CN108205520A - KPI and QoE regression models method for building up and device - Google Patents
KPI and QoE regression models method for building up and device Download PDFInfo
- Publication number
- CN108205520A CN108205520A CN201611186034.XA CN201611186034A CN108205520A CN 108205520 A CN108205520 A CN 108205520A CN 201611186034 A CN201611186034 A CN 201611186034A CN 108205520 A CN108205520 A CN 108205520A
- Authority
- CN
- China
- Prior art keywords
- regression model
- data
- training
- trained
- qoe
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/147—Network analysis or design for predicting network behaviour
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Computer Networks & Wireless Communication (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Signal Processing (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Biology (AREA)
- Algebra (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Operations Research (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The embodiment of the invention discloses a kind of KPI and QoE regression models method for building up and device.The method includes:W historical data of telecommunication service event is obtained, and a training data is selected from the w historical data, wherein, each historical data in the w historical data includes KPI data and QoE data;The a training data is divided into k station work data by the QoE ranks belonging to the QoE data of each training data in a training data;Regression model is trained respectively using the k station work data, so as to obtain k trained regression model, optimal trained regression model is selected from described k trained regression model, wherein, w, a, k are the integer more than zero, w>a≥k.
Description
Technical field
The present invention relates to MultiMedia Field more particularly to a kind of KPI and QoE regression models method for building up and device.
Background technology
Become the market of sufficient competition with the market development of telecommunications service, promoting the satisfaction of client has become telecommunications fortune
Battalion quotient possesses own customers and attracts the only way of rival client.The satisfaction of client can use client perception body
(Quality of Experience, QoE) is tested to represent.Technical staff has found that Quality of Experience is main in long-term practice
By the influence of Key Performance Indicator (Key Performance Indicators, KPI).So can be KPI and QoE it
Between establish a regression model.After Network Management System collects the real-time KPI data of telecommunication service event, by telecommunication service
The real-time KPI data of event is input to regression model, it is possible to predict the real-time QoE of telecommunication service event.If telecommunications industry
The real-time QoE of business event is less than predetermined threshold value, then finds that QoE is deteriorated, propose alarm.
But the accuracy of the regression model of the prior art is not generally high, it is impossible to meet the needs of actual use.
Invention content
The embodiment of the invention discloses a kind of KPI and QoE regression models method for building up and device, can improve KPI with
The accuracy of QoE regression models.
In a first aspect, provide a kind of KPI and QoE regression model method for building up.Network Management System obtains telecommunication service
W historical data of event, wherein, each historical data in the w historical data includes KPI data and QoE numbers
According to.Then, Network Management System selects a training data from the w historical data, according in a training data
Each training data QoE data belonging to QoE ranks, a training data is divided into k station work data.Net
Network management system reuses the k station work data and regression model is trained respectively, is returned so as to obtain k training
Model, and optimal trained regression model is selected from described k trained regression model, wherein, w, a, k are whole more than zero
Number, w>a≥k.
As can be seen that in above-mentioned technical proposal, a training data is divided into k station work data according to QoE ranks,
Then, k station work data are reused regression model is trained to obtain k trained regression model.Due to difference
The training datas of QoE ranks can be incorporated into different classification, the comparison in difference between the training data of same QoE ranks
It is small, so, the accuracy for training the k trained regression model come can be than by all training datas while to same recurrence mould
The accuracy of training regression model that type is trained will height.Moreover, also reselection goes out one from k trained regression model
A optimal training regression model, can further improve the accuracy of regression model.
With reference to first aspect, in the first possible embodiment of first aspect, Network Management System is gone through from the w
B verification data are selected in history data, and using described b verification data respectively to described k in trained regression model
Each training regression model is verified, to obtain the verification of each training regression model in described k trained regression model
As a result.Finally, the verification result of each training regression model of the Network Management System in described k trained regression model from
Described k trained regression model selects optimal trained regression model.Wherein, the b verification data and described a trained number
According to being belonging respectively to different data acquisition systems, that is, a training data belongs to set A, and the b training data belongs to set
B does not have intersection between set A and set B.
It is appreciated that in the prior art, the verification to regression model be all by by regression model put in the field into
Row verification, that is, regression model is actually used into a period of time, to observe the effect of regression model.And in the above-mentioned technical solutions,
It only needs that from usage history data verification to regression model can be completed, greatly improves the convenience of verification.
The possible embodiment of with reference to first aspect the first, Network Management System are selected from the w historical data
Select out c test data, and the verification result of each training regression model in described k trained regression model is from described
K trained regression model selects trained regression model undetermined.Later, Network Management System uses the c test data pair
The trained regression model undetermined is tested to obtain test result, if the test result is more than predetermined threshold value, by institute
It states trained regression model undetermined and is set as optimal trained regression model, if the test result is less than or equal to default threshold
Value abandons the trained regression model undetermined.Wherein, a training data, the b verification data and described c survey
Examination data are belonging respectively to different data acquisition systems, that is, a training data belongs to set A, and the b training data belongs to
Set B, the c test data belong to set C, do not have intersection between set A, set B and set C.
It is appreciated that only using the test result that test data carries out trained regression model undetermined test acquisition big
When predetermined threshold value, just the trained regression model undetermined can be set as optimal trained regression model, conversely, just by instruction undetermined
Practice regression model to abandon, it is ensured that the accuracy of optimal trained regression model can be just retained when sufficiently high.
Any of the above-described kind of possible embodiment with reference to first aspect, the 4th kind of possible embodiment of first aspect
In, the regression model is multinomial Logic Regression Models.
Any of the above-described kind of possible embodiment with reference to first aspect, the 5th kind of possible embodiment of first aspect
In, the granularity of the telecommunication service event is user setting.Wherein, the telecommunication service event can be the world transferred to
Calling, provincial calling, city-level calling, enterprise-level calling etc., alternatively, the telecommunication service event can be the world received
Calling, provincial calling, city-level calling, enterprise-level calling etc..
Any of the above-described kind of possible embodiment with reference to first aspect, the 6th kind of possible embodiment of first aspect
In, the KPI data includes at least one of bandwidth availability ratio, time delay, shake and packet loss.
Second aspect provides a kind of KPI and establishes device with QoE regression models, including:Acquisition module, sort module, instruction
Practice module and selecting module,
The acquisition module is used to obtain w historical data of telecommunication service event, and is selected from the w historical data
A training data is selected, wherein, each historical data in the w historical data includes KPI data and QoE data;
QoE of the sort module belonging to for the QoE data of each training data in a training data
The a training data is divided into k station work data by rank;
The training module is for being respectively trained regression model using the k station work data, so as to obtain
K trained regression model is obtained,
The selecting module is used to select optimal trained regression model from described k trained regression model, wherein, w,
A, k is the integer more than zero, w>a≥k.
With reference to second aspect, in the first possible embodiment of second aspect, the selecting module further includes:Verification
Unit and selecting unit,
Acquisition module is additionally operable to select b verification data from the w historical data, wherein, the b verification number
Different data acquisition systems are belonging respectively to according to from a training data;
The authentication unit is used for each in trained regression model to described k respectively using described b verification data
Training regression model is verified, to obtain the verification result of each training regression model in described k trained regression model;
The selecting unit is for each verification result for training regression model in described k trained regression model
Optimal trained regression model is selected from described k trained regression model.
With reference to the first possible embodiment of second aspect, in second of possible embodiment of second aspect,
The selecting unit further includes selecting unit undetermined, test cell and setting unit,
The acquisition module is used to select c test data from the w historical data, wherein, the c test
Data, the b verification data are belonging respectively to different data acquisition systems from a training data;
The verification result of each training regression model of the selecting unit undetermined in described k trained regression model
Trained regression model undetermined is selected from described k trained regression model;
The test cell is used to that the trained regression model undetermined to be tested to obtain using the c test data
Obtain test result;
The setting unit is used in the case where the test result is more than predetermined threshold value, and the training undetermined is returned
Model is set as optimal trained regression model.
With reference to second of possible embodiment of second aspect, in the third possible embodiment of second aspect,
The selecting unit further includes discarding unit,
The discarding unit is used in the case where the test result is less than or equal to predetermined threshold value, will be described undetermined
Training regression model abandons.
With reference to any of the above-described kind of possible embodiment of second aspect, the 4th kind of possible embodiment of second aspect
In, the regression model is multinomial Logic Regression Models.
With reference to any of the above-described kind of possible embodiment of second aspect, the 5th kind of possible embodiment of second aspect
In, the granularity of the telecommunication service event is user setting.
With reference to any of the above-described kind of possible embodiment of second aspect, the 6th kind of possible embodiment of second aspect
In, the KPI data includes at least one of bandwidth availability ratio, time delay, shake and packet loss.
In addition, the embodiment of the present application third aspect provides a kind of Network Management System.It is single that Network Management System includes storage
Member, communication interface and the processor coupled with the storage unit and communication interface.The storage unit for storing instruction, institute
Processor is stated for performing described instruction, the communication interface is used to other equipment be led under the control of the processor
Letter.The method in first aspect can be performed according to described instruction when the processor is performing described instruction.
In addition, the embodiment of the present application fourth aspect provides a kind of computer readable storage medium, it is described computer-readable
Storage medium stores the program code for KPI and QoE regression model method for building up.Said program code includes performing
The instruction of method in first aspect.
Description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described.
Fig. 1 is the flow chart of a kind of KPI and QoE regression model method for building up that the prior art proposes;
Fig. 2 is the structure diagram of the network architecture that the scheme possible application of the embodiment of the present application is arrived;
Fig. 3 is the flow chart that the embodiment of the present invention proposes a kind of KPI and QoE regression model method for building up;
Fig. 4 be telecommunication service event in the embodiment of the present invention granularity be same call object when schematic diagram;
Signal when Fig. 5 is call object of the granularity of telecommunication service event in the embodiment of the present invention for same enterprise
Figure;
Schematic diagram when Fig. 6 is call object of the granularity of telecommunication service event in the embodiment of the present invention for same county;
Fig. 7 is the structure diagram that the embodiment of the present invention proposes a kind of KPI and QoE regression models establish device;
Fig. 8 is that the embodiment of the present invention proposes a kind of structure diagram of Network Management System.
Specific embodiment
The technical solution in the embodiment of the present invention is explicitly described below in conjunction with attached drawing.
In order to make it easy to understand, the general modeling process of KPI and QoE regression models is introduced first.As shown in Figure 1,
KPI and the QoE regression model of the prior art include:
101:Network Management System obtains w historical data of telecommunication service event, wherein, it is every in w historical data
A historical data includes KPI data and QoE data.QoE data can be that A (very satisfied), B (satisfaction) and C are (discontented
Meaning) etc..
102:The w historical data is input in regression model and the regression model is instructed by Network Management System
Practice, so as to obtain trained regression model.
From the above, it can be seen that in existing technical solution, w historical data is input to same by Network Management System
Regression model is trained, so as to obtain a trained regression model.Both it is A including QoE data in this w historical data
When historical data, historical data when being also C including QoE data, also, when QoE data are A, KPI data and QoE data
Relationship and QoE data when being C, KPI data and the relationship difference of QoE data are very big.So by all different Q oE ranks
For historical data for training same trained regression model, the accuracy that can lead to the training regression model is not high.
Understand for ease of scheme, it first below may come the scheme of the lower the embodiment of the present application of introduction of illustrating with reference to relevant drawings
The network architecture being applied to.
As shown in Fig. 2, the scheme possible application of the embodiment of the present application to the network architecture relate generally to obtain KPI data
Network element, the network element and Network Management System for obtaining QoE data.Wherein, it is returned for performing the embodiment of the present application KPI and QoE
Method for establishing model is Network Management System, and Network Management System can be deployed among one or more application server.Network
Different function component in management system can be deployed among same or different application server.Wherein, Network Management System
Can be a newly-increased system or an existing system.When Network Management System is an existing system
When, be the equal of by carrying out function enhancing to existing system, so that the existing system after function enhancing can be held
KPI and QoE the regression model method for building up of row the application.
In a specific embodiment, the network element for obtaining KPI data is with the network element for obtaining KPI data ability.For example,
The network element for obtaining KPI data can be router (Router), Layer 2 switch (2 switches of Layer), three-tier switch
(3 switches of Layer) etc..Certainly, the network element that can obtain KPI data is far above above-mentioned network element.
In a specific embodiment, the network element for obtaining QoE data is with the network element for obtaining QoE data capabilities.For example,
Media quality monitoring center (Media Quality Monitor Center, MQMC) or business intelligence gateway (service
Intelligence gateway, SIG) etc..Certainly, the network element that can obtain QoE data is far above above-mentioned network element.
In practical applications, the network element for obtaining KPI data and the network element for obtaining QoE data can be same network elements,
It can not be same network element (that is, obtain KPI data is a network element, obtain QoE data is another network element), this hair
It is bright to be not especially limited.
The KPI data acquired is sent to network management device by the network element for obtaining KPI data, obtains QoE data
The QoE data acquired are sent to network management device by network element, and network management device is by KPI data and QoE data correlations
Get up, and as history data store in network management device.
In order to solve the problems, such as that the accuracy of KPI and QoE regression models in existing technical solution is not high, the present invention is implemented
Example proposes a kind of KPI and QoE regression model method for building up.As shown in figure 3, KPI and the QoE regression model of the present embodiment are established
Method includes:
210:Network Management System obtains w historical data of telecommunication service event, and is selected from the w historical data
Select a training data.Wherein, each historical data in the w historical data includes KPI data and QoE data, w
It is the integer more than zero with a.
Using KPI provided in an embodiment of the present invention KPI and QoE regression models are established with QoE regression model method for building up
Before, it is necessary first to determine the granularity of KPI and the targeted telecommunication service event of QoE regression models.Wherein, telecommunication service event
Telecommunication service session is may be, for example, specifically, the telecommunication service event may be, for example, the telecommunication service of same call object
Session collection (for example, as shown in Figure 4, multiple telecommunication service sessions of user A calling parties B) or different callings pair
The telecommunication service session collection of elephant.When the telecommunication service event is the telecommunication service session collection of different call objects, telecommunications industry
Business session collection can be the telecommunication service session collection for enterprise-level (for example, as shown in Figure 5, user A calling parties B1, using
Multiple telecommunication service sessions of family B2 and user B3, wherein, user B1, user B2 and user B3 belong to same local
Net) or at county level (for example, as shown in Figure 6, user A calling parties B1, user B2, user B3, user B4, user B5 and use
Multiple telecommunication service sessions of family B6, wherein, user B1, user B2 and user B3 belong to same LAN, user B4, use
Family B5 and user B6 belong to same LAN) telecommunication service session collection or for other region granularities (for example, city-level,
It is provincial or international) telecommunication service session collection.Wherein, the telecommunication service referred in the embodiment of the present application is, for example, text industry
Business, audio service, video traffic or multimedia (such as videoconference) business etc..In actual use, telecommunication service event
Granularity can be that user sets according to their needs.
It is understood that the granularity and KPI and QoE regression models of the historical data that Network Management System obtains are targeted
The granularity of telecommunication service event must be identical.Citing illustrates, if KPI and the targeted electricity of QoE regression models
Communication service event is the telecommunication service session that party A-subscriber calls party B-subscriber, then the KPI data and QoE data that Network Management System obtains
Must be that party A-subscriber calls the KPI data of telecommunication service session of party B-subscriber and QoE data.If KPI and QoE regression models institute needle
To telecommunication service event be telecommunication service session that party A-subscriber calls L enterprises, then Network Management System obtains KPI data and
QoE data must be the KPI data of telecommunication service session and QoE data that party A-subscriber calls L enterprises.If KPI and QoE is returned
The telecommunication service event that model is directed to is the telecommunication service session that party A-subscriber calls S cities, then the KPI of Network Management System acquisition
Data and QoE data must be the KPI data of telecommunication service session and QoE data that party A-subscriber calls S cities.
In embodiments of the present invention, KPI data is mainly used for describing the performance that the network equipment can be provided.It is specific one
In embodiment, KPI data can be at least one of bandwidth availability ratio, time delay, shake and packet loss.Implement in others
In example, KPI data can also be that packet data protocol message creates success rate (PDP Context Activation Success
Ratio), packet data protocol message creates duration (PDP Context Activation time), domain name system server name
When title is parsed into power (DNS Host Name Resolution Success Ratio), domain name system server name resolving
Long (DNS Host Name Resolution time), it obtains and transmits request success rate (Get and Post Request
Success Ratio), obtain and transmission request duration (Get and Post Request time) etc..
In embodiments of the present invention, QoE data be from the angle of user come weigh network quality quality index.
In one embodiment, QoE data value range is to include:A (very satisfied), B (satisfaction) and C (dissatisfied) three
QoE ranks.Network Management System cannot directly acquire to obtain QoE data, can only directly acquire to obtain MOS values, and according to MOS values
Corresponding QoE data are calculated with the correspondence of QoE data.Wherein, MOS values receive to assess in each users from networks
The index of the perceived quality of media.In one embodiment, this index value range is【1~5】, wherein, 1 is worst sense
Know quality, 5 be highest perceived quality.In a specific embodiment, MOS values and the correspondence of QoE data are as shown in table 1.
It is understood that the correspondence shown in table 1 is to parse the present invention, specific restriction should not be formed.
The mapping table of 1 MOS values of table and QoE data
More intuitively understand historical data for the ease of reader, citing below illustrates, and table 2 shows Network Management Department
The a plurality of historical data stored in system.Wherein, every a line in table 2 is a historical data.
2 historical data table of table
In embodiments of the present invention, Network Management System selects a training data from w historical data.With shown in Fig. 2
Table for, Network Management System selected from 9 historical datas row 1 to 5 totally 5 historical datas using as training data.
220:QoE belonging to the QoE data of each training data of the Network Management System in a training data
The a training data is divided into k station work data by rank.Wherein, k is the integer more than zero.
230:Network Management System is respectively trained regression model using the k station work data, so as to obtain
Obtain k trained regression model.
In order to make it easy to understand, step 220 and step 230 are illustrated with reference to table 2.In previous step,
Network Management System selected from 9 historical datas row 1 to 5 totally 5 historical datas as training data.This 5 trained numbers
According to altogether comprising three QoE ranks:A (very satisfied), B (satisfaction) and C (dissatisfied).Wherein, the second row historical data and
The QoE data of fifth line historical data are A (very satisfied), so, the second row historical data and fifth line can be gone through
History data are divided into first station work data;The QoE data of the first row historical data and fourth line historical data are B
(satisfaction), so, the first row historical data and fourth line historical data can be divided into second station work data;Third
The QoE data of row historical data are C (dissatisfied), so, the third line historical data can be divided into third station work number
According to.
First station work data is inputted regression model to be trained regression model, so as to obtain first training
Regression model;Second station work data is inputted regression model to be trained regression model, so as to obtain second instruction
Practice regression model;Third station work data input regression model is trained regression model, so as to obtain third
Training regression model.
In embodiments of the present invention, regression model can be multinomial Logic Regression Models, can also be multilayer perceptron
Neural network, random forest, support vector machines learning model and deep learning machine learning model etc..
240:Optimal trained regression model is selected from described k trained regression model.
In a specific embodiment, Network Management System can select one from k trained regression model at random
Training regression model is using as optimal trained regression model.
In a specific embodiment, Network Management System can also select b verification number from w historical data
According to, each training regression model using described b verification data respectively to described k in trained regression model is verified,
To obtain the verification result of each training regression model in described k trained regression model, and returned according to described k training
The verification result of each training regression model in model selects optimal training from described k trained regression model and returns mould
Type.
Further, c test data can also be selected from the w historical data, is trained back according to described k
The verification result for each training regression model returned in model selects training undetermined from described k trained regression model and returns mould
Type;The trained regression model undetermined is tested to obtain test result using the c test data;If the survey
Test result is more than predetermined threshold value, and the trained regression model undetermined is set as optimal trained regression model;If the test
As a result it is less than or equal to predetermined threshold value, the trained regression model undetermined is set as optimal trained regression model.
It is understood that a training data chosen from w historical data, b verification data and c
Test data belongs to different set.That is, a training data belongs to set A, b verification data belong to set B, c test number
According to set C is belonged to, there is no intersection between set A, set B and set C.
The above-mentioned method for illustrating the embodiment of the present invention, for the ease of preferably implementing the above-mentioned of the embodiment of the present invention
Correspondingly, the device of the embodiment of the present invention is provided below in scheme.
Refering to Fig. 7, Fig. 7 is the structural representation that a kind of KPI provided in an embodiment of the present invention and QoE regression models establish device
Figure.KPI and the QoE regression model of the present embodiment are established device and are included:Acquisition module 710, sort module 720, training module 730
And selecting module 740.
The acquisition module 710 is used to obtaining w historical data of telecommunication service event, and from the w historical data
A training data of middle selection, wherein, each historical data in the w historical data includes KPI data and QoE numbers
According to;
The sort module 720 is for belonging to the QoE data of each training data in a training data
The a training data is divided into k station work data by QoE ranks;
The training module 730 is used to respectively be trained regression model using the k station work data, so as to
Obtain k trained regression model;
The selecting module 740 is used to select optimal trained regression model from described k trained regression model,
In, w, a, k are the integer more than zero, w>a≥k.
Optionally, the selecting module 740 further includes:Authentication unit 741 and selecting unit 742,
Acquisition module 710 is additionally operable to select b verification data from the w historical data, wherein, the b are tested
Card data are belonging respectively to different data acquisition systems from a training data;
The authentication unit 741 is used for every in trained regression model to described k respectively using described b verification data
A trained regression model is verified, to obtain the verification knot of each training regression model in described k trained regression model
Fruit;
The selecting unit is for each verification result for training regression model in described k trained regression model
Optimal trained regression model is selected from described k trained regression model.
Optionally, the selecting unit further includes selecting unit 7411 undetermined, test cell 7412 and setting unit
7413,
The acquisition module 710 is used to select c test data from the w historical data, wherein, the c
Test data, the b verification data are belonging respectively to different data acquisition systems from a training data;
The verification of each training regression model of the selecting unit undetermined 7411 in described k trained regression model
As a result trained regression model undetermined is selected from described k trained regression model;
The test cell 7412 is used to test the trained regression model undetermined using the c test data
To obtain test result;
The setting unit 7413 is used in the case where the test result is more than predetermined threshold value, by the training undetermined
Regression model is set as optimal trained regression model.
Optionally, described device further includes discarding unit 7414, and the discarding unit 7414 is used in the test result
In the case of less than or equal to predetermined threshold value, the trained regression model undetermined is abandoned.
Optionally, the regression model is multinomial Logic Regression Models.
Optionally, the granularity of the telecommunication service event is user setting.
Optionally, the KPI data includes at least one of bandwidth availability ratio, time delay, shake and packet loss.
For example, KPI and QoE regression models establish device perform in above-described embodiment the KPI of step 210-240 with
The flow of QoE regression model method for building up.
Referring to Fig. 8, KPI provided by the embodiments of the present application and QoE regression models are established device and are included:Storage unit 810 is led to
Letter interface 820 and the processor 830 coupled with the storage unit 810 and communication interface 820.The storage unit 810 is used for
Store instruction, for the processor 820 for performing described instruction, the communication interface 820 is used for the control in the processor 830
System is lower to communicate with other equipment.This Shen can be performed according to described instruction when the processor 830 is performing described instruction
Any one KPI and QoE regression model method for building up that please be in above-described embodiment.
Processor 830 can also claim central processing unit (CPU, Central Processing Unit).Storage unit 810
It can include read-only memory and random access memory, and instruction and data etc. is provided to processor 830.Storage unit 810
A part may also include nonvolatile RAM.KPI establishes device with QoE regression models in specific application
Each component is for example coupled by bus system.Bus system can also include electricity other than it may include data/address bus
Source bus, controlling bus and status signal bus in addition etc..But for the sake of clear explanation, various buses are all designated as always in figure
Linear system system 840.The method that the embodiments of the present invention disclose can be applied to realize in processor 830 or by processor 830.Place
It may be a kind of IC chip to manage device 830, has the processing capacity of signal.During realization, each step of the above method
Suddenly it can be completed by the integrated logic circuit of the hardware in processor 830 or the instruction of software form.Wherein, above-mentioned processing
Device 830 can be general processor, digital signal processor, application-specific integrated circuit, ready-made programmable gate array or other can
Programmed logic device, discrete gate or transistor logic, discrete hardware components.Processor 830 can be realized or be performed
Disclosed each method, step and logic diagram in the embodiment of the present invention.General processor can be microprocessor or this at
It can also be any conventional processor etc. to manage device.The step of method with reference to disclosed in the embodiment of the present invention, can directly embody
Completion is performed for hardware decoding processor or performs completion with the hardware in decoding processor and software module combination.Software
Module can be located at random access memory, flash memory, read-only memory, programmable read only memory or electrically erasable programmable storage
In the storage medium of this fields such as device, register maturation.The storage medium is located at storage unit 810, such as processor 830 is readable
The step of taking the information in storage unit 810, the above method completed with reference to its hardware.
W historical data of telecommunication service event is obtained, and a training data is selected from the w historical data,
In, each historical data in the w historical data includes KPI data and QoE data;
QoE ranks belonging to the QoE data of each training data in a training data, by described a instruction
Practice data and be divided into k station work data;
Regression model is trained respectively using the k station work data, mould is returned so as to obtain k training
Type,
Optimal trained regression model is selected from described k trained regression model, wherein, w, a, k are more than zero
Integer, w>a≥k.
Optionally, b verification data are selected from the w historical data, wherein, the b verification data and institute
It states a training data and is belonging respectively to different data acquisition systems;Described k training is returned respectively using described b verification data
Each training regression model in model is verified, mould is returned with each training obtained in described k trained regression model
The verification result of type;The verification result of each training regression model in described k trained regression model is instructed from described k
Practice regression model and select optimal trained regression model.
Optionally, c test data is selected from the w historical data, wherein, it is the c test data, described
B verification data are belonging respectively to different data acquisition systems from a training data;According in described k trained regression model
The verification result of each training regression model select trained regression model undetermined from described k trained regression model;It uses
The c test data is tested to obtain test result to the trained regression model undetermined;It is big in the test result
In the case of predetermined threshold value, the trained regression model undetermined is set as optimal trained regression model.
Optionally, in the case where the test result is less than or equal to predetermined threshold value, the training undetermined is returned
Model abandons.
Optionally, the regression model is multinomial Logic Regression Models.
Optionally, the granularity of the telecommunication service event is user setting.
Optionally, the KPI data includes at least one of bandwidth availability ratio, time delay, shake and packet loss.
One of ordinary skill in the art will appreciate that realizing all or part of flow in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in computer read/write memory medium
In, the program is when being executed, it may include such as the flow of the embodiment of above-mentioned each method.And aforementioned storage medium includes:ROM、
The various media that can store program code such as RAM, magnetic disc or CD.
Claims (14)
1. a kind of KPI and QoE regression model method for building up, which is characterized in that including:
W historical data of telecommunication service event is obtained, and a training data is selected from the w historical data, wherein,
Each historical data in the w historical data includes KPI data and QoE data;
QoE ranks belonging to the QoE data of each training data in a training data, by described a trained number
According to being divided into k station work data;
Regression model is trained respectively using the k station work data, so as to obtain k trained regression model,
Optimal trained regression model is selected from described k trained regression model, wherein, w, a, k are the integer more than zero,
w>a≥k。
2. according to the method described in claim 1, it is characterized in that,
Include before optimal trained regression model is selected from described k trained regression model:
B verification data are selected from the w historical data, wherein, the b verification data and described a trained number
According to being belonging respectively to different data acquisition systems;
Optimal trained regression model is selected from described k trained regression model to include:
Each regression model is trained to verify in trained regression model to described k respectively using described b verification data,
To obtain the verification result of each training regression model in described k trained regression model;
The verification result of each training regression model in described k trained regression model returns mould from described k training
Type selects optimal trained regression model.
3. according to the method described in claim 2, it is characterized in that, each training in described k trained regression model
The verification result of regression model includes before selecting optimal trained regression model from described k trained regression model:
C test data is selected from the w historical data, wherein, the c test data, the b verification data
Different data acquisition systems are belonging respectively to from a training data;
The verification result of each training regression model in described k trained regression model returns mould from described k training
Type is selected optimal trained regression model and is included:
The verification result of each training regression model in described k trained regression model returns mould from described k training
Type selects trained regression model undetermined;
The trained regression model undetermined is tested to obtain test result using the c test data;
In the case where the test result is more than predetermined threshold value, the trained regression model undetermined is set as optimal training and is returned
Return model.
4. according to the method described in claim 3, it is characterized in that,
In the case where the test result is less than or equal to predetermined threshold value, the trained regression model undetermined is abandoned.
5. according to the method described in Claims 1-4 any claim, which is characterized in that the regression model is multinomial
Logic Regression Models.
6. according to the method described in claim 1 to 5 any claim, which is characterized in that the grain of the telecommunication service event
It spends for user setting.
7. according to the method described in claim 1 to 6 any claim, which is characterized in that the KPI data includes bandwidth profit
With at least one of rate, time delay, shake and packet loss.
8. a kind of KPI establishes device with QoE regression models, which is characterized in that including:Acquisition module, sort module, training module
And selecting module,
The acquisition module is used to obtain w historical data of telecommunication service event, and selects a from the w historical data
A training data, wherein, each historical data in the w historical data includes KPI data and QoE data;
QoE grade of the sort module belonging to for the QoE data of each training data in a training data
Not, a training data is divided into k station work data;
The training module is for being respectively trained regression model using the k station work data, so as to obtain k
Training regression model,
The selecting module is used to select optimal trained regression model from described k trained regression model, wherein, w, a, k
It is the integer more than zero, w>a≥k.
9. device according to claim 8, which is characterized in that the selecting module further includes:Authentication unit and selection
Unit,
Acquisition module is additionally operable to select b verification data from the w historical data, wherein, the b verification data and
The a training data is belonging respectively to different data acquisition systems;
The authentication unit is used for using described b verification data each training to described k in trained regression model respectively
Regression model is verified, to obtain the verification result of each training regression model in described k trained regression model;
The selecting unit for each trained regression model in described k trained regression model verification result from institute
It states k trained regression model and selects optimal trained regression model.
10. device according to claim 9, which is characterized in that the selecting unit further includes selecting unit undetermined, test
Unit and setting unit,
The acquisition module is used to select c test data from the w historical data, wherein, the c test number
According to, it is described b verification data be belonging respectively to different data acquisition systems from a training data;
The verification result of each training regression model of the selecting unit undetermined in described k trained regression model is from institute
It states k trained regression model and selects trained regression model undetermined;
The test cell is used to that the trained regression model undetermined to be tested to be surveyed using the c test data
Test result;
The setting unit is used in the case where the test result is more than predetermined threshold value, by the trained regression model undetermined
It is set as optimal trained regression model.
11. device according to claim 10, which is characterized in that described device further includes selecting unit,
The discarding unit is used in the case where the test result is less than or equal to predetermined threshold value, by the training undetermined
Regression model abandons.
12. according to the device described in claim 8 to 11 any claim, which is characterized in that the regression model is multinomial
Formula Logic Regression Models.
13. according to the device described in claim 8 to 12 any claim, which is characterized in that the telecommunication service event
Granularity is user setting.
14. according to the device described in claim 8 to 13 any claim, which is characterized in that the KPI data includes bandwidth
At least one of utilization rate, time delay, shake and packet loss.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611186034.XA CN108205520A (en) | 2016-12-20 | 2016-12-20 | KPI and QoE regression models method for building up and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611186034.XA CN108205520A (en) | 2016-12-20 | 2016-12-20 | KPI and QoE regression models method for building up and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108205520A true CN108205520A (en) | 2018-06-26 |
Family
ID=62603363
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611186034.XA Pending CN108205520A (en) | 2016-12-20 | 2016-12-20 | KPI and QoE regression models method for building up and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108205520A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110855669A (en) * | 2019-11-14 | 2020-02-28 | 北京理工大学 | Video QoE index prediction method suitable for encrypted flow based on neural network |
CN111954232A (en) * | 2019-05-16 | 2020-11-17 | 大唐移动通信设备有限公司 | Wireless network index classification method and device |
CN112311935A (en) * | 2019-07-26 | 2021-02-02 | 中国移动通信集团广东有限公司 | Abnormity detection method and device and electronic equipment |
US11922279B2 (en) | 2020-06-12 | 2024-03-05 | International Business Machines Corporation | Standard error of prediction of performance in artificial intelligence model |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103312531A (en) * | 2012-03-15 | 2013-09-18 | 华为技术有限公司 | Quality of experience (QOE) acquiring method, device and QOE guaranteeing method and device |
CN103389310A (en) * | 2013-07-31 | 2013-11-13 | 哈尔滨工业大学 | Online sub-pixel optical component damage detection method based on radiation calibration |
CN105554782A (en) * | 2015-12-09 | 2016-05-04 | 中国联合网络通信集团有限公司 | Prediction method and device for user perception index |
-
2016
- 2016-12-20 CN CN201611186034.XA patent/CN108205520A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103312531A (en) * | 2012-03-15 | 2013-09-18 | 华为技术有限公司 | Quality of experience (QOE) acquiring method, device and QOE guaranteeing method and device |
CN103389310A (en) * | 2013-07-31 | 2013-11-13 | 哈尔滨工业大学 | Online sub-pixel optical component damage detection method based on radiation calibration |
CN105554782A (en) * | 2015-12-09 | 2016-05-04 | 中国联合网络通信集团有限公司 | Prediction method and device for user perception index |
Non-Patent Citations (2)
Title |
---|
陈邓: "移动互联网新型数据业务QOE评价方法的研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
顾明亮等: "基于联合多样性密度的汉语方言辨识", 《计算机工程与应用》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111954232A (en) * | 2019-05-16 | 2020-11-17 | 大唐移动通信设备有限公司 | Wireless network index classification method and device |
CN111954232B (en) * | 2019-05-16 | 2022-01-28 | 大唐移动通信设备有限公司 | Wireless network index classification method and device |
CN112311935A (en) * | 2019-07-26 | 2021-02-02 | 中国移动通信集团广东有限公司 | Abnormity detection method and device and electronic equipment |
CN110855669A (en) * | 2019-11-14 | 2020-02-28 | 北京理工大学 | Video QoE index prediction method suitable for encrypted flow based on neural network |
CN110855669B (en) * | 2019-11-14 | 2020-10-09 | 北京理工大学 | Video QoE index prediction method suitable for encrypted flow based on neural network |
US11922279B2 (en) | 2020-06-12 | 2024-03-05 | International Business Machines Corporation | Standard error of prediction of performance in artificial intelligence model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108205520A (en) | KPI and QoE regression models method for building up and device | |
US8762475B2 (en) | Simultaneous instant messaging in single window | |
US9320006B2 (en) | Categorized location identification based on historical locations of a user device | |
US9894208B2 (en) | Connection routing system | |
CN104883657B (en) | A kind of virtual card management method and device | |
CN101119321A (en) | Network flux classification processing method and apparatus | |
CN105579976B (en) | Webpage automated testing method and device | |
CN108289053A (en) | Control method, the device and system of instant communication session | |
CN107908686B (en) | Information pushing method and device, server and readable storage medium | |
US10142924B2 (en) | Method and apparatus for generating recommended changes to communication behaviors | |
CN111654321B (en) | Satellite network management method and device and electronic equipment | |
CN106211229B (en) | Intelligent accelerated method, apparatus and system | |
CN106203989A (en) | A kind of information processing method and device | |
CN106982158A (en) | The sharing method and device of movement locus | |
CN110191131A (en) | A kind of file sharing method, device, system, server and storage medium | |
CN105337841A (en) | Information processing method and system, client, and server | |
CN109218370A (en) | Message method and device, electronic equipment | |
CN107659982B (en) | Wireless network access point classification method and device | |
CN108763017A (en) | Application of software data processing method, server-side and the storage medium of financial business | |
CN109039670A (en) | The sharing method and device of team's configuration method, team's allocation plan | |
CN106998386B (en) | Method and device for adding contact way and user terminal | |
CN109462683A (en) | Number sorted method, apparatus, storage medium and computer equipment | |
CN106330563B (en) | A kind of method and device of determining Intranet http communication stream service type | |
EP3318021B1 (en) | Managing and indexing communication data, and recommending communication type | |
CN110334273A (en) | Service search method, apparatus and computer equipment based on universal search platform |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180626 |