CN101345796B - Soft exchange user line intelligent test system and method based on user line test module - Google Patents

Abstract

The present invention discloses a soft switch door line intelligence test system based on door line test module and method thereof, pertaining to the soft-switching technology in the communication field. The system includes a core controller (100), a soft-switching support network (200), an access apparatus group (300), a door Line (400) and a phone (500) connected in sequence; the system is also provided with a door line test module (600) which is connected to other network components through the soft-switching support network (200); the door line test module (600) is a computer procedure installed on common PC, including a data sampler (610) and a categorizer (620) connected fore and after, which are connected to a fault reception center (630) through the soft-switching support network (200). The invention has extensive adaptability to the soft switching system, and can be conveniently transplanted to failure diagnosis applications of other fields due to flexible configuration.

Description

Soft switch family line intelligent test system and method thereof based on family line test module

Technical field

The present invention relates to soft switch (VoIP, the Voice of Internet Phone) technology in a kind of communications field, relate in particular to a kind of soft switch family line intelligent test system and method thereof based on family line test module.

Background technology

In the Softswitch technology field, the architecture of telephone service family line test (also claiming subscriber line test) is made up of the core controller, soft switch supporting network, access device, family line and the phone that connect successively.

General access device is distributed in the position near the user, thereby comparatively disperses.Access device generally comprises Integrated Access Device and IAD (AG).The subscriber line out of order data can obtain through the detecting unit that related access equipment provides, and supply Breakdown Maintenance network analysis location.

Flexible exchanging network have carry with control be separated, professional and characteristics that signaling is separated; With respect to traditional exchange network (carrying combines with control, business combines incorporate characteristics with signaling), flexible exchanging network has multiple user mode access and type, therefore family line fault test has been proposed new requirement.

The common measuring head testing scheme that adopts of the family line test of Plain Old Telephone Service must make up the special maintenance system, comprises composition such as " line center, family+professional branch office+remote test equipment " part, wherein:

Line center, family comprises: family line IVR (the automatic speech navigation is accepted, confirmed automatically); Database (subscriber data storehouse, the single information table of worker, branch office's routing table, test routing table, maintenance record table and system maintenance table etc.); WEB server (automatic flow management, artificial WEB application and statistical analysis etc.); Testing server (automatically, manual testing's dispatching management), family line central test operator station (manual work is accepted, manual work confirms, singly inquiry of worker, statistical analysis and manual testing etc.).

Professional branch office comprises: manual work is accepted, artificial confirmation, workform management, comprehensive inquiry, system maintenance, statistical analysis and manual testing etc.

Remote test equipment: remote test box.

This shows that the family line test configurations cost of Plain Old Telephone Service is very high; But because of the stored-program control exchange centralized configuration at the local side telecommunications room, and soft switchcall server decentralized configuration normally will adopt different family line testing schemes to this situation.

Soft switchcall server is different with the centralized programme-controlled exchange architecture of black phone; Telephone wire is distributed between user residence and Integrated Access Device or the IAD (AG); General access device widely dispersed, therefore the unsuitable traditional test head scheme that makes up can come the analysis of data collected modeling through the subsidiary detecting unit of access device; Form family line fault test analytical system, below be referred to as economical emulation testing scheme.

Emulation testing has convenience of deployment and advantage cheaply, but needs analyzing test data to come fault location, and accuracy depends on parser and reasonable analysis model.

Summary of the invention

For the family line emulation testing based on access device, to the subscribers feeder that connects access device port and phone, standard phone line comprises A line and B line.Through the detecting unit of access device AG, can obtain the A line to B line, A line over the ground with B line resistance, capacitance over the ground, and loop current value must could realize fault location by the following state of the data separation that test obtains.

(1) A line shorted to earth

(2) B line shorted to earth

(3) two-wire shorted to earth

(4) user's off-hook

(5) phone fault

(6) normal (on-hook)

(7) broken string

(8) short circuit

(9) circuit string direct current

To locate family line dependent failure according to the data that the access device detecting unit obtains; Usually need under the result can know the inside story condition to obtain experimental data (below be referred to as training data) through substantive test; Characteristics according to numeric distribution form empirical equation again, confirm certain specific user's line states thus, claim that the method is experience classification method (to call option b in the following text); Because aspect factors such as phone model, subscribers feeder length all influence the fault location conclusion; Therefore just must preserve these related informations, and the reference data of before every subscriber line test, preserving various definite states in advance, the reference data collection of large-scale consumer line is a challenge under the actual conditions; Almost can't implement; Though laboratory controlled condition capable of using is gathered, generally uncontrollable because of the scene, and the sample that the laboratory collects can not cover the correlative factor of locality.Thereby being necessary to introduce data mining technology (to call option A in the following text), the present invention is a kind of realization of option A.

Facts have proved; By computer software function, adopt the suitable data digging technology, also can satisfy the accuracy requirement of fault location; Can significantly reduce family line test system configurations and implementation cost thus, be applicable to that the line test of soft switch family is in interior resultant fault maintenance system demand.

The object of the invention just is to overcome the above-mentioned shortcoming and defect that prior art exists, and a kind of soft switch family line intelligent test system and method thereof based on family line test module is provided.

The objective of the invention is to realize like this:

One, based on the soft switch family line intelligent test system of family line test module

Like Fig. 1, native system comprises core controller 100, soft switch supporting network 200, access device crowd 300, family line 400 and the phone 500 that connects successively; Be provided with family line test module 600, family line test module 600 is connected with other network components through soft switch supporting network 200;

Like Fig. 2, family line test module 600 is a kind of computer programs, and installation and operation is on PC commonly used, and latter linked data acquisition unit 610 and grader 620 are connected to fault through soft switch supporting network 200 and accept center 630 before comprising;

Described data acquisition unit 610 is a kind of subprograms of obtaining data such as family line 400 resistance, electric capacity, voltage and current through access device crowd 300, provides grader 620 desired datas;

Described grader 620 is seed routine, the data that data collector 610 provides is classified, to confirm fault category;

It is a kind of systems that are connected on the soft switch supporting network 200 that described fault is accepted center 630, is responsible for accepting soft switch family wire system fault.

The operation principle of native system:

Fault for family, location line 400; Each IAD (AG) all has family line detecting unit among the access device crowd 300; Through in soft switch supporting network (200), disposing family line test module 600, comprise data acquisition unit 610 and grader 620 two parts, realize the fault detect and the location of family line (400); Grader 620 is set up training dataset 621 and is made up disaggregated model 623, and accomplish fault category on this basis and judge by intelligent algorithm 622 work, in time notifies family line fault to accept center 630 failure condition.

Wherein intelligent algorithm 622 is the subprogram according to the establishment of Naive Bayes Classification method, supplies grader 620 to call.

Two, based on the soft switch family line intelligent test method of family line test module

This method comprises the following steps:

(1) makes up disaggregated model

Like Fig. 3,1. under laboratory controlled or on-the-spot controlled condition, gather family line characteristic, comprise the numerical value of resistance, electric capacity, voltage and current;

2. the data conversion of gathering is become sample data, make up training dataset;

3. the sample data that every kind is gathered requires between 50～200 groups (too little can not fit curve well), and grader judges whether every type of training data reaches sample size;

4. if all kinds of training data quantity reach requirement, then make up disaggregated model, after this grader is used for fault location through intelligent algorithm;

5. if some training data quantity does not reach specified quantity, then change step (1)-1. over to, continue to gather the family line characteristic of respective classes through controlled way;

(2) handle the testing data tracing trouble through grader 620

1. the customary family line characteristic of gathering is sent to grader 620;

2. made up the grader processing collected data of disaggregated model, drawn the fault location conclusion, be sent to fault and accept center 630.

The present invention has advantage and good effect:

1, through introducing family line intelligent test method based on Naive Bayes Classification, make family line emulation testing scheme practicability, can satisfy the demand of fault location, greatly reduce cost than traditional test head scheme, be suitable for soft switch family line test macro demand.

2, do family line test (option A) with the method for Naive Bayes Classification, the family wire testing method (option b) that relies on reference data than conventional experience classification has very remarkable advantages:

1. data scale is little: the sample data that the A scheme needs about about 500～1000 groups, does not receive family line restricted number greatly; Sample (benchmark) data that the B scheme needs are almost suitable with family line quantity, suppose that soft switchcall server has 300,000 subscribers feeders, and the data scale of these two kinds of schemes differs more than 300 times;

2. accuracy is high: the data of A scheme collection are the data under the various situation; Data distribution curve through fitting under the various situation is done analysis; Find between the data and the association between data and the conclusion through the method self study of Bayes; The scientific theory foundation is arranged, can clearly distinguish common normal and malfunction.Even the B scheme has been accomplished 300,000 groups of data acquisitions, only to sort out by rule of thumb, how data use not rationale, are merely able to judge whether fault with normal, effectively fault location;

3. flexible: the A scheme is based on data, so even the test cell edition upgrading causes acquired data values to change, the fault discovery of new classification is arranged perhaps, only needs to gather and adds a desired data and get final product, and does not need update routine and algorithm.And the B scheme just needs the adjustment empirical equation, and may cause gathering again large quantities of data;

4. speed is fast: the A scheme only need be done learning process one time to 500～1000 groups of training datas; Learning process also can be very fast; Current common computer did not need for 1 second yet; Can generate grader, just can directly carry out fault judgement to test data with this grader later, the time of cost can be ignored (not can above 0.01 second) fully.The B scheme need be inquired about the data of 300,000 above data scales continually, and what its speed will inevitably be very is slow.

In a word, the present invention utilizes bayes classification method to realize data mining, makes up cheap and practical family line intelligent fault test macro, and traditional relatively family line cost of testing system reduces greatly; Realization is carried out intelligent diagnostics to complicated family line fault, helps improving the accuracy and the flexibility of diagnosis; This intelligent test method is easy to implement; Flexible configuration soft switchcall server had adaptability widely, owing to can also be transplanted in other field diagnosis application easily.

Description of drawings

Fig. 1 is the block diagram of native system;

Fig. 2 is the block diagram of family line test module 600;

Fig. 3 is the main flow chart of this method.

Wherein:

100-core controller.

200-soft switch supporting network.

300-access device crowd.

400-family line.

500-phone.

600-family line test module;

610-data acquisition unit,

621-training dataset, 622-intelligent algorithm, 623-disaggregated model;

620-grader;

630-fault is accepted the center.

1-laboratory controlled or on-the-spot controlled condition;

2-collection family line characteristic;

3-structure training dataset;

4-judge whether to reach sample size;

5-structure disaggregated model.

Embodiment

Different with traditional test head scheme; Emulation family line test data analysis belongs to typical classification problem; Promptly judge the fault that the relevant family of this equipment line possibly occur through collecting test data on equipment; Need training data to be used for training classifier as priori, thus the prediction fault category that real data reflected in the future.Because the data that the family line collects receive influence of various factors; Has uncertainty; Understand the relation between the complicated factor and be unusual difficulty result's influence; Yet the uncertain data Normal Distribution that this type influenced by complicated factor, so bayes classification method is suitable for handling this type problem.

Bayes is based on Bayes' theorem; Comparative studies through to bayesian algorithm is found; A kind of simple Bayes's sorting algorithm that is called Naive Bayes Classification can compare favourably with decision tree and neural network classification algorithm; Match with Computer Database, have high-accuracy and high-speed characteristics.

Attribute of Naive Bayes Classification hypothesis is independent of the value of other attributes to given type influence, and this hypothesis type of being called condition is independent, simplifies required calculating thus, and under this meaning, be called " simple ".

The line artificial intelligence test of soft switch family is introduced the naive Bayesian method and is realized data mining.

Family line test module 600 is the programs that operate on the ordinary PC; Constitute by data acquisition unit 610 and grader 620 subprograms; Data acquisition unit 610 is network-oriented work, and grader 620 comprises database engine, works based on frequently-used data storehouse (like MySQL).

Data acquisition unit 610 comprises the tcp/ip communication preset mechanism; Can send family line test command to access device crowd 300 AG through assigned ip address and COM1; Command parameter comprises family line interface number, receive the return data of corresponding family line interface from AG when data acquisition unit after, grader 620 adopts intelligent algorithms to handle; With these data construct training datasets 621; Up to composition and classification model 623, just can be used to judge the fault category of image data on this basis, the validity of conclusion depends on training dataset 621 and disaggregated model 623 certainly.

1, disaggregated model makes up

Like Fig. 2, grader 620 at first will use the data of 622 pairs of collections of intelligent algorithm to carry out sorting of operation, changes into training dataset 621.Training dataset 621 can be gathered through laboratory and scene on the spot; Not controlled way of controlled and routine is generally arranged; If can confirm specific category in advance when gathering; Or the people is image data when producing different classes of fault, then is called controlled way, comprising: laboratory controlled collection, on-the-spot controlled collection; On-the-spot customary the detection gathered normally not controlled way.

Simulated field various factors on the spot as much as possible under the laboratory controlled condition comprises AG port, family line length and phone model, and gathers training data different classes of under the various situation, comprises the data under the normal condition.

Form in the laboratory controlled collection on the basis of training dataset, also need be through on-the-spot controlled collection incremental data, the attendant cooperates down at the scene, confirms the concrete class of current image data; Random choose access device AG and make various types of faults equably as far as possible also in addition is saved in the data of this controlled collection in the training dataset by fault category.Attention remedies the situation that laboratory simulation does not go out through the data of collection in worksite, improves disaggregated model 623 thus.

Wherein, laboratory controlled makes up the plan such as the table 1 of disaggregated model:

Fault is divided into 9 types, and to far away 5 kilometers, data representation test group number in the form wherein adopts the telephone set of two kinds of models using always to the family line length from recently (10 meters), and 5|5 representes to exchange telephone set and respectively surveys 5 groups.

Table 1: family line data acquisition makes up the planning of experiments of disaggregated model

Family line typical length	10m	1km	2km	3km	4km	5km	Add up to
								A line shorted to earth	5	10	10	10	10	10	55

B line shorted to earth	5	10	10	10	10	10	55
								The two-wire shorted to earth	5	10	10	10	10	10	55
User's off-hook	5	5|5	5|5	5|5	5|5	5|5	55
								The phone fault	5	5|5	5|5	5|5	5|5	5|5	55
Normally (on-hook)	5	5|5	5|5	55	5|5	5|5	55
								Broken string	5	10	10	10	10	10	55
Short circuit	5	10	10	10	10	10	55
								Circuit string direct current	30	10	10	10	10	?	70

2, the maintenance work of newly-increased failure classes and device upgrade

If the disaggregated model precision that the controlled acquisition method of laboratory makes up is not high enough; Maybe need increase fault category newly; Or device upgrade causes the data value distribution situation to change; Generally need gather 200 to 300 groups of new training datas on the spot and revise, can adopt incremental modification method, so just can come dynamically to adjust nicety of grading through adjustment training data sample set.Even bigger variation takes place in the sample data that causes gathering owing to reasons such as device upgrades in the future, also only need gather training data again, the grader handling procedure also needn't be changed.

3, the test accuracy evaluation method of image data

According to the actual conditions of family line test, selected in the data mining Naive Bayes Classification method as self-learning algorithm, through the data that collect being done the accuracy that ten operation methods of repeatedly intersecting are come verification test.The data that collect can be used as training data, also can be used as test data to be measured.

Concrete implementation method comprises the following steps:

1. with random algorithm data are upset and be divided into 10 groups, get 1 group successively, and remaining 9 groups is training data as test data;

2. getting the 1st group is test data; And 2～10 groups be training data, and training data can make up disaggregated model after through the self study process, predicts the malfunction of test data (removal malfunction) again with this grader; Promptly accomplished a subseries, and the precision that obtains classifying;

3. getting the 2nd group is test data, and 1,3～10 group as training data, with 2. step operation;

4. and the like can obtain 10 group categories precision, average then as last test accuracy.

Conclusion (of pressure testing):

Under the laboratory controlled condition, set up disaggregated model with the Naive Bayes Classification method and can record classify accuracy and reached 99.41%.

Conventional empirical equation classification can't effectively be distinguished user's off-hook and short circuit both of these case, be difficult for confirming short trouble, and the Naive Bayes Classification method can make a distinction well; In addition, circuit string direct current also is that the empirical equation classification can't be discerned, and the Naive Bayes Classification device to this type Fault Identification degree near 100%.

The Naive Bayes Classification method can dynamically increase fault category newly, makes grader reach effective precision through said method.

The speed of test: 510 groups of data do ten repeatedly crossing operation time of reaching a conclusion less than 1 second.Machines configurations is CPU:2.93GHZ, MEM:1G.

[attaching] operation principle and theoretical foundation

1, access gateway AG test philosophy

AG inside can provide the emulation testing function, and emulation testing is to utilize AG close beta resource to realize a kind of means of testing of 112 line tests.The test command of analog simulation AG maintenance terminal is adopted in 112 emulation testings; And through the relevant terminal mouth test command is sent to AG and starts test; AG close beta unit starts test and with test result loopback relevant terminal after receiving the test command that sends at the terminal.

112 emulation testings are 112 the cheapest solutions of cost of investment from economic angle.Certainly, the measuring accuracy of emulation testing depends on the measuring accuracy of AG test cell itself.And the external emulation test request is that to handle rank lower in the message processing facility of AG, so test speed will receive the AG not busy degree affect of doing.

But see that through fixed network user's construction situation for many years Plain Ordinary Telephone Service (POTS, Plain OldTelephone Service) user is solved the test of all kinds of business, and generally all use test platform or measuring head mode solve.The mode of external testing equipment is precision, speed and stablizes the test that the aspect all is better than inner clamp greatly.

AG office has characteristics such as capacity is little, point wide, the common unattended operations that distribute more; Consideration based on high performance-price ratio; Can be built-in in AG " emulation remote test greatest service can " accomplish test; When fully guaranteeing test performance, take into full account the above-mentioned characteristic of AG office, reach the effect of optimality price ratio.

Testing process: after the subscriber phone fault was declared, the resource management system through the Soft core controlling platform can navigate to the access device port, and this port is corresponding with the fault telephone number.Can initiate the emulation testing order through the access device webmaster, the result data that obtains analyzed, conclusion is fed back to the Breakdown Maintenance system, and then accomplish fault and accept by expert system.

2, data mining principle

Bayes is based on Bayes' theorem; Comparative studies through to bayesian algorithm is found; A kind of simple Bayes's sorting algorithm that is called Naive Bayes Classification can compare favourably with decision tree and neural network classification algorithm; Match with Computer Database, show high-accuracy with high-speed.

Attribute of Naive Bayes Classification hypothesis is independent of the value of other attributes to given type influence.This hypothesis type of being called condition is independent.Do this supposition and be in order to simplify required calculating, and under this meaning, be called " simple ".

The line emulation testing of soft switch VOIP family is introduced the naive Bayesian method and is realized data mining.

3, Bayes' theorem

If the data sample that X type of being label is unknown.If H is certain supposition, belong to certain specific class C like data sample X.For classification problem, we hope to confirm P (H|X)---given observation data sample X, suppose the probability that H sets up.

P (H|X) is a posterior probability, or the posterior probability of H under the condition X, and P (H) is a prior probability, or the prior probability of H.Based on more information, P (H) is independent of X to posterior probability P (H|X) than prior probability P (H).

Similarly, P (X|H) is under the condition H, the posterior probability of X, and P (X) is the prior probability of X.

How to calculate these probability? As what below will see, P (X), P (H) and P (X|H) can be by given data computation.Bayes' theorem is useful, and it provides a kind of by P (X), and P (H) and P (X|H) calculate the method for posterior probability P (H|X).Bayes' theorem is:

$P\left(H\right|X)=\frac{P\left(X\right|H)P\left(H\right)}{P\left(X\right)}$

Next, will narrate how in Naive Bayes Classification, to use Bayes' theorem.

4, Naive Bayes Classification is theoretical

The course of work of Naive Bayes Classification is following:

(1) each data sample is with a n dimensional feature vector X=(X ₁, X ₂, X ₃..., X _n) expression, describe respectively n attribute A _iThe n of sample tolerance.

(2) supposition has m type of C ₁, C ₂, C ₃..., C _mThe data sample X of given the unknown (i.e. type of not having label), classification will predict that X belongs to the class with the highest posterior probability (under condition X).That is to say that Naive Bayes Classification is given a type C with the sample dispensing of the unknown _i, and if only if

P(C _i/X)>P(C _j/X)，1≤j≤m，j≠i.

Like this, maximization P (C _i/ X).Its P (C _i/ X) maximum class C _iBe called the maximum a posteriori hypothesis.According to Bayes' theorem,

$P\left(C_i\right|X)=\frac{P\left(X\right|C_i)P\left(C_i\right)}{P\left(X\right)}$

(3) because P (X) is a constant for all types, only need P (X/C _i) P (C _i) maximum getting final product.Prior probability like fruit is unknown, supposes usually that then these types are equiprobable, i.e. P (C ₁)=P (C ₂)=...=P (C _m).And in view of the above to P (C _i/ X) maximization.Otherwise, maximization P (X/C _i) P (C _i).Notice that the prior probability of class can be used $P\left(C_i\right)=\frac{S_i}{S}$ Calculate wherein S _iType of being C _iIn number of training, and S is a training sample sum.

(4) given data set with many attributes calculates P (X/C _i) expense maybe be very big.Calculate P (X/C for reducing _i) expense, can type of doing condition independently simple hypothesis.Given specimen number supposes that the property value condition of reciprocity is independent, promptly between attribute, does not have dependence.Like this,

$P\left(X\right|C_i)=\underset{k=1}{\overset{n}{\mathrm{\Π}}}P\left(x_k\right|C_i)$

Probability P (x _k| C _i) can estimate by training sample, wherein

If 1. A _kBe categorical attribute, then $P\left(x_k\right|C_i)=\frac{S_{\mathrm{Ik}}}{S_i}$ S wherein _IkBe at attribute A _kOn have value X _kThe class C _iNumber of training, and S _iBe C _iIn number of training.

If 2. A _kBe the successive value attribute, then suppose this attribute Normal Distribution usually.Thereby,

$P\left(X\right|C_i)=g(x_k,{\mathrm{\μ}}_{C_i},{\mathrm{\σ}}_{C_i})=\frac{1}{\sqrt{2\mathrm{\π}}{\mathrm{\σ}}_{C_i}}{e}^{-\left(\frac{x_k-{\mathrm{\μ}}_{C_i}}{{{2\mathrm{\σ}}_{C_i}}^2}\right)}$

Wherein, given type of C _iTraining sample attribute A _kValue,

Be attribute A _kDensity function, and

With

Be respectively mean value and standard deviation.

(5) be to unknown sample X classification, to each type C _i, calculate P (X/C _i) P (C _i).Sample X is assigned to class C _i, and if only if

P(X|C _i)P(C _i)>P(X|C _j)P(C _j)，1≤j≤m，j≠i

In other words, X is assigned to its P (X/C _i) P (C _i) maximum class C _i

Claims (1)

1. soft switch family line intelligent test method based on family line test module is characterized in that:

(1) makes up disaggregated model

1. under laboratory controlled or on-the-spot controlled condition, gather family line characteristic, comprise the numerical value of resistance, electric capacity, voltage and current;

2. the data conversion of gathering is become sample data, make up training dataset;

3. the sample data that every kind is gathered requires between 50～200 groups, and grader judges whether every type of training data reaches sample size;

4. if all kinds of training data quantity reach requirement, then make up disaggregated model, after this grader is used for fault location through intelligent algorithm;

5. if some training data quantity does not reach specified quantity, then change step (1)-1. over to, continue to gather the family line characteristic of respective classes through controlled way;

(2) handle the testing data tracing trouble through grader 620

1. the customary family line characteristic of gathering is sent to grader (620);

2. made up the grader processing collected data of disaggregated model, drawn the fault location conclusion, be sent to fault and accept center (630).

Images