CN110008253A - The industrial data association rule mining and unusual service condition prediction technique of strategy are generated based on two stages frequent item set - Google Patents
The industrial data association rule mining and unusual service condition prediction technique of strategy are generated based on two stages frequent item set Download PDFInfo
- Publication number
- CN110008253A CN110008253A CN201910244856.6A CN201910244856A CN110008253A CN 110008253 A CN110008253 A CN 110008253A CN 201910244856 A CN201910244856 A CN 201910244856A CN 110008253 A CN110008253 A CN 110008253A
- Authority
- CN
- China
- Prior art keywords
- line segment
- frequent
- data
- service condition
- association rule
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of industrial data association rule minings and unusual service condition prediction technique that strategy is generated based on two stages frequent item set, can be applied to the prognostic and health management of industrial process.The present invention is introduced into association rule mining in industrial equipment failure predication, finds the relevance between operating parameter by association rules mining algorithm.For industrial data feature, start with from the variation tendency of equipment operating parameter, by being most important quota student into transaction set with operating parameter variation tendency, and the association rule mining between parameter and parameter is carried out based on this, then association rule mining result is introduced into the prediction of industrial equipment unusual service condition, to obtain more accurate prediction result.For in engineering failure predication and health control have major application value.
Description
Technical field
The invention belongs to reliability maintenance field of engineering technology, it is related to a kind of generating strategy based on two stages frequent item set
Industrial data association rule mining and unusual service condition prediction technique.
Background technique
As the emergence of complication system and the demand of industrial process real-time monitoring are continuously increased, modern industrial equipment
Multiple sensors are often equipped in the process of running to be monitored its operating status.Meanwhile it may in equipment running process
There is various faults mode, a certain failure may correspond to several signs, and in the case, single-sensor information can not complete body
Existing equipment running status, the failure predication based on multi-sensor information are come into being.Failure predication based on multi-sensor information
It is intended to the operating status using comprehensive sensor information analytical equipment, to carry out more reliable device diagnostic and prediction.With
The sustainable development of sensing technology, using multiple sensors carry out equipment status monitoring, fault diagnosis and prediction have become
Development trend.
There are certain relevances between its operating parameter in equipment running process still rarely has at present in failure predication field
The work that association rule mining is combined with failure predication.And in fact, for time series data, equipment fault or failure
The correlated characteristic extracted often by parameter or from parameter is embodied, and prediction is often to parameter or correlated characteristic
Variation tendency predicted.Excavate the correlation rule between parameter, available more complete parameter, that is, equipment running status letter
Breath provides certain foundation for subsequent prediction.
Summary of the invention
For the status of the prior art, present invention aim to address rarely have consideration in the Predicting Technique of available data driving
Sensing data the case where there are correlation rules, propose a kind of unit exception operating condition prediction side based on operating parameter correlation rule
Method constructs more applicable wavelet neural network and carries out unusual service condition prediction (failure predication).
Now design of the invention is described below:
The present invention is portrayed using relevance of the correlation rule to industrial process operating parameter, have studied based on when ordinal number
According to the unusual service condition forecasting problem of association rule mining.In order to which association rule are excavated in sequence level for time series data
Then, the invention proposes a kind of time series data association rules mining algorithms that process is generated comprising two stages frequent item set.
In the first stage, basic model of the variation tendency information of extraction time sequence as association rule mining, discovery time sequence
The frequent item set of change shape;In second stage, based on the frequent item set of time series variation form, discovery sequence is base
The frequent item set of this mode, and association rule mining has been carried out to sequence two-by-two.Then, the resulting correlation rule phase of excavation is utilized
The system variable of pass carries out unusual service condition prediction, and correlation rule is introduced into wavelet neural network and improves forecasting accuracy.This hair
The method of bright proposition accounts for operating parameter correlation rule, can obtain more accurate failure predication result.
According to the above inventive concept, the invention proposes a kind of industrial datas that strategy is generated based on two stages frequent item set
Association rule mining and prediction technique, the specific steps are as follows:
Step 1: to time series data piece-wise linearization expression and symbolism, construction be suitable for association rule mining from
Dissipate type data set;
Step 2: the frequent item set of data set is generated using two stage Frequent Itemsets Mining Algorithm;
Step 3: correlation rule being generated according to frequent item set, extracts the pass for meeting minimum support and minimal confidence threshold
Connection rule;
Step 4: association rule mining result being introduced into wavelet neural network, and the unusual service condition for industrial equipment is pre-
It surveys.
Based on above scheme, each step can specifically use following implementation:
Preferably, the step 1 includes following sub-step:
Step 1.1: note sensor measurement time series isN is sensor
Quantity, k length of time series;Initially fitting starting point isInitially fitting terminal isNote is fitted starting pointBeing fitted terminal isError of fitting threshold value is ωE;
Step 1.2: for eachPiecewise fitting is carried out as follows:
1.2.1 waypoint count value count=1 is initialized;
1.2.2 successively to each fitting starting pointExecute step 1)-step 4):
1) end=start+h is calculated first;
2) for dataAnd be fitted using least square method, it counts
Calculate error of fitting ERR;
3) if error of fitting ERR is not more than error of fitting threshold value ωE, then 1) h=h+1, gos to step again;
4) if error of fitting ERR is greater than error of fitting threshold value ωE, obtainLine segment be fitted sequenceStart=start+h records waypointReset h=2, count=count+1;
1.2.3 circulation executes 1.2.2 and terminates greater than k until end, the line segment time series after being fittedAnd waypointThe segmentation point sequence P of compositioni;
Step 1.3: the time series after any sensor is fittedIt is denoted as Yk={ y1,y2,…,yk, it extracts every and intends
The trend and numerical information of zygonema section, and a matching line segment s is indicated by the way of following triplei:
Wherein, kiIndicate the slope of the line segment,Indicate the span of the line segment on a timeline, riIndicate the segment data
Growth rate, data { y corresponding for the line segmentj,yj+1,…,yj+h,J is the starting point of the line segment;
To line segment time series YkIn all line segments carry out triple expression, obtain triad sequence Sn={ s1,
s2,…,sn, wherein n indicates time series XkLine segment number after segmentation;
Step 1.4: cluster being carried out to a serial of line sections in triad sequence and symbolism is carried out to line segment, is set for indicating
Standby or different system version describes line segment s using Euclidean distanceiAnd sjSimilarity dij:
Wherein, dijIndicate line segment siAnd sjSimilarity, dijIt is smaller, then it represents that two lines section has more like variation shape
State, ωkAnd ωrFor weight;
Then according to index of similarity dij, using K-means clustering algorithm to SnIt is clustered, and is same class line segment point
The variation pattern that operating parameter is indicated with a same symbol, obtains the sequence F of symbolismn={ f1,f2,…,fn, f1,
f2,…,fnRespectively indicate the 1,2nd ..., the symbol that n line segment is assigned to;
Step 1.5: for the time of measuring sequence of every two sensorWithMerge it and is segmented point sequence PiAnd Pj,
It is denoted as Pij, nij- 1 is PiAnd PjWaypoint number after merging;And by the waypoint after merging to its symbolism sequenceWith
It is split reconstruct, the symbolism sequence after being reconstructedWith
Preferably, the step 2 includes following sub-step:
Step 2.1: for time of measuring sequenceWithCorresponding operating parameter ViAnd Vj, it is obtained by step 1
Measurement sequence symbolism data beWithTransaction set is constituted by it,
I.e. each affairs are denoted as WithIncluded in line segment class code be denoted as respectivelyWithRemember that two stage minimum support threshold value is respectively minsup1And minsup2;
Step 2.2: by single sweep operation data set, calculating the support of each single item, obtain frequent 1- item collection, by as follows
2.2.1~2.2.3 process carries out:
2.2.1: note σ () is the support counting of item or item collection, is initially 0;IfClass code be tk, t expression
A or b;
2.2.2: for each affairsCalculate σ (tk)=σ (tk)+1;
2.2.3: for each tkIfNot less than minimum support threshold value minsup1, then it is assumed that tkFor frequent 1-
Item collection retains tkAnd record corresponding support counting;IfLess than minimum support threshold value minsup1, then it is assumed that tk
It is not frequent 1- item collection;
Step 2.3: using the frequent 1- item collection t obtained in step 2.2k2- item collection is constituted, and calculates its support, to
It was found that frequent 2- item collection, carries out according to the following procedure:
2.3.1: note apAnd bqRespectively pass through step 2.2 from former line segment class codeWithThe item of middle reservation;
2.3.2 for each { ap,bq, execute following steps:
1) each is present inIn { ap,bq, calculate σ ({ ap,bq)=σ ({ ap,bq})+1
If 2)Not less than minsup1, then it is assumed that { ap,bqIt is frequent 2- item collection, retain { ap,bqAnd record
Corresponding support counting;
Step 2.4: using the frequent 2- item collection { a obtained in step 2.3p,bqCalculate every two operating parameter entirely counting
According to the support of concentration, and the frequent item set of parameter level is obtained, carried out according to the following procedure: to every two operating parameter ViAnd Vj
Item collection { the V of compositioni,Vj, calculate σ ({ Vi,Vj)=sum (σ ({ ap,bq)), ifNot less than minimum support threshold value
minsup2, then retain { Vi,VjAnd corresponding support is recorded, and calculate σ (Vi)=sum (σ (ap));σ(Vj)=sum (σ
(bq))。
Preferably, the step 3 includes following sub-step:
Step 3.1: to the every group of { V for meeting support threshold obtained in step 2i,Vj, generate following correlation rule: Vj
→ViAnd Vi→Vj, note minimal confidence threshold is minconf;
Step 3.2: according to every group of correlation rule of generation, calculating its confidence threshold value, extract the process of correlation rule such as
Under: for each correlation rule Vi→Vj, calculateIf conf (Vi→Vj) be not less than
Minimal confidence threshold is minconf, then retains correlation rule Vi→VjAnd record corresponding support and confidence level ωi。
Preferably, the step 4 includes following sub-step:
Step 4.1: for any group association parameter extracted from correlation rule, being denoted as { V1,V2,…,Vu, wherein u table
Show the quantity of relevant parameter, VuFor rule it is consequent i.e. target component, for each correlation rule Vi→Vu, i=1,2 ... u-
1, there is a confidence level, is denoted as ωi;For target component Vu, using wavelet neural network, carry out unusual service condition prediction;
Step 4.2: construction training sample: remembering that preset prediction step is l, the one group of association extracted by association rule mining
Parameter is { V1,V2,…,Vu, the complete training dataset being made of it is denoted as
Construct following matrix ItrainIt is inputted for the training of neural network:
Wherein, ItrainIn it is each be classified as a trained input sample, construct training output OtrainAre as follows:
Step 4.3: be trained using the training sample of construction to wavelet neural network: input parameter is Vi, i=1,
2 ... u-1, output parameter Vu, wherein utilizing the confidence level ω obtained by correlation rule in netiniti, i=1,
The initial weight between network input layer and hidden layer is arranged in 2 ... u-1;
Step 4.4: new data prediction: remembering that threshold value occurs for preset unusual service condition is ωp, for new collected sensor
Measurement data carries out l step prediction using model trained in step 4.3, if obtained target component predicted value is relative to first
The normal drift value that begins is more than set threshold value ωp, then it is assumed that unusual service condition occurs.
Preferably, before equipment does not fail, with the update of data, after every measurement data for updating predetermined quantity,
Simultaneously training pattern need to be reconfigured, to obtain more accurate prediction result.
It is proposed by the present invention a kind of the industrial data association rule mining of strategy and pre- to be generated based on two stages frequent item set
Survey method can be used for the Complex Industrial Systems of sensor measurement.By being associated rule digging to industrial equipment operating parameter,
Corresponding parameter association is obtained, and is introduced into wavelet neural network prediction, more accurate prediction effect can be obtained.
This will provide solid support to subsequent plant maintenance plan, for the equipment maintenance and management stringent to reliability requirement
It is of great advantage, there are bright prospects in terms of practical implementation.
Detailed description of the invention
Fig. 1 compares for 7 prediction result of IDV (13) variable in embodiment and with true value;
Fig. 2 compares for 11 prediction result of IDV (13) variable in embodiment and with true value;
Fig. 3 is that IDV (13) variable 7 predicts error rate in embodiment;
Fig. 4 is that IDV (13) variable 11 predicts error rate in embodiment.
Specific embodiment
A specific embodiment of the invention is further described now in conjunction with attached drawing.
This example is specifically described concrete operation step by Tennessee-Yi Siman (TE) process simulation data and tests below
The effect of card method.
The sampling interval of the data set is 3 minutes, each collected change of sensor under the data set record sampling interval
Measurement.Under each service condition (the failure operation state under normal operating condition and 21 kinds of preset failures), imitate
The measurement data of true process will all generate two class data sets, i.e. training set and test set.Wherein, for the acquisition of training set
Journey is the measured value of all 52 variables obtained in the case of simulation process runs 25 small, wherein except normal operation
Outside the training set that state acquisition arrives, the acquisition of remaining 21 training set data is as a child to introduce failure in simulation process operation 1,
And only record the measurement data of subsequent 24 hours.In other words, the training set of normal operating condition has 500 observation samples,
The training set acquired under remaining 21 malfunction is 480 observation samples.In addition, for 22 test sets, data are
Simulation process runs 48 collected all variable measurements of hour institute, that is to say, that includes 960 in each test set
Sample data.It should be noted that corresponding failure is at simulation run 8 hours when emulating to 21 kinds of procedure faults
It introduces afterwards.Therefore, for the test set under 21 failure operation states, preceding 160 observation samples are normal data, after
800 observation samples are fault data.In TE process simulation model, only IDV (13) is a soft fault, therefore,
In this example, we are tested using the related data of IDV (13).Industrial data association rule mining and unusual service condition prediction side
Detailed process is as follows for method:
Step 1: to time series data piece-wise linearization expression and symbolism, construction be suitable for association rule mining from
Dissipate type data set.This step specifically includes following sub-step:
Step 1.1: note sensor measurement time series isN is sensor
Quantity, k length of time series;Initially fitting starting point isInitially fitting terminal isNote is fitted starting pointBeing fitted terminal isError of fitting threshold value is ωE.It should be noted that in the present invention, i, j are tables as subscript
The number for showing sensor is only to indicate ordinal number as subscript, unrelated with sensor number.
Step 1.2: for eachPiecewise fitting is carried out as follows:
1.2.1 waypoint count value count=1 is initialized;
1.2.2 successively to each fitting starting pointExecute step 1)-step 4):
1) end=start+h is calculated first;
2) for dataAnd be fitted using least square method, it counts
Calculate error of fitting ERR;
3) if error of fitting ERR is not more than error of fitting threshold value ωE, then 1) h=h+1, gos to step again;
4) if error of fitting ERR is greater than error of fitting threshold value ωE, obtainLine segment be fitted sequenceRecord waypointReset h=2, count=count+1;
1.2.3 circulation executes 1.2.2 and terminates greater than k until end, the line segment time sequence after obtaining least square method fitting
ColumnAnd waypointThe segmentation point sequence P of compositioni;
Step 1.3: the time series after any sensor is fittedIt is denoted as Yk={ y1,y2,…,yk, wherein having more
The line segment of the aforementioned least square method fitting of item.The trend and numerical information of every matching line segment are extracted, and uses following ternary
The mode of group indicates a matching line segment si:
Wherein, kiIndicate the slope of the line segment,Indicate the span of the line segment on a timeline, riIndicate the segment data
Growth rate, data { y corresponding for the line segmentj,yj+1,…,yj+h,J is the starting point of the line segment;
To line segment time series YkIn all line segments carry out triple expression, obtain triad sequence Sn={ s1,
s2,…,sn, wherein n indicates time series XkLine segment number after segmentation;
Step 1.4: cluster being carried out to a serial of line sections in triad sequence and symbolism is carried out to line segment, is set for indicating
Standby or different system version, to prepare for subsequent association rule mining.Line segment s is described using Euclidean distancei
And sjSimilarity dij:
Wherein, dijIndicate line segment siAnd sjSimilarity, dijIt is smaller, then it represents that two lines section has more like variation shape
State, ωkAnd ωrFor weight;
Then according to index of similarity dij, using K-means clustering algorithm to SnIt is clustered, and is same class line segment point
The variation pattern that operating parameter is indicated with a same symbol, obtains the sequence F of symbolismn={ f1,f2,…,fn, f1,
f2,…,fnRespectively indicate the 1,2nd ..., the symbol that n line segment is assigned to;
Step 1.5: for the time of measuring sequence of every two sensorWithMerge it and is segmented point sequence PiAnd Pj,
It is denoted as Pij, nij- 1 is PiAnd PjWaypoint number after merging;And by the waypoint after merging respectively to its symbolism sequence
WithIt is split reconstruct, the symbolism sequence after being reconstructedWith
Step 2: the frequent item set of data set is generated using two stage Frequent Itemsets Mining Algorithm.This step specifically includes
Following sub-step:
Step 2.1: for time of measuring sequenceWithCorresponding operating parameter ViAnd Vj, it is obtained by step 1
Measurement sequence symbolism data beWithTransaction set is constituted by it,
I.e. each affairs are denoted as WithIncluded in line segment class code be denoted as respectivelyWithRemember that two stage minimum support threshold value is respectively minsup1And minsup2.?
In this example, minimum support threshold value is set are as follows: minsup1=0.2, minsup2=0.2.
Step 2.2: by single sweep operation data set, calculating the support of each single item, obtain frequent 1- item collection, by as follows
2.2.1~2.2.3 process carries out:
2.2.1: note σ () is the support counting of item or item collection, is initially 0;IfClass code be tk, t expression
A or b;
2.2.2: for each affairsCalculate σ (tk)=σ (tk)+1;
2.2.3: for each tkIfNot less than minimum support threshold value minsup1, then it is assumed that tkIt is frequent
1- item collection retains tkAnd record corresponding support counting;IfLess than minimum support threshold value minsup1, then it is assumed that
tkIt is not frequent 1- item collection;
Step 2.3: using the frequent 1- item collection t obtained in step 2.2k2- item collection is constituted, and calculates its support, to
It was found that frequent 2- item collection, carries out according to the following procedure:
2.3.1: note apAnd bqRespectively pass through step 2.2 from former line segment class codeWithThe item of middle reservation;
2.3.2 for each { ap,bq, execute following steps:
1) each is present inIn { ap,bq, calculate σ ({ ap,bq)=σ ({ ap,bq})+1
If 2)Not less than minsup1, then it is assumed that { ap,bqIt is frequent 2- item collection, retain { ap,bqAnd record
Corresponding support counting;
Step 2.4: using the frequent 2- item collection { a obtained in step 2.3p,bqCalculate every two operating parameter entirely counting
According to the support of concentration, and the frequent item set of parameter level is obtained, carried out according to the following procedure: to every two operating parameter ViAnd Vj
Item collection { the V of compositioni,Vj, calculate σ ({ Vi,Vj)=sum (σ ({ ap,bq)), ifNot less than minimum support threshold value
minsup2, then retain { Vi,VjAnd corresponding support is recorded, and calculate σ (Vi)=sum (σ (ap));σ(Vj)=sum (σ
(bq))。
Step 3: correlation rule being generated according to frequent item set, extracts the pass for meeting minimum support and minimal confidence threshold
Connection rule.This step specifically includes following sub-step:
Step 3.1: to the every group of { V for meeting support threshold obtained in step 2i,Vj, generate following correlation rule: Vj
→ViAnd Vi→Vj, note minimal confidence threshold is minconf;In this example, in this example, minimal confidence threshold is set are as follows:
Minconf=0.7;
Step 3.2: according to every group of correlation rule of generation, calculating its confidence threshold value, extract the process of correlation rule such as
Under: for each correlation rule Vi→Vj, calculateIf conf (Vi→Vj) be not less than
Minimal confidence threshold is minconf, then retains correlation rule Vi→VjAnd record corresponding support and confidence level ωi。
This step generates the correlation rule for meeting threshold condition, and extracts part relevant parameter and its confidence value such as table 1
It is shown.As seen from the results in Table 1, this example will use variable 7 and variable 11 to carry out predicted operation as target component.
Step 4: association rule mining result being introduced into wavelet neural network, and the unusual service condition for industrial equipment is pre-
It surveys.This step specifically includes following sub-step:
Step 4.1: for any group association parameter extracted from correlation rule, being denoted as { V1,V2,…,Vu, wherein u table
Show the quantity of relevant parameter, VuFor rule it is consequent i.e. target component, for each correlation rule Vi→Vu, i=1,2 ... u-
1, there is a confidence level, is denoted as ωi;For target component Vu, using wavelet neural network, carry out unusual service condition prediction;
Step 4.2: construction training sample: remembering that preset prediction step is l, in this example, prediction step is set as 10.By
The group association parameter that association rule mining extracts is { V1,V2,…,Vu, the complete training dataset being made of it is denoted asConstruct following matrix ItrainIt is inputted for the training of neural network:
Wherein, ItrainIn it is each be classified as a trained input sample, construct training output OtrainAre as follows:
Specifically, training set herein not merely uses the fault data of IDV (13) correlated variables, while also using
Data under correlated variables normal operating condition.
Step 4.3: be trained using the training sample of construction to wavelet neural network: input parameter is Vi, i=1,
2 ... u-1, output parameter Vu, wherein utilizing the confidence level ω obtained by correlation rule in netiniti, i=1,
The initial weight between network input layer and hidden layer is arranged in 2 ... u-1.In this example, for variable 7, input layer is 4 sections
Point, hidden layer are 8 nodes;For variable 11, input layer is 3 nodes, and hidden layer is 6 nodes, the output of two variables
Layer is 1 node, wherein the wavelet basis function used is Morlet morther wavelet basic function, and with the related confidence in table 1
Initialization weight of the angle value as neural network input layer and hidden layer;
Step 4.4: new data prediction: remembering that threshold value occurs for preset unusual service condition is ωp, for new collected sensor
Measurement data carries out l step prediction using model trained in step 4.3, if obtained target component predicted value is relative to first
The normal drift value that begins is more than set threshold value ωp, then it is assumed that unusual service condition occurs.Before equipment does not fail, with data
It updates, every update predetermined quantity NlMeasurement data after, need to reconfigure and training pattern, to obtain more accurately predicting knot
Fruit, wherein NlDepending on sensor sample frequency and actual industrial field demand.This example utilizes test set (totally 960 samplings
Point) preceding 300 data to verify prediction effect, and neural network is updated according to every 10 data.It is arranged herein different
The threshold value that (i.e. its normal value certain percentage of parameter drift-out) occurs for normal operating condition (failure) is ωp=0.015.
1 correlation rule of table
Regular preceding paragraph | Rule is consequent | Confidence level |
Variable 13 | Variable 7 | 0.7527 |
Variable 16 | Variable 7 | 0.7446 |
Variable 36 | Variable 7 | 0.7017 |
Variable 35 | Variable 11 | 0.7513 |
Variable 36 | Variable 11 | 0.7390 |
Table 2 always predicts error rate
Introduce correlation rule | It is not introduced into correlation rule | |
Variable 7 | 1.0482 | 1.8548 |
Variable 11 | 0.8536 | 1.2135 |
Fig. 1 and Fig. 2 is 13 prediction result of variable 7 and variable, and the advantage of correlation rule is introduced for verifying, and this example ties prediction
Fruit is compared with the neural network prediction result being not introduced under conditions of correlation rule.In fig. 1 and 2, perpendicular solid line is
The actual unusual service condition time of origin under our threshold value setting, erect dotted line and perpendicular chain-dotted line be respectively introduce correlation rule and
It is not introduced into the predicted value of abnormal time of origin under the premise of correlation rule.By Fig. 1 and Fig. 2 it is found that the method institute that the present invention is mentioned
Obtained prediction result can preferably approaching to reality value obtain very well especially in the prediction of first half test data
Prediction result, this is because first half is the operation data under normal condition, training set is more complete and numerical value is relatively concentrated.
In the prediction of out-of-service time, the method that is mentioned of the present invention has also obtained preferable prediction result, in Fig. 1, predicted value and true
Real value is compared to 8 sampled points have been lagged, and in Fig. 2, predicted value has then lagged 5 sampled points compared with true value.Be not introduced into
The prediction result of correlation rule is compared, and the mentioned method of the present invention obviously obtains more accurate prediction result.Variable 7 and variable 11
Prediction error rate calculated result it is as shown in Figure 3 and Figure 4.Meanwhile being further quantized result, this example also calculates total prediction and misses
Rate, as shown in table 2.From the point of view of integrally predicting error, the introducing of correlation rule significantly reduces the prediction error of neural network,
This point has also obtained good embodiment in the data that chart 2 is presented.
Claims (6)
1. a kind of industrial data association rule mining and unusual service condition prediction technique that strategy is generated based on two stages frequent item set,
It is characterized in that, specific step is as follows:
Step 1: to time series data piece-wise linearization expression and symbolism, construction is suitable for the discrete type of association rule mining
Data set;
Step 2: the frequent item set of data set is generated using two stage Frequent Itemsets Mining Algorithm;
Step 3: correlation rule being generated according to frequent item set, extracts the association rule for meeting minimum support and minimal confidence threshold
Then;
Step 4: association rule mining result being introduced into wavelet neural network, and is predicted for the unusual service condition of industrial equipment.
2. a kind of industrial data association rule mining for generating strategy based on two stages frequent item set according to claim 1
And unusual service condition prediction technique, it is characterised in that the step 1 includes following sub-step:
Step 1.1: note sensor measurement time series isN is number of sensors,
K length of time series;Initially fitting starting point isInitially fitting terminal isNote is fitted starting pointIt is quasi-
Closing terminal isError of fitting threshold value is ωE;
Step 1.2: for eachPiecewise fitting is carried out as follows:
1.2.1 waypoint count value count=1 is initialized;
1.2.2 successively to each fitting starting pointExecute step 1)-step 4):
1) end=start+h is calculated first;
2) for dataAnd be fitted using least square method, it calculates quasi-
Close error E RR;
3) if error of fitting ERR is not more than error of fitting threshold value ωE, then 1) h=h+1, gos to step again;
4) if error of fitting ERR is greater than error of fitting threshold value ωE, obtainLine segment be fitted sequence
Start=start+h records waypointReset h=2, count=count+1;
1.2.3 circulation executes 1.2.2 and terminates greater than k until end, the line segment time series after being fittedAnd waypointThe segmentation point sequence P of compositioni;
Step 1.3: the time series after any sensor is fittedIt is denoted as Yk={ y1,y2,…,yk, extract every fit line
The trend and numerical information of section, and a matching line segment s is indicated by the way of following triplei:
Wherein, kiIndicate the slope of the line segment,Indicate the span of the line segment on a timeline, riIndicate the growth of the segment data
Rate, data { y corresponding for the line segmentj,yj+1,…,yj+h,J is the starting point of the line segment;
To line segment time series YkIn all line segments carry out triple expression, obtain triad sequence Sn={ s1,s2,…,
sn, wherein n indicates time series XkLine segment number after segmentation;
Step 1.4: to a serial of line sections in triad sequence carry out cluster and to line segment carry out symbolism, for indicate equipment or
The different version of system describes line segment s using Euclidean distanceiAnd sjSimilarity dij:
Wherein, dijIndicate line segment siAnd sjSimilarity, dijIt is smaller, then it represents that two lines section has more like change shape, ωk
And ωrFor weight;
Then according to index of similarity dij, using K-means clustering algorithm to SnIt is clustered, and is same class line segment distribution one
A the same symbol obtains the sequence F of symbolism to indicate the variation pattern of operating parametern={ f1,f2,…,fn, f1,f2,…,
fnRespectively indicate the 1,2nd ..., the symbol that n line segment is assigned to;
Step 1.5: for the time of measuring sequence of every two sensorWithMerge it and is segmented point sequence PiAnd Pj, it is denoted as
Pij, nij- 1 is PiAnd PjWaypoint number after merging;And by the waypoint after merging to its symbolism sequenceWithIt carries out
Segmentation reconstruct, the symbolism sequence after being reconstructedWith
3. a kind of industrial data association rule mining for generating strategy based on two stages frequent item set according to claim 2
And unusual service condition prediction technique, it is characterised in that the step 2 includes following sub-step:
Step 2.1: for time of measuring sequenceWithCorresponding operating parameter ViAnd Vj, it is obtained by step 1 and measures sequence
The symbolism data of column areWithTransaction set is made of it, i.e., it is each
Affairs are denoted as WithIncluded in line segment class code be denoted as respectivelyWithRemember that two stage minimum support threshold value is respectively minsup1And minsup2;
Step 2.2: by single sweep operation data set, calculating the support of each single item, frequent 1- item collection is obtained, by following 2.2.1
~2.2.3 process carries out:
2.2.1: note σ () is the support counting of item or item collection, is initially 0;IfClass code be tk, t expression a or b;
2.2.2: for each affairsCalculate σ (tk)=σ (tk)+1;
2.2.3: for each tkIfNot less than minimum support threshold value minsup1, then it is assumed that tkIt is 1- frequent
Collection retains tkAnd record corresponding support counting;IfLess than minimum support threshold value minsup1, then it is assumed that tkIt is not
Frequent 1- item collection;
Step 2.3: using the frequent 1- item collection t obtained in step 2.2k2- item collection is constituted, and calculates its support, to find
Frequent 2- item collection, carries out according to the following procedure:
2.3.1: note apAnd bqRespectively pass through step 2.2 from former line segment class codeWith
The item of middle reservation;
2.3.2 for each { ap,bq, execute following steps:
1) each is present inIn { ap,bq, calculate σ ({ ap,bq)=σ ({ ap,bq})+1
If 2)Not less than minsup1, then it is assumed that { ap,bqIt is frequent 2- item collection, retain { ap,bqAnd record corresponding
Support counting;
Step 2.4: using the frequent 2- item collection { a obtained in step 2.3p,bqEvery two operating parameter is calculated in entire data set
In support, and obtain the frequent item set of parameter level, carry out according to the following procedure: to every two operating parameter ViAnd VjIt constitutes
Item collection { Vi,Vj, calculate σ ({ Vi,Vj)=sum (σ ({ ap,bq)), ifNot less than minimum support threshold value
minsup2, then retain { Vi,VjAnd corresponding support is recorded, and calculate σ (Vi)=sum (σ (ap));σ(Vj)=sum (σ
(bq))。
4. a kind of industrial data association rule mining for generating strategy based on two stages frequent item set according to claim 3
And unusual service condition prediction technique, it is characterised in that the step 3 includes following sub-step:
Step 3.1: to the every group of { V for meeting support threshold obtained in step 2i,Vj, generate following correlation rule: Vj→Vi
And Vi→Vj, note minimal confidence threshold is minconf;
Step 3.2: according to every group of correlation rule of generation, calculating its confidence threshold value, the process for extracting correlation rule is as follows: right
In each correlation rule Vi→Vj, calculateIf conf (Vi→Vj) set not less than minimum
Confidence threshold is minconf, then retains correlation rule Vi→VjAnd record corresponding support and confidence level ωi。
5. a kind of industrial data association rule mining for generating strategy based on two stages frequent item set according to claim 4
And unusual service condition prediction technique, it is characterised in that the step 4 includes following sub-step:
Step 4.1: for any group association parameter extracted from correlation rule, being denoted as { V1,V2,…,Vu, wherein u indicates to close
Join the quantity of parameter, VuFor rule it is consequent i.e. target component, for each correlation rule Vi→Vu, i=1,2 ... u-1 have
One confidence level, is denoted as ωi;For target component Vu, using wavelet neural network, carry out unusual service condition prediction;
Step 4.2: construction training sample: remembering that preset prediction step is l, the group association parameter extracted by association rule mining
For { V1,V2,…,Vu, the complete training dataset being made of it is denoted asConstruction
Following matrix ItrainIt is inputted for the training of neural network:
Wherein, ItrainIn it is each be classified as a trained input sample, construct training output OtrainAre as follows:
Step 4.3: be trained using the training sample of construction to wavelet neural network: input parameter is Vi, i=1,2 ... u-
1, output parameter Vu, wherein utilizing the confidence level ω obtained by correlation rule in netiniti, i=1,2 ... u-1,
Initial weight between network input layer and hidden layer is set;
Step 4.4: new data prediction: remembering that threshold value occurs for preset unusual service condition is ωp, for new collected sensor measurement
Data carry out l step prediction using model trained in step 4.3, if obtained target component predicted value is relative to initially just
Normal drift value is more than set threshold value ωp, then it is assumed that unusual service condition occurs.
6. a kind of industrial data association rule mining for generating strategy based on two stages frequent item set according to claim 1
And unusual service condition prediction technique, it is characterised in that before equipment does not fail, with the update of data, every update predetermined quantity
After measurement data, simultaneously training pattern need to be reconfigured, to obtain more accurate prediction result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910244856.6A CN110008253B (en) | 2019-03-28 | 2019-03-28 | Industrial data association rule mining and abnormal working condition prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910244856.6A CN110008253B (en) | 2019-03-28 | 2019-03-28 | Industrial data association rule mining and abnormal working condition prediction method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110008253A true CN110008253A (en) | 2019-07-12 |
CN110008253B CN110008253B (en) | 2021-02-23 |
Family
ID=67168723
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910244856.6A Active CN110008253B (en) | 2019-03-28 | 2019-03-28 | Industrial data association rule mining and abnormal working condition prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110008253B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112130541A (en) * | 2020-10-20 | 2020-12-25 | 陕西煤业新型能源科技股份有限公司 | Energy comprehensive management control system based on Internet of things |
CN112380274A (en) * | 2020-11-16 | 2021-02-19 | 北京航空航天大学 | Control process-oriented anomaly detection system |
CN112801426A (en) * | 2021-04-06 | 2021-05-14 | 浙江浙能技术研究院有限公司 | Industrial process fault fusion prediction method based on correlation parameter mining |
CN112800686A (en) * | 2021-03-29 | 2021-05-14 | 国网江西省电力有限公司电力科学研究院 | Transformer DGA online monitoring data abnormal mode judgment method |
CN113032912A (en) * | 2021-04-20 | 2021-06-25 | 上海交通大学 | Ship diesel engine fault detection method based on association rule |
CN113792754A (en) * | 2021-08-12 | 2021-12-14 | 国网江西省电力有限公司电力科学研究院 | Method for processing DGA (differential global alignment) online monitoring data of converter transformer by removing different elements and then repairing |
CN114936581A (en) * | 2022-06-01 | 2022-08-23 | 中国人民解放军63796部队 | Multi-parameter association mining method based on time sequence data segmentation |
CN115497267A (en) * | 2022-09-06 | 2022-12-20 | 江西小手软件技术有限公司 | Equipment early warning platform based on time sequence association rule |
CN115689071A (en) * | 2023-01-03 | 2023-02-03 | 南京工大金泓能源科技有限公司 | Equipment fault fusion prediction method and system based on correlation parameter mining |
CN116204842A (en) * | 2023-03-10 | 2023-06-02 | 广东省建设工程质量安全检测总站有限公司 | Abnormality monitoring method and system for electrical equipment |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN201898519U (en) * | 2010-09-01 | 2011-07-13 | 燕山大学 | Equipment maintenance early-warning device with risk control |
US20120239600A1 (en) * | 2009-12-21 | 2012-09-20 | International Business Machines Corporation | Method for training and using a classification model with association rule models |
CN103676645A (en) * | 2013-12-11 | 2014-03-26 | 广东电网公司电力科学研究院 | Mining method for association rules in time series data flows |
CN108873859A (en) * | 2018-05-31 | 2018-11-23 | 浙江工业大学 | Based on the bridge-type grab ship unloader fault prediction model method for improving correlation rule |
-
2019
- 2019-03-28 CN CN201910244856.6A patent/CN110008253B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120239600A1 (en) * | 2009-12-21 | 2012-09-20 | International Business Machines Corporation | Method for training and using a classification model with association rule models |
CN201898519U (en) * | 2010-09-01 | 2011-07-13 | 燕山大学 | Equipment maintenance early-warning device with risk control |
CN103676645A (en) * | 2013-12-11 | 2014-03-26 | 广东电网公司电力科学研究院 | Mining method for association rules in time series data flows |
CN108873859A (en) * | 2018-05-31 | 2018-11-23 | 浙江工业大学 | Based on the bridge-type grab ship unloader fault prediction model method for improving correlation rule |
Non-Patent Citations (1)
Title |
---|
王玲等: ""基于频繁项集树的时态关联规则挖掘算法"", 《控制与决策》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112130541A (en) * | 2020-10-20 | 2020-12-25 | 陕西煤业新型能源科技股份有限公司 | Energy comprehensive management control system based on Internet of things |
CN112380274B (en) * | 2020-11-16 | 2023-08-22 | 北京航空航天大学 | Abnormality detection method for control process |
CN112380274A (en) * | 2020-11-16 | 2021-02-19 | 北京航空航天大学 | Control process-oriented anomaly detection system |
CN112800686A (en) * | 2021-03-29 | 2021-05-14 | 国网江西省电力有限公司电力科学研究院 | Transformer DGA online monitoring data abnormal mode judgment method |
CN112801426A (en) * | 2021-04-06 | 2021-05-14 | 浙江浙能技术研究院有限公司 | Industrial process fault fusion prediction method based on correlation parameter mining |
CN113032912A (en) * | 2021-04-20 | 2021-06-25 | 上海交通大学 | Ship diesel engine fault detection method based on association rule |
CN113792754A (en) * | 2021-08-12 | 2021-12-14 | 国网江西省电力有限公司电力科学研究院 | Method for processing DGA (differential global alignment) online monitoring data of converter transformer by removing different elements and then repairing |
CN114936581A (en) * | 2022-06-01 | 2022-08-23 | 中国人民解放军63796部队 | Multi-parameter association mining method based on time sequence data segmentation |
CN114936581B (en) * | 2022-06-01 | 2024-04-26 | 中国人民解放军63796部队 | Multi-parameter association mining method based on time sequence data segmentation |
CN115497267A (en) * | 2022-09-06 | 2022-12-20 | 江西小手软件技术有限公司 | Equipment early warning platform based on time sequence association rule |
CN115689071A (en) * | 2023-01-03 | 2023-02-03 | 南京工大金泓能源科技有限公司 | Equipment fault fusion prediction method and system based on correlation parameter mining |
CN116204842A (en) * | 2023-03-10 | 2023-06-02 | 广东省建设工程质量安全检测总站有限公司 | Abnormality monitoring method and system for electrical equipment |
CN116204842B (en) * | 2023-03-10 | 2023-09-08 | 广东省建设工程质量安全检测总站有限公司 | Abnormality monitoring method and system for electrical equipment |
Also Published As
Publication number | Publication date |
---|---|
CN110008253B (en) | 2021-02-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110008253A (en) | The industrial data association rule mining and unusual service condition prediction technique of strategy are generated based on two stages frequent item set | |
CN110018670A (en) | A kind of industrial process unusual service condition prediction technique excavated based on dynamic association rules | |
JP7240691B1 (en) | Data drive active power distribution network abnormal state detection method and system | |
CN109522600A (en) | Complex equipment remaining life prediction technique based on combined depth neural network | |
Nathan et al. | Estimating low flow characteristics in ungauged catchments | |
CN112508105B (en) | Fault detection and retrieval method for oil extraction machine | |
CN109857090B (en) | Health evaluation system and method for balanced air cylinder braking device | |
CN102789545A (en) | Method for predicating remaining life of turbine engine based on degradation model matching | |
AU2016287383A1 (en) | Method for detecting anomalies in a water distribution system | |
CN112200237B (en) | Time sequence monitoring data abnormality diagnosis method for structural health monitoring system | |
CN111122162B (en) | Industrial system fault detection method based on Euclidean distance multi-scale fuzzy sample entropy | |
CN111898644B (en) | Intelligent identification method for health state of aerospace liquid engine under fault-free sample | |
CN104950875A (en) | Fault diagnosis method by combining correlation analysis and data fusion | |
CN114004137A (en) | Multi-source meteorological data fusion and pretreatment method | |
CN112507479B (en) | Oil drilling machine health state assessment method based on manifold learning and softmax | |
Son et al. | Deep learning-based anomaly detection to classify inaccurate data and damaged condition of a cable-stayed bridge | |
CN109471698A (en) | System and method for detecting abnormal behavior of virtual machine in cloud environment | |
CN115310361A (en) | Method and system for predicting underground dust concentration of coal mine based on WGAN-CNN | |
CN114492642A (en) | Mechanical fault online diagnosis method for multi-scale element depth residual shrinkage network | |
CN114882069A (en) | Taxi track abnormity detection method based on LSTM network and attention mechanism | |
CN113029619A (en) | Underground scraper fault diagnosis method based on C4.5 decision tree algorithm | |
CN115982658A (en) | Hydrological data anomaly identification and repair method based on federated learning framework | |
Mangaraj et al. | A Markov chain analysis of daily rainfall occurrence at western Orissa of India | |
CN109299201A (en) | Power plant's production subsystem method for monitoring abnormality and device based on two-phase analyzing method | |
CN104063622A (en) | Complex system monitoring data visualization method based on similarity measurement |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |