The localization method of automatic calibration of electric energy meter assembly line epitope failure
Technical field
The present invention relates to electrical energy meter fault field of locating technology, are a kind of automatic calibration of electric energy meter assembly line epitope failures
Localization method.
Background technique
With the lasting popularization of concentration calibrating mode, the annual verification task amount of measurement centre is constantly promoted, automation inspection
The construction scale for determining facility constantly expands, and extensive, prolonged continuous calibrating proposes the quality control of automatic calibration facility
Bigger test is gone out.Epitope failure is the failure of the most frequent generation of automatic calibration of electric energy meter assembly line, epitope failure at present
It is broadly divided into the epitope intermittent defect as caused by the reasons such as contact pin abrasion, epitope displacement and curved, epitope bottom plate is hit by contact pin
Epitope caused by the reasons such as breakdown permanently damages failure.Taking place frequently for epitope failure not only influences calibrating efficiency, but also electric energy
Table verification result lacks reliability.
In order to improve automatic calibration assembly line O&M efficiency, shorten the automatic calibration pipeline stall time, monitoring is set
Standby operating condition improves calibrating quality, needs to improve calibrating abnormal quality monitoring mechanism, thus explores the wiring of automatic calibration assembly line
Fault location technology is very necessary.
Summary of the invention
The present invention provides a kind of localization methods of automatic calibration of electric energy meter assembly line epitope failure, overcome above-mentioned existing
There is the deficiency of technology, can effectively solve existing automatic calibration method and search epitope failure and waste time length, seriously affects
The problem of calibrating efficiency.
Technical solution of the present invention first is that being realized by following measures: a kind of automatic calibration of electric energy meter assembly line
The localization method of epitope failure, comprising the following steps:
S1, data preparation: obtain respectively it is a certain calibrating assembly line on electric energy meter calibration data, connection box of electric energy meter data and
Calibrating installation failure handling data;
S2 constructs epitope exception temporal aspect collection using time series window algorithm: according to time series window algorithm point
Not Gou Jian electric energy meter calibration qualification rate and connection box of electric energy meter success rate in watch window feature set;
S3 constructs automatic calibration of electric energy meter assembly line epitope abnormity diagnosis model: using the support based on sliding window algorithm
Vector machine model, abbreviation SVM model, structural classification device is to distinguish whether epitope has exception;
S4 constructs automatic calibration of electric energy meter assembly line epitope failure cause diagnostic model: judging fault type;
Electric energy meter calibration quality tracing: S5 obtains the calibrating uncertain electric energy meter inventory of quality.
Here is the further optimization and/or improvements to invention technology described above scheme:
In above-mentioned S2, epitope exception temporal aspect is constructed using time series window algorithm: being calculated according to time series window
Method constructs electric energy meter calibration qualification rate and connection box of electric energy meter success rate feature set in watch window, including following procedure respectively:
S21, according to electric energy meter epitope association electric energy meter calibration data, connection box of electric energy meter data and calibrating installation troubleshooting
Data;
S22, after electric energy meter calibration data, connection box of electric energy meter data are integrated using time series sliding window algorithm according to
Examine and determine the arrangement of time ascending order;
S23, according to fixed data window size M to the electric energy meter calibration data and the progress of connection box of electric energy meter data after sequence
Data subscript is moved backward a data unit by fragment processing, each window fragment, obtains N number of window, after forming fragment
Data set D;
S24 is calculated and is examined and determine in each window in data set D using electric energy meter calibration data and connection box of electric energy meter data
The assay approval rate x of record1With wiring success rate x2;
S25 constructs the calibrating epitope state y of each window phase according to calibrating installation failure handling data, forms data set
Di, formula are as follows:
Training set D is respectively obtained according to formula (1)1, test set D2, real time data collection D3, wherein x1、x2, y represent feature, n
Represent data set line number.
In above-mentioned S3, using the supporting vector machine model based on sliding window algorithm, structural classification device is to distinguish whether epitope has
It is abnormal, including following procedure:
S31 constructs SVM model to the calibrating epitope state y of each window phase, utilizes training set D1, including calibrating
Qualification rate x1, wiring success rate x2, calibrating epitope state y, training the model;Support vector machines mould of the training based on sliding window algorithm
The process of type is as follows:
Enabling Optimal Separating Hyperplane is wx+b=0, supporting vector (xs,ys) distance away from Optimal Separating Hyperplane are as follows:Enabling the function of supporting vector and Optimal Separating Hyperplane distance is 1, then it is super flat according to classifying to maximize supporting vector
The distance in face can be exchanged into:
Lagrangian is constructed to (2) formula:
(3) formula is solved according to the method for solving of KKT condition and dual problem, (2) formula is converted are as follows:
It is solved in above-mentioned transformed formula (4):
The super flat wx+b=0 that classifies is calculated according to formula (5), utilizes test set D2Verify the validity of model;
S32, using trained SVM model, to real time data collection D3Corresponding calibrating epitope state y classifies, i.e.,
By D3Assay approval rate x1, wiring success rate x2Decision function f (x)=sign (wx+b) after bringing training into obtains calibrating
Epitope state, and recording exceptional information.
S33, breaks down if calibrating epitope state is divided into, and issues epitope abnormal alarm to the calibrating epitope.
In above-mentioned S4, automatic calibration of electric energy meter assembly line epitope failure cause diagnostic model is constructed, judges fault type
Include following procedure:
S41 constructs Bayesian model according to the abnormal failure warning message of calibrating epitope:
Enabling continuous abnormal alarm times is J, and continuous J abnormal alarm time T, continuous J abnormal alarm is event B, is enabled
AiFor wiring faults event, wherein setting A1For epitope intermittent defect event, A2Event is permanently damaged for epitope, Bayes is seemingly
Right estimation formulas is as follows:
P(Ai|B)∝P(B|Ai)·P(Ai) (6)
S42 calculates separately A according to formula (6)1Event and A2The probability of event;
S44, according to A1Event and A2The probability calculation of event is as a result, judge fault type, i.e., as P (A1| B) > P (A2|B)
When, then report A1The wiring faults event of class, as P (A1|B)≤P(A2| B) when, then report A2The wiring faults event of class.
In above-mentioned S5, electric energy meter calibration quality tracing obtains the calibrating uncertain electric energy meter inventory of quality, including following mistake
Journey:
S51, for exact AiEvent marks the N determined by SVM modeliExist under (i=1,2 ... N) a time window
Examine and determine the electric energy meter inventory L of risk1;
S52, for exact AiEvent marks the N determined by SVM modeli(i=1,2 ... N) a time window is backward extremely
The time span t of first time abnormal alarm, Bayesian model judge the time T of J abnormal alarm, are recorded in t+T time span
The interior electric energy meter inventory L that there is calibrating risk2;
S53 obtains the calibrating uncertain electric energy meter inventory of quality are as follows: L1+L2。
Above-mentioned further includes S6, Modifying model: is modeled again to the fault sample of newly-increased tape label, according to the SVM constructed
The case where model and Bayesian model are not inconsistent diagnostic result and actual field, corrects SVM model parameter, then increase SVM model
Sample size, to improve model output accuracy.
By examining and determine conclusion using the analysis automated calibrating assembly line of big data technology, quick positioning failure is set the present invention
It is standby, accurate feedback epitope failure cause, the abnormal mesh to reach raising calibrating efficiency, promote calibrating quality of timely prosthetic appliance
's.
Detailed description of the invention
Attached drawing 1 is one method flow diagram of the embodiment of the present invention.
Attached drawing 2 is the method flow diagram of the building epitope exception temporal aspect collection of another embodiment of the invention.
Attached drawing 3 is the building automatic calibration of electric energy meter assembly line epitope abnormity diagnosis mould of another embodiment of the invention
The method flow diagram of type.
Attached drawing 4 is that the building automatic calibration of electric energy meter assembly line epitope failure cause of another embodiment of the invention is examined
Disconnected model, method flow diagram.
Attached drawing 5 is the method flow diagram of the electric energy meter calibration quality tracing of another embodiment of the invention.
Attached drawing 6 is the method flow diagram comprising Modifying model of another embodiment of the invention.
Specific embodiment
The present invention is not limited by the following examples, can determine according to the technique and scheme of the present invention with actual conditions specific
Embodiment.
Below with reference to examples and drawings, the invention will be further described:
Embodiment one: as shown in Fig. 1, a kind of localization method of automatic calibration of electric energy meter assembly line epitope failure, packet
Include following steps:
S1, data preparation: obtain respectively it is a certain calibrating assembly line on electric energy meter calibration data, connection box of electric energy meter data and
Calibrating installation failure handling data;
S2 constructs epitope exception temporal aspect collection using time series window algorithm: according to time series window algorithm point
Not Gou Jian electric energy meter calibration qualification rate and connection box of electric energy meter success rate in watch window feature set;
S3 constructs automatic calibration of electric energy meter assembly line epitope abnormity diagnosis model: using the support based on sliding window algorithm
Vector machine model, abbreviation SVM model, structural classification device is to distinguish whether epitope has exception;
S4 constructs automatic calibration of electric energy meter assembly line epitope failure cause diagnostic model: judging fault type;
Electric energy meter calibration quality tracing: S5 obtains the calibrating uncertain electric energy meter inventory of quality.
Above-mentioned electric energy meter calibration data and connection box of electric energy meter data: main includes calibrating epitope number, calibrating time, electricity
It can table calibrating conclusion (qualified, unqualified) and connection box of electric energy meter state (successful, unsuccessful);
Above-mentioned calibrating installation failure handling data: main includes examining and determine epitope number, wiring faults time of origin and connecing
Line fault type (such as epitope intermittent defect or epitope are permanently damaged).
The localization method of above-mentioned automatic calibration of electric energy meter assembly line epitope failure can be made further according to actual needs
Optimization or/and improvement:
As another embodiment of the invention: as shown in Fig. 2, in S2, being constructed using time series window algorithm
Epitope exception temporal aspect collection: according to time series window algorithm construct respectively electric energy meter calibration qualification rate in watch window and
Connection box of electric energy meter success rate feature set, including following procedure:
S21, according to electric energy meter epitope association electric energy meter calibration data, connection box of electric energy meter data and calibrating installation troubleshooting
Data;
S22, after electric energy meter calibration data, connection box of electric energy meter data are integrated using time series sliding window algorithm according to
Examine and determine the arrangement of time ascending order;
S23, according to fixed data window size M to the electric energy meter calibration data and the progress of connection box of electric energy meter data after sequence
Data subscript is moved backward a data unit by fragment processing, each window fragment, obtains N number of window, after forming fragment
Data set D;
S24 is calculated and is examined and determine in each window in data set D using electric energy meter calibration data and connection box of electric energy meter data
The assay approval rate x of record1(0: it is unqualified, 1: qualified) and wiring success rate x2(0: success, 1: unsuccessful);
S25, according to calibrating installation failure handling data, construct each window phase calibrating epitope state y (- 1: it is normal, 1:
Failure), form data set DiFormula are as follows:
Training set D is respectively obtained according to formula (1)1, test set D2, real time data collection D3, wherein x1、x2, y represent feature, n
Represent data set line number.
Electric energy meter calibration data in S23 refer mainly to electric energy meter calibration conclusion (0: unqualified, 1: qualified) and electric energy meter
Wired data refers mainly to connection box of electric energy meter state (0: success, 1: unsuccessful).
As another embodiment of the invention: as shown in Fig. 3, in S3, constructing automatic calibration of electric energy meter flowing water
Line epitope abnormity diagnosis model: supporting vector machine model (hereinafter referred to as SVM model) structural classification based on sliding window algorithm is used
Device is to distinguish whether epitope has exception, including following procedure:
S31 constructs SVM model to the calibrating epitope state y of each window phase, utilizes training set D1, including calibrating
Qualification rate x1, wiring success rate x2, calibrating epitope state y, training the model;Support vector machines mould of the training based on sliding window algorithm
The process of type is as follows:
Enabling Optimal Separating Hyperplane is wx+b=0, supporting vector (xs,ys) distance away from Optimal Separating Hyperplane are as follows:Enabling the function of supporting vector and Optimal Separating Hyperplane distance is 1, then it is super flat according to classifying to maximize supporting vector
The distance in face can be exchanged into:
Lagrangian is constructed to (2) formula:
(3) formula is solved according to the method for solving of KKT condition and dual problem, (2) formula is converted are as follows:
It is solved in above-mentioned transformed formula (4):
The super flat wx+b=0 that classifies is calculated according to formula (5), utilizes test set D2Verify model validation;
S32, using trained SVM model to real time data collection D3Corresponding calibrating epitope state y classifies, i.e., will
D3Assay approval rate x1, wiring success rate x2Decision function f (x)=sign (wx+b) after bringing training into obtains calibrating table
Position state, and recording exceptional information.
S33, breaks down if calibrating epitope state is divided into, and issues epitope abnormal alarm to the calibrating epitope.
As another embodiment of the invention: as shown in Fig. 4, in S4, constructing automatic calibration of electric energy meter flowing water
Line epitope failure cause diagnostic model, including following procedure:
S41 constructs Bayesian model according to the abnormal failure warning message of calibrating epitope:
Enabling continuous abnormal alarm times is J, and continuous J abnormal alarm time T, continuous J abnormal alarm is event B, is enabled
AiFor wiring faults event, wherein setting A1For epitope intermittent defect event, A2Event is permanently damaged for epitope, Bayes is seemingly
Right estimation formulas is as follows:
P(Ai|B)∝P(B|Ai)·P(Ai) (6)
S42 calculates separately A according to formula (6)1Event and A2The probability of event;
S43, according to A1Event and A2The probability calculation of event is as a result, judge fault type, i.e., as P (A1|B)>P(A2|B)
When, then report A1The wiring faults event of class, as P (A1|B)≤P(A2| B) when, report A2The wiring faults event of class.
As another embodiment of the invention: as shown in Fig. 5, in S5, electric energy meter calibration quality tracing:
S51, for exact AiEvent marks the N determined by SVM modeliExist under (i=1,2 ... N) a time window
Examine and determine the electric energy meter inventory L of risk1;
S52, for exact AiEvent marks the N determined by SVM modeli(i=1,2 ... N) a time window is backward extremely
The time span t of first time abnormal alarm, Bayesian model judge the time T of J abnormal alarm, are recorded in t+T time span
The interior electric energy meter inventory L that there is calibrating risk2;
S53 obtains the calibrating uncertain electric energy meter inventory of quality are as follows: L1+L2。
As another embodiment of the invention: on the basis of the above embodiments further including S6, mould as shown in Fig. 6
Type amendment: modeling the fault sample of newly-increased tape label again, is tied according to the SVM model and Bayesian model constructed to diagnosis
The case where fruit and actual field are not inconsistent corrects SVM model parameter, then increases SVM model sample amount, accurate to improve model output
Property.
The above technical features constitute embodiments of the present invention, can basis with stronger adaptability and implementation result
Actual needs increases and decreases non-essential technical characteristic, to meet the needs of different situations.