to be identified
Word frequency-inverse document frequency (Term Frequency-Inverse the Document Frequency, TF- of end equipment
IDF) feature) etc., but not limited to this.Based on this, a kind of optional embodiment of step 201 are as follows: tieed up according to the classification of setting
Degree classifies the network data of every network equipment in an at least network equipment;According to every network equipment each
Network data under dimension of classifying, obtains K low order feature vector of network to be identified, and by K low order of network to be identified
Network characterization vector of the feature vector as network to be identified.
Further, in order to improve the accuracys rate of the application scenarios to network belonging to be identified, one kind of step 201 is optional
Embodiment are as follows: on the basis of K low order feature vector of above-mentioned acquisition, K low order feature vector is carried out from coding
Reason, obtains Q high-order feature vector of network to be identified;And together by K low order feature vector and Q high-order feature vector
As the network characterization vector of network to be identified, wherein Q is positive integer, and specific value can flexibly be set according to actual needs
It sets.
Further, noise reduction self-encoding encoder can be used to carry out from coded treatment K low order feature vector, thus obtain to
Identify Q high-order feature vector of network.
Further, it is contemplated that the distribution situation of network data of the every network equipment under each classification dimension, Mei Gefen
The classification of class dimension is different, in order to improve the accuracy rate of the application scenarios to network belonging to be identified, step 201 it
Before, correction can be carried out to the network data that an at least network equipment reports or numeralization pre-processes.Wherein, at least one
Network data that the network equipment reports carries out pretreated optional embodiment can be with are as follows: according to every network equipment each
Network data distribution situation under dimension of classifying carries out correction processing to the feature under each classification dimension, to obtain each point
The network data after correction under class dimension;And/or if network data of the every network equipment under each classification dimension includes
The device type field of terminal device in network to be identified then carries out dummy variable processing to device type field;If and/or
Network data of the every network equipment under each classification dimension includes timing field, then by timing field with preset
Time interval is that unit carries out sliding-model control.
Since scene Recognition is as the classification problem in machine learning field, often realized using disaggregated model.?
In the embodiment of the present application, model training can be carried out by machine learning, obtain decision-tree model, and utilize decision-tree model pair
The application scenarios of network belonging to be identified are predicted and are identified.The process of decision-tree model training is carried out below exemplary
Explanation.
Step 1: the network data of P network of samples is obtained.Wherein, P network of samples respectively belonging to application scenarios be
It is known, it is assumed that the classification number of application scenarios belonging to P network of samples is M, wherein M≤P, and P and M are positive integer.
Wherein, P improves the precision for training the decision-tree model come, it may be preferable that P sample net to meet probability statistics principle
The data volume of the network data of network can be big as far as possible, to reflect the network characterization of P network of samples comprehensively.Optionally, it can obtain
The historical network data of certain time of P network of samples is taken, for example, one week, one month, two months, 1 year etc., Huo Zhegeng
For a long time.
Step 2: according to the network data of P network of samples, the network characterization vector of P network of samples is obtained respectively.
Optionally, it when above-mentioned server can be used being identified to the application scenarios of network belonging to be identified, obtains wait know
The mode of the network characterization of other network specifically describes to obtain the network characterization vector of P network of samples and can be found in above-mentioned phase
Hold inside the Pass, details are not described herein.
It further, can also be according to P network of samples before the network characterization vector that server obtains P network of samples
Network data distribution situation, pre-processed using above embodiment 1-3, details are not described herein.
Step 3: according to the network characterization vector of P network of samples, building represents M application belonging to P network of samples
The M stalk decision tree of scene, to generate decision-tree model.Wherein, every stalk decision tree represents an application scenarios.For every
Stalk decision tree, P sample are respectively fallen in the leaf node of the sub-tree, then for each leaf node, are corresponded to
Numerical value represent the possibility of the application scenarios that application scenarios belonging to the network of samples in the leaf node represent as the sub-tree
Property.For example, the result of decision of the sub-tree is belonging to P network of samples if a stalk decision tree represents shops's scene
A possibility that application scenarios are respectively shops's scene.
Optionally, weight, which can be used, indicates the numerical value of each leaf node.
For every stalk decision tree in the M stalk decision tree in decision-tree model, building process is identical.Under
Face is illustrated by taking the first sub-tree in M stalk decision tree as an example.Wherein, the first sub-tree can be M
Any sub-tree in sub-tree, and the first sub-tree represents the first application scenarios.For in M stalk decision tree
First sub-tree can construct generation using GBDT algorithm or XGBOOST algorithm according to the network characterization vector of P network sample
First sub-tree of the first application scenarios of table.Wherein, it determines for constructing the first son using GBDT algorithm or XGBOOST algorithm
The specific embodiment of plan tree can be found in the above system embodiment the related content of step S1-S6 and to P network sample
Network data pre-processed and generated the related content of network characterization vector, details are not described herein.
Based on the above-mentioned decision-tree model trained, a kind of optional embodiment of step 202 are as follows: by network to be identified
Network characterization vector be sent into decision-tree model as input parameter, and then obtain network to be identified and at least one
Affiliated probability between application scenarios;The decision-tree model includes M stalk decision tree, and every stalk decision tree represents a kind of application
Scene.
Further, since the above-mentioned decision-tree model trained includes M stalk decision tree, and every stalk decision tree packet
Containing N CART tree, a kind of optional reality of the affiliated probability between network to be identified and at least one application scenarios is calculated
Apply mode are as follows: the every stalk decision being sent into decision-tree model using the network characterization vector of network to be identified as input parameter
Tree;And using N CART tree in every stalk decision tree, network to be identified is calculated representated by every stalk decision tree
Predicted value under application scenarios;Later, numerical value meter is carried out to corresponding predicted value using the activation primitive in every stalk decision tree
It calculates, obtains the affiliated probability between application scenarios representated by network to be identified and every stalk decision tree.
Optionally, activation primitive can be for softmax function, sigmoid function, Relu function or tanh function etc., but not
It is limited to this.Correspondingly, predicted value of the network to be identified under every stalk decision tree can be brought into corresponding activation primitive, calculated
Each network sample belongs to the probability of application scenarios representated by the sub-tree.
Further, every a kind of application scenarios of stalk decision tree representation in decision-tree model, and every stalk decision tree is predicted
The prediction process of the probability of the application scenarios of network belonging to be identified is identical.Below with the first sub- decision in M stalk decision tree
For tree, process, which illustrates, to be predicted to it.Wherein the first sub-tree is that any son in M stalk decision tree is determined
Plan tree.It is calculated using N CART tree in the first sub-tree wait know for the first sub-tree in M stalk decision tree
A kind of optional embodiment of predicted value under other network application scenarios representated by first sub-tree are as follows: utilize
N CART tree in first sub-tree, calculates separately weight of the network to be identified under each CART tree;And by N
Weight results obtained by CART tree are summed, and are obtained under network to be identified application scenarios representated by the first sub-tree
Predicted value.
Optionally, after calculating network to be identified using M stalk decision tree and belonging to the probability of each application scenarios, according to
Statistical principle, the network application scene of prediction and the true affiliated application scenarios of network to be identified are more close or identical, should
Probability belonging to the application scenarios of prediction is higher.Based on this, a kind of optional embodiment of step 203 are as follows: by network to be identified
Application scenarios corresponding to the maximum probability in affiliated probability between at least one application scenarios, as network to be identified
Affiliated application scenarios.
Alternatively, minimum probability threshold value can be set.Based on minimum probability threshold value, a kind of optional embodiment of step 203
Are as follows: preset probability threshold value will be greater than or equal in the affiliated probability between network to be identified and at least one application scenarios
Application scenarios corresponding to probability, the application scenarios as network belonging to be identified.Preferably, in order to improve to net to be identified
The accuracy rate of the affiliated application scenarios of network, minimum probability threshold value be generally arranged it is higher, for example, 90% or 90% or more each probability
Etc..
It should be noted that the executing subject of each step of above-described embodiment institute providing method may each be same equipment,
Alternatively, this method is also by distinct device as executing subject.For example, step 201 and 202 executing subject can be equipment A;
For another example, the executing subject of step 201 can be equipment A, and the executing subject of step 202 can be equipment B;Etc..
In addition, containing in some processes of the description in above-described embodiment and attached drawing according to particular order appearance
Multiple operations, but it should be clearly understood that these operations can not execute or parallel according to its sequence what appears in this article
It executes, serial number of operation such as 201,202 etc. is only used for distinguishing each different operation, and serial number itself does not represent any
Execute sequence.In addition, these processes may include more or fewer operations, and these operations can execute in order
Or parallel execution.
Fig. 3 is a kind of structural schematic diagram of network management device provided by the embodiments of the present application.As shown in figure 3, network pipe
Managing equipment includes: memory 30a, processor 30b and communication component 30c.
Wherein, memory 30a for storing computer program, and can be configured to store various other data to support
Operation on network management device.Wherein, the computer program stored in memory 30a can be performed in processor 30b, with reality
Now corresponding control logic.Memory 30a can be by any kind of volatibility or non-volatile memory device or their group
Close and realize, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), it is erasable can
Program read-only memory (EPROM), programmable read only memory (PROM), read-only memory (ROM), magnetic memory, flash memory
Reservoir, disk or CD.
Wherein, communication component 30c is configured to facilitate wired or wireless way between network management device and other equipment
Communication.Network management device can access the wireless network based on communication standard, such as WiFi, 2G or 3G or their group
It closes.In one exemplary embodiment, communication component receives the broadcast from external broadcasting management system via broadcast channel and believes
Number or broadcast related information.In one exemplary embodiment, the communication component further includes near-field communication (NFC) module, with
Promote short range communication.For example, NFC module can be based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology,
Ultra wide band (UWB) technology, bluetooth (BT) technology and other technologies are realized.
Wherein, communication component 30c, the network number that at least network equipment for receiving in network to be identified reports
According to.
Memory 30a, for storing computer program.
Processor 30b is coupled with memory 30a, for executing the computer program, to be used for: according to network to be identified
In the network data that reports of an at least network equipment, generate the network characterization vector of network to be identified;According to net to be identified
The network characterization vector of network predicts the affiliated probability between network to be identified and at least one application scenarios;According to net to be identified
The affiliated probability of network and at least one application scenarios determines the net to be identified from least one application scenarios
Application scenarios belonging to network.
Optionally, processor 30c is also used to: the physical region or achieved where whole Webweb network equipment
Business function divides to form at least one group of networks the network equipment in whole net according to physical region;And from least
Network to be identified is determined in one group of networks.
In an alternative embodiment, processor 30c is before the network characterization vector for generating the network to be identified, also
For: according to network data distribution situation of the every network equipment under each classification dimension, under each classification dimension
Feature carry out correction processing, with obtain it is each classification dimension under correction after network data;And/or
If network data of the every network equipment under each classification dimension includes the terminal device in network to be identified
Device type field then carries out dummy variable processing to device type field;
If and/or every network equipment it is each classification dimension under network data include timing field, by timing
Property field be at predetermined intervals unit carry out sliding-model control.
In another alternative embodiment, processor 30c is specific to use when generating the network characterization vector of network to be identified
In: the network data of every network equipment in an at least network equipment is classified according to the classification dimension of setting;According to
Network data of the every network equipment under each classification dimension, obtains K low order feature vector of network to be identified;To K
Low order feature vector is carried out from coded treatment, to obtain Q high-order feature vector of network to be identified;By K low order feature to
Amount and network characterization vector of the Q high-order feature vector as network to be identified;Wherein, K is the positive integer more than or equal to 2,
Q is positive integer.
Further, processor 30c is specifically used for when carrying out K low order feature vector from coded treatment: using drop
Self-encoding encoder of making an uproar carries out from coded treatment, to obtain Q high-order of the network to be identified the K low order feature vector
Feature vector.
In a further alternative embodiment, processor 30c is being predicted between network to be identified and at least one application scenarios
When affiliated probability, it is specifically used for: is sent into the network characterization vector of network to be identified as input parameter in decision-tree model,
Obtain the affiliated probability between network to be identified and at least one application scenarios;Wherein, decision-tree model includes M stalk
Decision tree, every stalk decision tree represent a kind of application scenarios, and M is positive integer.
Further, processor 30c is sending the network characterization vector of network to be identified as input parameter into decision tree mould
When obtaining the affiliated probability between network to be identified and at least one application scenarios in type, it is specifically used for: by net to be identified
The network characterization vector of network is sent into every stalk decision tree in decision-tree model as input parameter;Utilize every stalk decision tree
In N CART tree, calculate the predicted value under network to be identified application scenarios representated by every stalk decision tree;Benefit
Numerical value calculating is carried out to corresponding predicted value with the activation primitive in every stalk decision tree, network to be identified is obtained and determines with every stalk
Affiliated probability between the representative application scenarios of plan tree;Wherein, N is the positive integer more than or equal to 2.
Further, for the first sub-tree in M stalk decision tree, processor 30c is using in every stalk decision tree
N CART tree, when calculating the predicted value under network to be identified application scenarios representated by every stalk decision tree, really
It says with cutting, using N CART tree in the first sub-tree, calculates network to be identified in the first sub-tree institute's generation
When predicted value under the application scenarios of table, it is specifically used for: using N CART tree in the first sub-tree, calculates separately wait know
Weight of the other network under each CART tree;Weight results obtained by N CART tree are summed to obtain network to be identified
Predicted value under the application scenarios representated by first sub-tree;Wherein, the first sub-tree is that the M stalk is determined
Any sub-tree in plan tree.
Further, processor 30c is specific when determining the application scenarios of network to be identified from least one application scenarios
For: by applied field corresponding to the maximum probability in the affiliated probability between network to be identified and at least one application scenarios
Scape, the application scenarios as network belonging to be identified;Or by the institute between network to be identified and at least one application scenarios
Belong to and be greater than or equal to application scenarios corresponding to the probability of preset probability threshold value in probability, as network belonging to be identified
Application scenarios.
In yet another alternative embodiment, processor 30c is using the network characterization vector of network to be identified as input parameter
It before being sent into decision-tree model, is also used to: obtaining the network data of P network of samples;According to the network of P network of samples
Data obtain the network characterization vector of P network of samples respectively;According to the network characterization vector of P network of samples, generation is constructed
The M stalk decision tree of M application scenarios belonging to P network of samples of table, to generate the decision-tree model;Wherein, P is big
In or equal to M positive integer.
Further, for the first sub-tree in M stalk decision tree, the first decision tree represents the first application scenarios, place
Device 30c is managed in the network characterization vector according to P network sample, building represents M application scenarios belonging to P network of samples
M stalk decision tree when, exactly, the first application scenarios are being represented according to the network characterization vector building of P network sample
The first sub-tree when, be specifically used for: according to the network characterization vector of P network sample, using GBDT algorithm or
XGBOOST algorithm constructs the first sub-tree for representing the first application scenarios;Wherein, the first sub-tree is M stalk decision tree
In any sub-tree.
In some optional embodiments, as shown in figure 3, the network management device can also include: power supply module 30d
Etc. optional component.Members are only schematically provided in Fig. 3, are not meant to that network management device must be comprising complete shown in Fig. 3
Parts do not mean that network management device can only include component shown in Fig. 3 yet.
Wherein, the various assemblies that power supply module 30d is configured as network management device provide electric power.Power supply module 30d can
To include power-supply management system, one or more power supplys and other generated, managed, and distributed with for equipment where power supply module
The associated component of electric power.
It should be noted that in the present embodiment, network management device refers to equipment required for carrying out network management,
It configures the requirement that should meet network management, can be network management unit, the equipment of network management center etc. of each node, can
To be planned in industrial environment network, be controlled and be monitored, it can be ensured that the normal operation of network.Network management device
It can be achieved as server, network management unit etc., but not limited to this.
In the present embodiment, the network data that network management device is reported according to the network equipment in network to be identified, it is raw
At the network characterization vector of the network to be identified, it is no longer dependent on manual identification, the efficiency of network scenarios identification can be improved;Into
And according to the network characterization vector of network to be identified, calculate the affiliated probability of network to be identified and plurality of application scenes, and root
The application scenarios of network to be identified are determined according to probability, can network management device improve the precision of scene Recognition, and then help
Suitability optimization is carried out to network in network management device, improves network performance.
The embodiment of the present application also provides a kind of computer readable storage medium for being stored with computer program, when the meter
Calculation machine program is performed the step, it can be achieved that in the above method.
It should be noted that the description such as herein " first ", " second ", be for distinguish different message, equipment,
Module etc. does not represent sequencing, does not also limit " first " and " second " and is different type.
It should be understood by those skilled in the art that, the embodiment of the present invention can provide as method, system or computer journey
Sequence product.Therefore, complete hardware embodiment, complete software embodiment or combining software and hardware aspects can be used in the present invention
The form of embodiment.Moreover, it wherein includes the calculating of computer usable program code that the present invention, which can be used in one or more,
The computer program implemented in machine usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.)
The form of product.
The present invention be referring to according to the method for the embodiment of the present invention, the process of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that can be realized by computer program instructions each in flowchart and/or the block diagram
The combination of process and/or box in process and/or box and flowchart and/or the block diagram.It can provide these computers
Processor of the program instruction to general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices
To generate a machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute
For realizing the function of being specified in one or more flows of the flowchart and/or one or more blocks of the block diagram
Device.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that instruction stored in the computer readable memory generation includes
The manufacture of command device, the command device are realized in one box of one or more flows of the flowchart and/or block diagram
Or the function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that
Series of operation steps are executed on computer or other programmable devices to generate computer implemented processing, thus calculating
The instruction executed on machine or other programmable devices is provided for realizing in one or more flows of the flowchart and/or side
The step of function of being specified in block diagram one box or multiple boxes.
In a typical configuration, calculate equipment include one or more processors (CPU), input/output interface,
Network interface and memory.
Memory may include the non-volatile memory in computer-readable medium, random access memory (RAM) and/
Or the forms such as Nonvolatile memory, such as read-only memory (ROM) or flash memory (flash RAM).Memory is computer-readable medium
Example.
Computer-readable medium includes permanent and non-permanent, removable and non-removable media can be by any side
Method or technology realize that information stores.Information can be computer readable instructions, data structure, the module of program or other numbers
According to.The example of the storage medium of computer includes, but are not limited to phase change memory (PRAM), static random access memory
(SRAM), dynamic random access memory (DRAM), other kinds of random access memory (RAM), read-only memory
(ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory techniques, CD-ROM are read-only
Memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, tape magnetic disk storage or
Other magnetic storage devices or any other non-transmission medium, can be used for storage can be accessed by a computing device information.It presses
It is defined according to herein, computer-readable medium does not include temporary computer readable media (transitory media), is such as modulated
Data-signal and carrier wave.
It should also be noted that, the terms "include", "comprise" or its any other variant are intended to nonexcludability
It include so that the process, method, commodity or the equipment that include a series of elements not only include those elements, but also to wrap
Include other elements that are not explicitly listed, or further include for this process, method, commodity or equipment intrinsic want
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including described
There is also other identical elements in the process, method of element, commodity or equipment.
The above description is only an example of the present application, is not intended to limit this application.For those skilled in the art
For, various changes and changes are possible in this application.All any modifications made within the spirit and principles of the present application, etc.
With replacement, improvement etc., should be included within the scope of the claims of this application.
Claims (14)
1. a kind of network scenarios recognition methods characterized by comprising
According to the network data that at least network equipment in network to be identified reports, the network of the network to be identified is generated
Feature vector;
According to the network characterization vector of the network to be identified, predict between the network to be identified and at least one application scenarios
Affiliated probability;
According to the affiliated probability between the network to be identified and at least one application scenarios, from least one application
The application scenarios of the network belonging to be identified are determined in scene.
2. the method according to claim 1, wherein further include:
Physical region or achieved business function where whole Webweb network equipment, set the network in the whole net
It is standby to be divided according to physical region to form at least one group of networks;
Network to be identified is determined from least one described group of networks.
3. the method according to claim 1, wherein at least network according in network to be identified is set
The standby network data reported, generates the network characterization vector of the network to be identified, comprising:
The network data of every network equipment in an at least network equipment is classified according to the classification dimension of setting;
According to network data of the every network equipment under each classification dimension, K for obtaining the network to be identified are low
Rank feature vector;
The K low order feature vector is carried out from coded treatment, with obtain Q high-order feature of the network to be identified to
Amount;
Using the K low order feature vector and the Q high-order feature vector as the network characterization of the network to be identified to
Amount;
Wherein, K is the positive integer more than or equal to 2, and Q is positive integer.
4. according to the method described in claim 3, it is characterized in that, described carry out from coding the K low order feature vector
It handles, includes: to obtain Q high-order feature vector of the network to be identified
The K low order feature vector is carried out from coded treatment, to obtain the network to be identified using noise reduction self-encoding encoder
Q high-order feature vector.
5. the method according to claim 1, wherein the network characterization vector for generating the network to be identified it
Before, further includes:
According to network data distribution situation of the every network equipment under each classification dimension, to each classification dimension
Under feature carry out correction processing, with obtain it is described it is each classification dimension under correction after network data;And/or
If network data of the every network equipment under each classification dimension includes that the terminal in the network to be identified is set
Standby device type field then carries out dummy variable processing to the device type field;And/or
If network data of the every network equipment under each classification dimension includes timing field, by the timing
Field is that unit carries out sliding-model control at predetermined intervals.
6. according to the method described in claim 5, it is characterized in that, the network characterization according to the network to be identified to
Amount predicts the affiliated probability between the network to be identified and at least one application scenarios, comprising:
It is sent into decision-tree model, obtains described to be identified using the network characterization vector of the network to be identified as input parameter
Affiliated probability between network and at least one application scenarios;Wherein, the decision-tree model includes M stalk decision tree,
Every stalk decision tree represents a kind of application scenarios, and M is positive integer.
7. according to the method described in claim 6, it is characterized in that, using the network characterization vector of the network to be identified as defeated
Enter in parameter feeding decision-tree model and obtain the affiliated probability between the network to be identified and at least one application scenarios,
Include:
Every stalk that the network characterization vector of the network to be identified is sent into the decision-tree model as input parameter is determined
Plan tree;
Using N CART tree in every stalk decision tree, the network to be identified is calculated in every stalk decision tree institute
Predicted value under the application scenarios of representative;
Numerical value calculating is carried out to corresponding predicted value using the activation primitive in every stalk decision tree, obtains the net to be identified
Affiliated probability between application scenarios representated by network and every stalk decision tree;
Wherein, N is the positive integer more than or equal to 2.
8. the method according to the description of claim 7 is characterized in that for the first sub-tree in the M stalk decision tree,
The N CART tree using in every stalk decision tree calculates the network to be identified in every stalk decision tree institute
Predicted value under the application scenarios of representative, comprising:
Using N CART tree in first sub-tree, the network to be identified is calculated separately under each CART tree
Weight;
Weight results obtained by the N CART tree are summed to obtain the network to be identified in the first sub-tree institute
Predicted value under the application scenarios of representative;Wherein, first sub-tree is any sub- decision in the M stalk decision tree
Tree.
9. according to the method described in claim 6, it is characterized in that, using the network characterization vector of the network to be identified as
Before inputting in parameter feeding decision-tree model, further includes:
Obtain the network data of P network of samples;
According to the network data of the P network of samples, the network characterization vector of the P network of samples is obtained respectively;
According to the network characterization vector of the P network of samples, building represents M applied field belonging to the P network of samples
The M stalk decision tree of scape, to generate the decision-tree model;
Wherein, P is the positive integer more than or equal to M.
10. according to the method described in claim 9, it is characterized in that, for the first sub- decision in the M stalk decision tree
Tree, first decision tree represent the first application scenarios, the network characterization vector according to the P network sample, building
Represent the M stalk decision tree of M application scenarios belonging to the P network of samples, comprising:
According to the network characterization vector of the P network sample, described the is represented using GBDT algorithm or the building of XGBOOST algorithm
First sub-tree of one application scenarios;Wherein, first sub-tree is any sub- decision in the M stalk decision tree
Tree.
11. -10 described in any item methods according to claim 1, which is characterized in that described according to the network to be identified and institute
The affiliated probability between at least one application scenarios is stated, the network to be identified is determined from least one application scenarios
Application scenarios, comprising:
It will be corresponding to the maximum probability in the affiliated probability between the network to be identified and at least one application scenarios
Application scenarios, the application scenarios as the network belonging to be identified;Or
It is preset general by being greater than or equal in the affiliated probability between the network to be identified and at least one application scenarios
Application scenarios corresponding to the probability of rate threshold value, the application scenarios as the network belonging to be identified.
12. a kind of network management device characterized by comprising memory, processor and communication component;
The communication component, the network data that at least network equipment for receiving in network to be identified reports;
The memory, for storing computer program;
The processor is coupled with the memory, for executing the computer program, to be used for:
According to the network data that at least network equipment in the network to be identified reports, the network to be identified is generated
Network characterization vector;According to the network characterization vector of the network to be identified, predict that the network to be identified is answered at least one
With the affiliated probability between scene;According to the affiliated probability of the network to be identified and at least one application scenarios, from institute
State the application scenarios that the network belonging to be identified is determined at least one application scenarios.
13. a kind of network scenarios identifying system characterized by comprising server and by least the one of the server admin
A group of networks;Wherein, each group of networks includes an at least network equipment, is used for the server report network data;
The server, is used for: network to be identified is determined from least one described group of networks, according in the network to be identified
The network data that an at least network equipment reports generates the network characterization vector of the network to be identified;According to described wait know
The network characterization vector of other network predicts the affiliated probability between the network to be identified and at least one application scenarios;According to
The affiliated probability of the network to be identified and at least one application scenarios determines institute from least one application scenarios
State the application scenarios of network belonging to be identified.
14. a kind of computer readable storage medium for being stored with computer program, which is characterized in that the computer program is held
, it can be achieved that step in any one of claim 1-11 the method when row.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811216526.8A CN109586950B (en) | 2018-10-18 | 2018-10-18 | Network scene recognition method, network management device, network scene recognition system and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811216526.8A CN109586950B (en) | 2018-10-18 | 2018-10-18 | Network scene recognition method, network management device, network scene recognition system and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109586950A true CN109586950A (en) | 2019-04-05 |
CN109586950B CN109586950B (en) | 2022-08-16 |
Family
ID=65920650
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811216526.8A Active CN109586950B (en) | 2018-10-18 | 2018-10-18 | Network scene recognition method, network management device, network scene recognition system and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109586950B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020098679A1 (en) * | 2018-11-15 | 2020-05-22 | 华为技术有限公司 | Method for recognizing network scene, and access network device |
CN111931797A (en) * | 2019-05-13 | 2020-11-13 | 中国移动通信集团湖南有限公司 | Method, device and equipment for identifying network to which service belongs |
CN114140140A (en) * | 2020-09-03 | 2022-03-04 | 中国移动通信集团浙江有限公司 | Scene screening method, device and equipment |
CN114513399A (en) * | 2021-12-31 | 2022-05-17 | 锐捷网络股份有限公司 | Device identification method and device, computer-readable storage medium and electronic device |
CN114630356A (en) * | 2020-12-11 | 2022-06-14 | 中移(成都)信息通信科技有限公司 | Base station determination method, device, equipment and storage medium |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101359372A (en) * | 2008-09-26 | 2009-02-04 | 腾讯科技(深圳)有限公司 | Training method and device of classifier, and method apparatus for recognising sensitization picture |
CN103297427A (en) * | 2013-05-21 | 2013-09-11 | 中国科学院信息工程研究所 | Unknown network protocol identification method and system |
CN103617417A (en) * | 2013-11-25 | 2014-03-05 | 中国科学院深圳先进技术研究院 | Automatic plant identification method and system |
CN105740909A (en) * | 2016-02-02 | 2016-07-06 | 华中科技大学 | Text recognition method under natural scene on the basis of spatial transformation |
CN107180245A (en) * | 2016-03-10 | 2017-09-19 | 滴滴(中国)科技有限公司 | A kind of indoor and outdoor scene recognition method and device |
CN107633207A (en) * | 2017-08-17 | 2018-01-26 | 平安科技(深圳)有限公司 | AU characteristic recognition methods, device and storage medium |
CN107682216A (en) * | 2017-09-01 | 2018-02-09 | 南京南瑞集团公司 | A kind of network traffics protocol recognition method based on deep learning |
-
2018
- 2018-10-18 CN CN201811216526.8A patent/CN109586950B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101359372A (en) * | 2008-09-26 | 2009-02-04 | 腾讯科技(深圳)有限公司 | Training method and device of classifier, and method apparatus for recognising sensitization picture |
CN103297427A (en) * | 2013-05-21 | 2013-09-11 | 中国科学院信息工程研究所 | Unknown network protocol identification method and system |
CN103617417A (en) * | 2013-11-25 | 2014-03-05 | 中国科学院深圳先进技术研究院 | Automatic plant identification method and system |
CN105740909A (en) * | 2016-02-02 | 2016-07-06 | 华中科技大学 | Text recognition method under natural scene on the basis of spatial transformation |
CN107180245A (en) * | 2016-03-10 | 2017-09-19 | 滴滴(中国)科技有限公司 | A kind of indoor and outdoor scene recognition method and device |
CN107633207A (en) * | 2017-08-17 | 2018-01-26 | 平安科技(深圳)有限公司 | AU characteristic recognition methods, device and storage medium |
CN107682216A (en) * | 2017-09-01 | 2018-02-09 | 南京南瑞集团公司 | A kind of network traffics protocol recognition method based on deep learning |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020098679A1 (en) * | 2018-11-15 | 2020-05-22 | 华为技术有限公司 | Method for recognizing network scene, and access network device |
CN111931797A (en) * | 2019-05-13 | 2020-11-13 | 中国移动通信集团湖南有限公司 | Method, device and equipment for identifying network to which service belongs |
CN111931797B (en) * | 2019-05-13 | 2023-09-08 | 中国移动通信集团湖南有限公司 | Method, device and equipment for identifying network to which service belongs |
CN114140140A (en) * | 2020-09-03 | 2022-03-04 | 中国移动通信集团浙江有限公司 | Scene screening method, device and equipment |
CN114140140B (en) * | 2020-09-03 | 2023-03-21 | 中国移动通信集团浙江有限公司 | Scene screening method, device and equipment |
CN114630356A (en) * | 2020-12-11 | 2022-06-14 | 中移(成都)信息通信科技有限公司 | Base station determination method, device, equipment and storage medium |
CN114630356B (en) * | 2020-12-11 | 2024-02-27 | 中移(成都)信息通信科技有限公司 | Base station determining method, device, equipment and storage medium |
CN114513399A (en) * | 2021-12-31 | 2022-05-17 | 锐捷网络股份有限公司 | Device identification method and device, computer-readable storage medium and electronic device |
Also Published As
Publication number | Publication date |
---|---|
CN109586950B (en) | 2022-08-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109586950A (en) | Network scenarios recognition methods, network management device, system and storage medium | |
CN110363449B (en) | Risk identification method, device and system | |
US20230244199A1 (en) | Dynamic Digital Twin of Distributed Energy Demand | |
Goyal et al. | Optimal configuration selection for reconfigurable manufacturing system using NSGA II and TOPSIS | |
CN106067080B (en) | Configurable workflow capabilities are provided | |
CN110020913A (en) | Products Show method, equipment and storage medium | |
CN107645403A (en) | Terminal rule engine apparatus, terminal rule operation method | |
US20180004859A1 (en) | Computation platform agnostic data classification workflows | |
US11595269B1 (en) | Identifying upgrades to an edge network by artificial intelligence | |
CN110188210A (en) | One kind is based on figure regularization and the independent cross-module state data retrieval method of mode and system | |
CN110471821A (en) | Abnormal alteration detection method, server and computer readable storage medium | |
CN106097161A (en) | Water affairs management system and data processing method thereof | |
KR20220072311A (en) | Method for designing intelligent integrated logistics platform | |
US20220058499A1 (en) | Multidimensional hierarchy level recommendation for forecasting models | |
CN111582341A (en) | User abnormal operation prediction method and device | |
CN107016115A (en) | Data export method, device, computer-readable recording medium and electronic equipment | |
CN104951442A (en) | Method and device for determining result vector | |
US11538048B1 (en) | Predictively identifying activity subscribers | |
CN115237592A (en) | Mixed cloud service flow scheduling method based on privacy perception | |
Subramanian et al. | Constrained prescriptive trees via column generation | |
CN109903079A (en) | Information processing method, equipment and storage medium | |
CN111967611A (en) | Feature generation method and apparatus, electronic device, and storage medium | |
CN114611015B (en) | Interactive information processing method and device and cloud server | |
CN109582476A (en) | Data processing method, apparatus and system | |
CN109068286A (en) | A kind of method, medium and the equipment of information parsing |
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 |