CN104052612B - A kind of Fault Identification of telecommunication service and the method and system of positioning - Google Patents
A kind of Fault Identification of telecommunication service and the method and system of positioning Download PDFInfo
- Publication number
- CN104052612B CN104052612B CN201310080584.3A CN201310080584A CN104052612B CN 104052612 B CN104052612 B CN 104052612B CN 201310080584 A CN201310080584 A CN 201310080584A CN 104052612 B CN104052612 B CN 104052612B
- Authority
- CN
- China
- Prior art keywords
- feature value
- failure
- service feature
- grader
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Test And Diagnosis Of Digital Computers (AREA)
Abstract
The invention provides a kind of Fault Identification of telecommunication service and the method and system of positioning, methods described includes:According to the coefficient correlation between the service feature value and phenomenon of the failure in representing fault source, the service feature value of back transfer BP neural network training is selected as from sample data;A Weak Classifier based on BP neural network is determined for each service feature value for having selected, and Weak Classifier is subjected to linear combination, corresponding strong classifier is obtained;According to the size order of the coefficient correlation between each service feature value and phenomenon of the failure, the corresponding strong classifier of each service feature value is connected, and the strong classifier after series connection is trained, the layer grader for multiple faults identifing source is obtained;Fault Identification and positioning are carried out to the business datum and/or network data that are produced during telecommunication service using the layer grader.The present invention can reduce the training sample amount needed for neutral net, improve fault detect real-time, and be easy to position failure cause.
Description
Technical field
The present invention relates to the core net in communication network, webmaster and business support field, and in particular to a kind of telecommunication service
Fault Identification with positioning method and system.
Background technology
With the development of telecommunications industry and information services industry, telecommunication traffic also rapid growth, the number of faults occurred
Amount will also be lifted therewith, and this turns into glutinous between lifting user satisfaction, the communication service for strengthening operator company and user
One huge obstacle of property.
Currently, an important step for solving failure is to carry out Fault Identification and positioning, determines failure.The step for it is usual
Implementation method be:Manual identified failure, i.e., according to business datum and the experience of maintenance staff, judge which aspect is present
Failure, the flow specifically judged is illustrated in fig. 1 shown below.
In Fig. 1, OCS is Online Charging System(Online Charging System);HLR is home location register system
(Host Location Register);BOSS is Business Operating Support System.Above three system
It is referred to as operation system.The data produced in operation system running are sampled, business datum can be obtained.Work as business
When system breaks down, current processing mode is:The contact staff on foreground is received after the fault message of client's submission, by failure
Processing task gives background maintenance personnel, and background maintenance personnel confirm event according to the information of client and foreground contact staff offer
The reason for hindering, and position failure, then solve problem.This is a kind of passive, delay pattern, i.e., must be after customer complaint
Or foreground could find failure when providing related error information.Obviously, the mode of this handling failure would not be timely, from
And the satisfaction of client is had influence on, degree of recognition of the reduction user to carrier service.
Neutral net is also known as artificial neural network, nerve and calculated, connectionism artificial intelligence, Serial Distribution Processing etc..
One neutral net is a large-scale Serial Distribution Processing device being made up of simple process member, with storage Heuristics
Be allowed to available characteristic.By using the neutral net shown in Fig. 2, failure can in real time be detected, to shorten failure
The time handled during generation.
In Fig. 2, the data that system is gathered from OCS, HLR, BOSS and network are inputted in real time by neural network input layer
Into trained neutral net, according to the species of the failure of prediction, it is defeated that the different input neurons of output layer are produced
Enter to show the different system failures.
The Fault Identification mode of use manual identified shown in Fig. 1 is a kind of passive, delay pattern, i.e., must be in visitor
Family complain after or foreground have correlation when reporting an error just go discovery failure, then so handling failure would not be timely, so that shadow
Ring approval of the user to service.And the Fault Model based on neutral net shown in Fig. 2 then has problems with:
(1)All failures are detected all in a neutral net, cause, this resistance very big for the sample data volume of training
The practicality of the class model is hindered.
The input of neutral net for detecting failure has accessed all system operation datas, and input then produce it is all
The fault-signal of representing fault phenomenon.In training, the generation of any kind phenomenon is required for the data of all inputs, from defeated
Enter and the dimension that exports from the point of view of, the sample data volume required for determining any kind failure is all very big, and this makes to train
Good such a detection neutral net needs substantial amounts of sample data, and this is for a fault detect system for being in first stage of construction
For system and the just travelling mechanism of concern fault detect, difficulty is larger.
(2)For the neutral net trained, because its is complicated, while to calculate all kinds of failures, cause meter
The big great increase of calculation amount, this make it that the delay detected in real time is larger.
The purpose of fault detect is the response time for shortening failure as far as possible, therefore, including Fault Identification and fault location
The real-time of the fault detect of two processes is extremely important.If neural network structure is complicated, calculative complexity increases
Greatly, calculating the time will extend, and this will be unfavorable for shortening the time of Fault Identification and fault location.
(3)For recognizing and positioning the detection neutral net of failure not in phenomenon of the failure(Output)And failure cause(It is defeated
Enter)Between set up clear and definite relation, so cause to be not easy to position failure cause.
The content of the invention
In view of this, the purpose of the embodiment of the present invention be to provide a kind of Fault Identification of telecommunication service with positioning method and
System, can reduce the training sample amount needed for neutral net, improve fault detect real-time, and be easy to position failure cause.
In order to solve the above technical problems, offer of embodiment of the present invention scheme is as follows:
The embodiments of the invention provide a kind of Fault Identification of telecommunication service and the method for positioning, including:
According to the coefficient correlation between the service feature value and phenomenon of the failure in representing fault source, phase is selected from sample data
Relation number is more than the service feature value of predetermined threshold value, the service feature value trained as back transfer BP neural network;
A Weak Classifier based on BP neural network is determined for each service feature value for having selected, and will be determined
Weak Classifier carry out linear combination, obtain the corresponding strong classifier of the service feature value;
According to the size order of the coefficient correlation between each service feature value and phenomenon of the failure, by each service feature value
Corresponding strong classifier series connection, and the sample data for including multiple faults source data and phenomenon of the failure is used to the strong classification after series connection
Device is trained, and obtains the layer grader for multiple faults identifing source;
Failure knowledge is carried out to the business datum and/or network data that are produced during telecommunication service using the layer grader
Other and positioning.
Further, in such scheme, it is described using the layer grader to the business number that is produced during telecommunication service
According to and/or network data carry out Fault Identification and positioning, including:By the business datum and/or net that are produced during telecommunication service
Layer grader described in network data input, the output produced according to each strong classifier positions failure cause.
Further, in such scheme, each service feature value to have selected determines one based on BP nerves
The Weak Classifier of network, including:
For each service feature value selected, a Weak Classifier based on BP neural network is set up;
Using the combination of the sample data comprising positive and negative business datum being previously obtained, the single event of Weak Classifier identification is trained
Barrier source, until predetermined frequency of training is reached, the Weak Classifier after being trained, wherein, estimate instruction after often performing once training
Practice target error, the minimum grader of selection wherein error simultaneously reduces the weights of the grader.
Further, it is described to be distinguished using the sample data for including multiple faults source data and phenomenon of the failure in such scheme
Strong classifier after series connection is trained, the layer grader for multiple faults identifing source is obtained, including:
Using the sample data of multiple faults source data and phenomenon of the failure is included, the strong classifier after series connection is trained,
Until reaching predetermined frequency of training, the layer grader is obtained, wherein, often perform estimation training objective after once training and miss
Difference, the minimum grader of selection wherein error simultaneously reduces the weights of the grader.
Further, in such scheme, further the coefficient correlation between selection and phenomenon of the failure is more than predetermined threshold value
Service feature value, the series connection is carried out by the corresponding strong classifier of selected service feature value.
The embodiment of the present invention additionally provides a kind of Fault Identification of telecommunication service and the system of positioning, including:
Eigenvalue unit, for the phase relation between the service feature value and phenomenon of the failure according to representing fault source
Number, selects coefficient correlation to be more than the service feature value of predetermined threshold value from sample data, is instructed as back transfer BP neural network
Experienced service feature value;
A single fault identifing source unit, for determining that is based on a BP neural network for each service feature value for having selected
Weak Classifier, and by identified Weak Classifier carry out linear combination, obtain the service feature value it is corresponding one by force classify
Device;
Multiple faults identifing source unit, for the size according to the coefficient correlation between each service feature value and phenomenon of the failure
Sequentially, the corresponding strong classifier of each service feature value is connected, and use includes the sample of multiple faults source data and phenomenon of the failure
Notebook data is trained to the strong classifier after series connection, obtains the layer grader for multiple faults identifing source;
Service Processing Unit, for using the layer grader to the business datum that is produced during telecommunication service and/or
Network data carries out Fault Identification and positioning.
Further, in such scheme, the Service Processing Unit, the business that will further be produced during telecommunication service
Data and/or network data input the layer grader, and the output produced according to each strong classifier positions failure cause.
Further, in such scheme, the single fault identifing source unit, each industry for being further used for having selected
Business characteristic value, sets up a Weak Classifier based on BP neural network;Using the sample for including positive and negative business datum being previously obtained
Notebook data is combined, and trains the Weak Classifier identification single fault source, until predetermined frequency of training is reached, it is weak after being trained
Grader, wherein, training objective error is estimated after often performing once training, the wherein minimum grader of error is selected and reduction should
The weights of grader.
Further, in such scheme, the multiple faults identifing source unit, being further used for using includes multiple faults source number
According to the sample data with phenomenon of the failure, the strong classifier after series connection is trained, until reaching predetermined frequency of training, obtained
The layer grader, wherein, often perform and training objective error is estimated after once training, the minimum grader of selection wherein error is simultaneously
Reduce the weights of the grader.
Further, in such scheme, the multiple faults identifing source unit is further used between selection and phenomenon of the failure
Coefficient correlation be more than the service feature value of predetermined threshold value, the corresponding strong classifier of selected service feature value is carried out described in
Series connection.
From it is described above as can be seen that telecommunication service provided in an embodiment of the present invention Fault Identification with positioning method and
System, can accurately recognize and be out of order and position, it is possible to reduce the artificial workload for checking business datum.The embodiment of the present invention
Also have the advantages that training sample amount is little, fault detect real-time is high and failure cause positioning is quick.
Brief description of the drawings
Fig. 1 is the schematic diagram of manual identified failure in the prior art;
Fig. 2 is the schematic diagram that application neutral net carries out Fault Identification;
Fig. 3 is the schematic flow sheet of methods described of the embodiment of the present invention;
Fig. 4 is the structural representation of system described in the embodiment of the present invention;
Fig. 5 is the schematic diagram for being laminated Artificial neural network ensemble structure in the embodiment of the present invention;
Fig. 6 is the structural representation of layer grader provided in an embodiment of the present invention.
Embodiment
Evade from existing failure(Failure is solved)Experience from the point of view of, be easiest to reduce user satisfaction link, be also
It is fault location most to spend time taking link, i.e., the reason for can not being quickly found failure, Failure elimination.Therefore, if energy
Failure, positioning failure can be recognized in a short time, and this solves the time of failure by greatly shortening, and improves the satisfaction of user
Degree.Meanwhile, if establishing clear and definite relation between phenomenon of the failure and failure cause, the time is solved for shortening failure, is also had
There is very important meaning.
Based on more than analyze, the embodiments of the invention provide a kind of Fault Identification of telecommunication service with positioning method and be
System, can reduce required training sample amount, improve fault detect real-time, and be easy to position failure cause.The present invention is implemented
Example method, be capable of active go analyze existing business datum or network data, so as to find failure, and quickly determine
Failure cause, to help attendant to quickly cope with failure problems.Here, business datum is the number exported from operation system
According to network data is communication network(Such as access network, core net)The data produced in running.
It is right below in conjunction with the accompanying drawings and the specific embodiments to make the object, technical solutions and advantages of the present invention clearer
The present invention is described in detail.
It refer to Fig. 3, the Fault Identification and the method for positioning of the telecommunication service described in the embodiment of the present invention, including following step
Suddenly:
Step 31, according to the coefficient correlation between the service feature value and phenomenon of the failure in representing fault source, from sample data
Middle selection coefficient correlation is more than the service feature value of predetermined threshold value, is used as back transfer (BP, Back Propagation) nerve
The service feature value of network training.
In this step, dimension-reduction treatment is carried out to sample data by coefficient correlation, suitable characteristic value is chosen and carries out subsequently
Calculate, to reduce amount of calculation, improve calculating speed.
Step 32, it is that each service feature value selected determines one based on BP nerves according to AdaBoost algorithms
The Weak Classifier of network, and identified Weak Classifier is subjected to linear combination, obtain the service feature value corresponding one strong
Grader.
In this step, it is each service feature value selected, sets up a Weak Classifier based on BP neural network;
Using the combination of the sample data comprising positive and negative business datum being previously obtained, the Weak Classifier identification single fault source is trained, directly
To reaching predetermined frequency of training, the Weak Classifier after being trained, wherein, often perform estimation training objective after once training and miss
Difference, the minimum grader of selection wherein error simultaneously reduces the weights of the grader;Then, line is entered to the Weak Classifier after training
The combination of property, obtains a strong classifier.
Step 33, according to the size order of the coefficient correlation between each service feature value and phenomenon of the failure, by each industry
Characteristic value of being engaged in corresponding strong classifier series connection, and using the sample data of multiple faults source data and phenomenon of the failure is included to series connection after
Strong classifier be trained, obtain the layer grader for multiple faults identifing source.
In this step, further the coefficient correlation between selection and phenomenon of the failure is more than the service feature value of predetermined threshold value,
The corresponding strong classifier of selected service feature value is subjected to the series connection.Then, by using including multiple faults source data
With the sample data of phenomenon of the failure, the strong classifier after series connection is trained, until reaching predetermined frequency of training, institute is obtained
A layer grader is stated, wherein, often perform and training objective error is estimated after once training, the minimum grader of selection wherein error and drop
The weights of the low grader.
Step 34, the business datum and/or network data that are produced during telecommunication service are entered using the layer grader
Row Fault Identification and positioning.
In this step, the layer point is inputted by the business datum and/or network data that will be produced during telecommunication service
Class device, the output then produced according to each strong classifier positions failure cause.
From above method as can be seen that the embodiment of the present invention is by correlation analysis, to the Data Dimensionality Reduction trained and detected,
Reduce the requirement to training samples number so that the fault detection technique proposed in the embodiment of the present invention is pratical and feasible.Moreover,
The design arrangement for the tracer that the embodiment of the present invention passes through layering, simple calculating is resolved into by complicated calculating, is improved
The real-time of fault detect.Finally, the embodiments of the invention provide a kind of layer grader of hierarchy, this layer of grader
The output of each strong classifier corresponds to a failure, so, can be navigated to when some strong classifier produces output pair
The failure cause answered, improves the speed of failure cause positioning.
Based on above method, the embodiment of the present invention has also correspondingly provided a kind of Fault Identification of telecommunication service and positioning
System, as shown in figure 4, the system includes:
Eigenvalue unit, for the phase relation between the service feature value and phenomenon of the failure according to representing fault source
Number, selects coefficient correlation to be more than the service feature value of predetermined threshold value from sample data, is instructed as back transfer BP neural network
Experienced service feature value;
Single fault identifing source unit, for according to AdaBoost algorithms, being that each service feature value selected is determined
One Weak Classifier based on BP neural network, and identified Weak Classifier is subjected to linear combination, obtain the service feature
It is worth a corresponding strong classifier;
Multiple faults identifing source unit, for the size according to the coefficient correlation between each service feature value and phenomenon of the failure
Sequentially, the corresponding strong classifier of each service feature value is connected, and use includes the sample of multiple faults source data and phenomenon of the failure
Notebook data is trained to the strong classifier after series connection, obtains the layer grader for multiple faults identifing source;
Service Processing Unit, for using the layer grader to the business datum that is produced during telecommunication service and/or
Network data carries out Fault Identification and positioning.
Preferably, the Service Processing Unit, further by the business datum and/or network that are produced during telecommunication service
Layer grader described in data input, the output produced according to each strong classifier positions failure cause.
Preferably, the single fault identifing source unit, each service feature value for being further used for having selected is set up
One Weak Classifier based on BP neural network;Using the combination of the sample data comprising positive and negative business datum being previously obtained, instruction
Practice the Weak Classifier identification single fault source, until predetermined frequency of training is reached, the Weak Classifier after being trained, wherein,
Often perform and training objective error is estimated after once training, the minimum grader of selection wherein error simultaneously reduces the power of the grader
Value.
Preferably, the multiple faults identifing source unit, being further used for using includes multiple faults source data and phenomenon of the failure
Sample data, the strong classifier after series connection is trained, until reach predetermined frequency of training, the layer classification is obtained
Device, wherein, often perform and training objective error is estimated after once training, the minimum grader of selection wherein error simultaneously reduces the classification
The weights of device.
Preferably, the multiple faults identifing source unit, the coefficient correlation being further used between selection and phenomenon of the failure is big
In the service feature value of predetermined threshold value, the corresponding strong classifier of selected service feature value is subjected to the series connection.
To more fully understand above method, the specific embodiment of the present invention will be made by more detailed description below
It is expanded on further.
Fig. 5 is refer to, Fig. 5 provides the stacking Artificial neural network ensemble that the embodiment of the present invention is used for identification and the positioning of failure
Structure.The Artificial neural network ensemble structure mainly realizes three functions:(1)The correlation of the source of trouble and phenomenon of the failure data is analyzed,
Point phenomenon of the failure determines sample data;(2)Train neutral net(Grader)For single fault source(Single eigenvalue)Identification;
(3)Integrated neural network, forms cascade filtering, for multiple faults source(Multiple characteristic values)Identification.Illustrate individually below.
(1)Correlation analysis, determines the input feature vector value of sample data
The characteristic value of sample data(Dimension)Numerous, in order to reduce the amount of calculation of neutral net, the embodiment of the present invention is to sample
The input of notebook data carries out dimension-reduction treatment, that is, selects suitable characteristic value.
As a kind of preferred embodiment, characteristic value is selected using the correlation analysis of equation below.
In formula, X is characteristic value to be selected, and Y is the output valve relevant with failure.For example, X is empty for the residue of server hard disc
Between ratio with normed space, Y is the number of times of interior inquiry per second.Subscript i represents different time points, and N represents that sample data is adopted
The quantity of sample,Represent XiAverage value,Represent YiAverage value, rXYRepresent the correlation between X and Y.
In this step, the present embodiment selection rXYFor 0.5 as threshold value, i.e. characteristic value of the selection more than or equal to 0.5, enter
Enter follow-up training to calculate.
(2)Single fault identifing source device
In the present embodiment, using BP neural network, setting up one has three layers of network model:Input layer, hidden layer, output
Layer.First, the business datum or network data as sample data being obtained ahead of time are inputted in input layer, by the effect of hidden layer
The effect of function, just can be with output valve in output end, and in the present embodiment, output valve only has two kinds:1 and 0, wherein 1 represents the number
Represent and do not have according to faulty, 0, error is then calculated according to error calculation function, compared with error expected, if be not reaching to
The requirement of error expected, by the feedback of the information exported just now to input layer and hidden layer, so as to act on input layer and hidden layer, is adjusted
Weights and related function, to reach the effect of optimization output valve.
It is theoretical according to Artificial neural network ensemble, synthesized by individually training multiple neutral nets, and by its result, can be notable
Ground improves the generalization ability of neutral net, and in the present embodiment, this generalization ability is Fault Identification ability.Here use
As a result synthetic method is AdaBoost algorithms, is that each characteristic value determines a Weak Classifier, then by these weak typings
Device carries out linear combination, just constitutes a kind of strong classifier.In the algorithm, each service feature value is endowed a power
Value, if some sample is not classified correctly, its weights will be enhanced, and conversely can then be reduced, and its arthmetic statement is as follows:
1) selected characteristic value is directed to, the sample of a number of positive and negative business datum is obtained, training grader is used as
Sample data, it is assumed here that there is positive and negative data sample respectively to have some.
2) weight w of initialization training classifiers combination, by positive and negative business datum sample according to default different proportion(Example
Such as, 10:1、10:2、10:3、…、10:10)Carry out the multiple sample combinations of sample combination producing.Generally, negative sample is fewer than positive sample
A lot, therefore, do not reduced per the positive sample of class sample combination, and the generation of negative sample can be entered by the way of repeatedly generating
Row sample size polishing;
3) vertical a grader is built jointly for each sample group(BP neural network algorithm), combine, carry out for each sample
Pre-determined number(Such as 100 training), then, simulation error estimation is carried out to the grader for having performed training, error is:Train mould
Intend output and the difference of reality output;
4) all samples are selected to combine the grader of the minimum predetermined number of error in corresponding grader, such as selection is missed
Poor 3 minimum graders;
5)Using the simulation correct recognition rata of grader as combining weights, linear group is carried out to the grader of above-mentioned selection
Close, obtain the corresponding strong classifier of service feature value.
A strong classifier is can be obtained by by process above, what the grader can be quickly tells the data
Whether failure, improve the efficiency of Fault Identification, but can only be for having a case that the failure of single features value.
(3)Multiple faults identifing source device
Single strong classifier is previously obtained, and when the quantity of characteristic value is more than one, it is impossible to use above
Single fault identifing source device.Therefore, the model of a kind of layer of grader as shown in Figure 6 is present embodiments provided, this layer of grader is
With the mode of series connection by multiple neutral nets(Strong classifier i.e. above)It is combined.Each nerve is arranged according to coefficient correlation
Position of the network in layer grader.Distinguished using all source of trouble X and phenomenon of the failure Y of multiple faults source data and normal data
Each neutral net is trained.Concretely comprise the following steps:
1)Using(1)In correlation analysis the step of, selection more than correlation coefficient threshold 0.5 all and a certain failure
Characteristic value X related phenomenon Y;One phenomenon of the failure may be caused by several sources of trouble, and a source of trouble corresponds to a spy
Value indicative;Here a certain phenomenon of the failure Y can be from the phenomenon of the failure for needing to position optionally.
2)By the size order of the coefficient correlation between each characteristic value and selected phenomenon of the failure Y, by each feature
The represented corresponding neutral net of the source of trouble of value is discharged into a layer grader(As shown in Figure 6)In;The size of coefficient correlation is reflected
Cause the size of the possibility of the source of trouble of phenomenon of the failure;The big neutral net of coefficient correlation, sorts in closer input
Position(Such as the position more kept right in Fig. 6);
3)Neutral net is trained with sample data respectively, the process of training with(2)Single fault identifing source device 1)、2)
With 3)Step is identical.
Arranged by such, each layer of fault grader can just exclude a source of trouble, the higher grader of level
(Sort closer to the grader of input end position), its corresponding characteristic value causes the possibility of failure bigger, therefore, this kind of group
Conjunction efficiently can detect and position failure, first layer grader only used one and can just filter out be not failure letter
Breath, the remaining business datum not filtered out then gone to perform by other graders successively, thus can be very high
Effect ground identification failure and the output positioning failure cause produced by neutral net(The source of trouble).
Through testing the conclusion drawn:Identification failure using the hierarchy than the single Neural structure without layering
Efficiency can averagely improve several times.
To sum up, the embodiment of the present invention is according to the demand and its feature of the fault location of telecommunication service, it is proposed that grader
Cascade mode, in the selection of service feature value, the embodiment of the present invention is by the phenomenon of the failure and the source of trouble in business datum(Failure
Reason)Association analysis is done between representative data, the investigation scope of failure is determined(Coefficient correlation is selected to be more than the feature of threshold value
Value), so as to reduce dimension, reduce the computation complexity subsequently calculated, improve calculating speed;Also, the embodiment of the present invention
Using the hierarchy schema of grader, the advantage of hierarchy is:(1)Treatment effeciency is high.The place of hierarchy grader
Reason efficiency is higher than the grader being made up of single Neural, experiments verify that:Input identical characteristic value, hierarchy schema
The efficiency of sorter model can improve several times;(2)Quick positioning failure:Output by neutral net in hierarchy is
The source of trouble can be positioned.
Described above is only embodiments of the present invention, it is noted that come for those skilled in the art
Say, under the premise without departing from the principles of the invention, some improvements and modifications can also be made, these improvements and modifications also should be regarded as
Protection scope of the present invention.
Claims (10)
1. a kind of Fault Identification of telecommunication service and the method for positioning, it is characterised in that including:
According to the coefficient correlation between the service feature value and phenomenon of the failure in representing fault source, phase relation is selected from sample data
Number is more than the service feature value of predetermined threshold value, the service feature value trained as back transfer BP neural network;
A Weak Classifier based on BP neural network is determined for each service feature value for having selected, and will be identified weak
Grader carries out linear combination, obtains the corresponding strong classifier of the service feature value;
According to the size order of the coefficient correlation between each service feature value and phenomenon of the failure, by each service feature value correspondence
Strong classifier series connection, and using including multiple faults source data and the sample data of phenomenon of the failure is entered to the strong classifier after series connection
Row training, obtains the layer grader for multiple faults identifing source;
Using the layer grader business datum and/or network data that are produced during telecommunication service are carried out Fault Identification and
Positioning.
2. the method as described in claim 1, it is characterised in that described to utilize the layer grader to being produced during telecommunication service
Raw business datum and/or network data carry out Fault Identification and positioning, including:By the business number produced during telecommunication service
According to and/or network data input the layer grader, the output produced according to each strong classifier positions failure cause.
3. the method as described in claim 1, it is characterised in that each service feature value to have selected determines one
Weak Classifier based on BP neural network, including:
For each service feature value selected, a Weak Classifier based on BP neural network is set up;
Using the combination of the sample data comprising positive and negative business datum being previously obtained, the Weak Classifier identification single fault is trained
Source, until predetermined frequency of training is reached, the Weak Classifier after being trained, wherein, estimate training after often performing once training
Target error, the minimum grader of selection wherein error simultaneously reduces the weights of the grader.
4. the method as described in claim 1, it is characterised in that described use includes the sample of multiple faults source data and phenomenon of the failure
Notebook data is trained to the strong classifier after series connection respectively, obtains the layer grader for multiple faults identifing source, including:
Using the sample data of multiple faults source data and phenomenon of the failure is included, the strong classifier after series connection is trained, until
Predetermined frequency of training is reached, the layer grader is obtained, wherein, often perform and training objective error, choosing are estimated after once training
Select the wherein minimum strong classifier of error and reduce the weights of the strong classifier.
5. the method as described in claim 1, it is characterised in that further the coefficient correlation between selection and phenomenon of the failure is more than
The service feature value of predetermined threshold value, the series connection is carried out by the corresponding strong classifier of selected service feature value.
6. a kind of Fault Identification of telecommunication service and the system of positioning, it is characterised in that including:
Eigenvalue unit, for the coefficient correlation between the service feature value and phenomenon of the failure according to representing fault source, from
Coefficient correlation is selected to be more than the service feature value of predetermined threshold value, the industry trained as back transfer BP neural network in sample data
Business characteristic value;
Single fault identifing source unit, for determining one based on the weak of BP neural network for each service feature value for having selected
Grader, and identified Weak Classifier is subjected to linear combination, obtain the corresponding strong classifier of the service feature value;
Multiple faults identifing source unit, it is suitable for the size according to the coefficient correlation between each service feature value and phenomenon of the failure
Sequence, the corresponding strong classifier of each service feature value is connected, and use includes the sample of multiple faults source data and phenomenon of the failure
Data are trained to the strong classifier after series connection, obtain the layer grader for multiple faults identifing source;
Service Processing Unit, for using the layer grader to the business datum and/or network that are produced during telecommunication service
Data carry out Fault Identification and positioning.
7. system as claimed in claim 6, it is characterised in that the Service Processing Unit, further by telecommunication service process
The business datum and/or network data of middle generation input the layer grader, the output produced according to each strong classifier, positioning
Failure cause.
8. system as claimed in claim 6, it is characterised in that the single fault identifing source unit, is further used for having selected
Each service feature value selected out, sets up a Weak Classifier based on BP neural network;Using being previously obtained comprising positive and negative
The sample data combination of business datum, trains the Weak Classifier identification single fault source, until reaching predetermined frequency of training, obtains
Weak Classifier after to training, wherein, often perform and training objective error is estimated after once training, minimum point of selection wherein error
Class device and the weights for reducing the grader.
9. system as claimed in claim 6, it is characterised in that the multiple faults identifing source unit, is further used for using bag
The sample data of multiple faults source data and phenomenon of the failure is included, the strong classifier after series connection is trained, until reaching predetermined
Frequency of training, obtains the layer grader, wherein, often perform and training objective error is estimated after once training, selection wherein error
The strong classifier of minimum simultaneously reduces the weights of the strong classifier.
10. system as claimed in claim 6, it is characterised in that the multiple faults identifing source unit, be further used for selection with
Coefficient correlation between phenomenon of the failure is more than the service feature value of predetermined threshold value, and selected service feature value is corresponding strong point
Class device carries out the series connection.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310080584.3A CN104052612B (en) | 2013-03-13 | 2013-03-13 | A kind of Fault Identification of telecommunication service and the method and system of positioning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310080584.3A CN104052612B (en) | 2013-03-13 | 2013-03-13 | A kind of Fault Identification of telecommunication service and the method and system of positioning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104052612A CN104052612A (en) | 2014-09-17 |
CN104052612B true CN104052612B (en) | 2017-08-25 |
Family
ID=51504997
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310080584.3A Active CN104052612B (en) | 2013-03-13 | 2013-03-13 | A kind of Fault Identification of telecommunication service and the method and system of positioning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104052612B (en) |
Families Citing this family (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106685674B (en) | 2015-11-05 | 2020-01-10 | 华为技术有限公司 | Method and device for predicting network event and establishing network event prediction model |
CN105554786B (en) * | 2015-12-11 | 2019-08-09 | 中国联合网络通信集团有限公司 | A kind of method and device of addressing network problems |
CN106909811B (en) * | 2015-12-23 | 2020-07-03 | 腾讯科技(深圳)有限公司 | Method and device for processing user identification |
CN106713016A (en) * | 2016-12-07 | 2017-05-24 | 中国联合网络通信集团有限公司 | Fault reasoning method and apparatus of indoor distribution system |
CN108696379A (en) * | 2017-04-07 | 2018-10-23 | 南京航空航天大学 | Cellular network fault diagnosis system based on integrated study and SMOTE |
CN107169434A (en) * | 2017-05-10 | 2017-09-15 | 广东工业大学 | One kind possesses the electric personal identification method of exclusive brain |
CN107067519B (en) * | 2017-05-12 | 2023-02-03 | 广东工业大学 | Safe box based on brain electricity identification |
CN107370618B (en) * | 2017-06-05 | 2020-06-05 | 北京奇艺世纪科技有限公司 | Troubleshooting method and device and electronic equipment |
CN113490857B (en) * | 2019-02-06 | 2024-04-02 | 利萨·德雷克塞迈尔有限责任公司 | Method and test device |
CN110716127A (en) * | 2019-10-29 | 2020-01-21 | 恩平市高尔电子科技有限公司 | Microphone receiver circuit board detecting system |
CN112129869B (en) * | 2020-09-23 | 2022-11-18 | 清华大学深圳国际研究生院 | Data-driven field mass spectrometer stable output control system and method |
CN113887414A (en) * | 2021-09-30 | 2022-01-04 | 北京百度网讯科技有限公司 | Target detection method, target detection device, electronic equipment and storage medium |
CN116805926B (en) * | 2023-08-21 | 2023-11-17 | 上海飞旗网络技术股份有限公司 | Network service type identification model training method and network service type identification method |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102957579A (en) * | 2012-09-29 | 2013-03-06 | 北京邮电大学 | Network anomaly traffic monitoring method and device |
-
2013
- 2013-03-13 CN CN201310080584.3A patent/CN104052612B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102957579A (en) * | 2012-09-29 | 2013-03-06 | 北京邮电大学 | Network anomaly traffic monitoring method and device |
Non-Patent Citations (1)
Title |
---|
Research of Identification and Positioning System for Network Failure based on BP neural network;Han-YongZhong等;《IEEE》;20121231;全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN104052612A (en) | 2014-09-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104052612B (en) | A kind of Fault Identification of telecommunication service and the method and system of positioning | |
CN105046279B (en) | A kind of analog circuit fault method for classifying modes | |
CN110348713A (en) | A kind of platform area line loss calculation method based on association analysis and data mining | |
CN108173708A (en) | Anomalous traffic detection method, device and storage medium based on incremental learning | |
CN109299741B (en) | Network attack type identification method based on multi-layer detection | |
CN106443259A (en) | Transformer fault diagnosis new method based on Euclidean clustering and SPO-SVM | |
CN106021771A (en) | Method and device for diagnosing faults | |
CN108958217A (en) | A kind of CAN bus message method for detecting abnormality based on deep learning | |
CN108520357A (en) | A kind of method of discrimination, device and the server of line loss abnormal cause | |
CN105372087A (en) | Fault diagnosis method based on multi-sensor signal analysis | |
CN106845526B (en) | A kind of relevant parameter Fault Classification based on the analysis of big data Fusion of Clustering | |
CN106548230A (en) | Diagnosis Method of Transformer Faults based on Modified particle swarm optimization neutral net | |
CN108932527A (en) | Using cross-training model inspection to the method for resisting sample | |
CN104809722A (en) | Electrical device fault diagnosis method based on infrared thermography | |
CN107656152B (en) | One kind being based on GA-SVM-BP Diagnosis Method of Transformer Faults | |
CN109033612A (en) | A kind of Diagnosis Method of Transformer Faults based on vibration noise and BP neural network | |
CN109102157A (en) | A kind of bank's work order worksheet processing method and system based on deep learning | |
CN108537259A (en) | Train control on board equipment failure modes and recognition methods based on Rough Sets Neural Networks model | |
CN109858503A (en) | The traction converter failure diagnostic method of decision tree is promoted based on gradient | |
CN107817404A (en) | A kind of Portable metering automatization terminal trouble-shooter and its diagnostic method | |
CN106874963B (en) | A kind of Fault Diagnosis Method for Distribution Networks and system based on big data technology | |
CN106198551A (en) | Method and device for detecting defects of power transmission line | |
CN110309884A (en) | Electricity consumption data anomalous identification system based on ubiquitous electric power Internet of Things net system | |
CN106327357A (en) | Load identification method based on improved probabilistic neural network | |
CN112200263B (en) | Self-organizing federal clustering method applied to power distribution internet of things |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |